{ "$schema": "https://themachinepress.com/schemas/edition-v1.schema.json", "schema_version": "1.0.0", "document_type": "machine_press_edition", "edition_id": "mp-2026-10-20-morning-0102", "source_edition_id": "mp-2026-10-20-morning-0102", "edition_number": 102, "edition_label": "Morning edition", "status": "published", "revision": 1, "published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "timezone": "America/Detroit", "canonical_url": "https://themachinepress.com/edition/2026-10-20", "html_url": "https://themachinepress.com/edition/2026-10-20", "json_url": "https://themachinepress.com/edition/2026-10-20.json", "markdown_url": "https://themachinepress.com/edition/2026-10-20.md", "lead_story_id": "mp-2026-10-20-001", "lead_summary": "A general multimodal model interpreted vision-based tactile images and used their contact cues across nineteen simulated manipulation tasks.", "coverage_window": { "start": "2026-10-19T13:15:53.000-04:00", "end": "2026-10-20T08:15:06.176-04:00" }, "story_count": 27, "stories": [ { "story_id": "mp-2026-10-20-001", "source_story_id": "tmp-lead-visuotactile-astra", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 1, "story_type": "lead", "section": "robotics", "editorial_classification": "editorial", "headline": "The Robot Learned Touch Through Its Eyes", "slug": "the-robot-learned-touch-through-its-eyes", "dek": "A general multimodal model interpreted vision-based tactile images and used their contact cues across nineteen simulated manipulation tasks.", "summary": "A general multimodal model interpreted vision-based tactile images and used their contact cues across nineteen simulated manipulation tasks.", "body_text": "Touch has usually been treated as specialist territory in robot learning. In a new simulation study, researchers passed RGB images from vision-based tactile sensors through GPT-6 Astra's existing visual interface, then asked the model to control nineteen contact-rich tasks with few-shot demonstrations and closed-loop feedback. Across 210 sampled scene seeds per setting, the authors report mean success of 93.8 percent on UniVTAC tasks and 81.8 percent on OpenViTac tasks, exceeding the strongest specialist baselines in their comparison by 22.9 and 13.1 percentage points. Trajectories showed active probing, cautious grasps, contact-guided alignment and recovery; tangential loading and corrective execution remained difficult. This is not a physical-robot validation, and the authors warn of possible benchmark saturation. The narrower result is still useful: a general visual model may already possess primitives for reading touch when tactile sensing is rendered as images.", "why_it_matters": "A general multimodal model interpreted vision-based tactile images and used their contact cues across nineteen simulated manipulation tasks.", "limitations": [ "This is not a physical-robot validation, and the authors warn of possible benchmark saturation." ], "importance": 10, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-001/the-robot-learned-touch-through-its-eyes", "json_url": "https://themachinepress.com/story/mp-2026-10-20-001.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-001-001", "text": "A general multimodal model interpreted vision-based tactile images and used their contact cues across nineteen simulated manipulation tasks.", "source_ids": [ "source-2026-10-20-001" ], "qualification": "This is not a physical-robot validation, and the authors warn of possible benchmark saturation." } ], "source_ids": [ "source-2026-10-20-001" ], "tags": [ "robotics", "tactile sensing", "multimodal models", "manipulation" ], "image_url": "https://themachinepress.com/issues/2026-10-20/lead-visuotactile-robot.png", "corrections": [] }, { "story_id": "mp-2026-10-20-002", "source_story_id": "tmp-feature-weak-magnetism-neutron", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 2, "story_type": "secondary", "section": "research", "editorial_classification": "editorial", "headline": "Weak Magnetism Appears in the Free Neutron", "slug": "weak-magnetism-appears-in-the-free-neutron", "dek": "Combining beta asymmetry, neutron lifetime and superallowed-decay measurements lifted a long-standing degeneracy and produced a 5.1-sigma signal.", "summary": "Combining beta asymmetry, neutron lifetime and superallowed-decay measurements lifted a long-standing degeneracy and produced a 5.1-sigma signal.", "body_text": "Nearly seven decades after Murray Gell-Mann introduced weak magnetism as a consequence of the conserved vector current hypothesis, researchers report its first observation in free-neutron decay. Beta asymmetry alone cannot separate the weak-magnetism form factor from the axial-to-vector coupling ratio, so the analysis combines asymmetry measurements with the neutron lifetime and the comparative half-life of superallowed nuclear decays. Current data yield a form factor of 4.47 with an uncertainty of 0.87, consistent with the predicted 4.71 and excluding no weak magnetism at 5.1 sigma. Across variations of the internally tense neutron dataset, the reported significance remains between 5.0 and 6.1 sigma. This is an inference from combined existing measurements, not a new detector event; its value is that three familiar observables expose a property each could not isolate alone.", "why_it_matters": "Combining beta asymmetry, neutron lifetime and superallowed-decay measurements lifted a long-standing degeneracy and produced a 5.1-sigma signal.", "limitations": [ "Beta asymmetry alone cannot separate the weak-magnetism form factor from the axial-to-vector coupling ratio, so the analysis combines asymmetry measurements with the neutron lifetime and the comparative half-life of superallowed nuclear decays.", "This is an inference from combined existing measurements, not a new detector event; its value is that three familiar observables expose a property each could not isolate alone." ], "importance": 9, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-002/weak-magnetism-appears-in-the-free-neutron", "json_url": "https://themachinepress.com/story/mp-2026-10-20-002.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-002-001", "text": "Combining beta asymmetry, neutron lifetime and superallowed-decay measurements lifted a long-standing degeneracy and produced a 5.1-sigma signal.", "source_ids": [ "source-2026-10-20-002" ], "qualification": "Beta asymmetry alone cannot separate the weak-magnetism form factor from the axial-to-vector coupling ratio, so the analysis combines asymmetry measurements with the neutron lifetime and the comparative half-life of superallowed nuclear decays." } ], "source_ids": [ "source-2026-10-20-002" ], "tags": [ "neutron decay", "weak magnetism", "particle physics", "conserved vector current" ], "image_url": "https://themachinepress.com/issues/2026-10-20/feature-weak-magnetism.png", "corrections": [] }, { "story_id": "mp-2026-10-20-003", "source_story_id": "tmp-story-functional-cad-mechanisms", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 3, "story_type": "dispatch", "section": "developer-tools", "editorial_classification": "editorial", "headline": "The CAD Assembly Has to Move, Not Merely Look Right", "slug": "the-cad-assembly-has-to-move-not-merely-look-right", "dek": "FunCAD-Harness turns underspecified mechanism requests into candidate-specific physical tests, then uses simulation failures to guide revisions.", "summary": "FunCAD-Harness turns underspecified mechanism requests into candidate-specific physical tests, then uses simulation failures to guide revisions.", "body_text": "A plausible rendering can conceal a mechanism that cannot perform its job. FunCAD-Harness asks agents to construct and revise mechanisms against physical simulation, generating tests for each candidate rather than matching one reference geometry. Its benchmark spans 46 mechanism families with interdependent, multistep operation. The authors report that the harness exposed failures invisible to visual inspection and improved functional success over evaluated simulation-assisted baselines. It does not certify generated designs for manufacture, but it moves CAD evaluation from appearance toward behavior.", "why_it_matters": "FunCAD-Harness turns underspecified mechanism requests into candidate-specific physical tests, then uses simulation failures to guide revisions.", "limitations": [ "A plausible rendering can conceal a mechanism that cannot perform its job.", "It does not certify generated designs for manufacture, but it moves CAD evaluation from appearance toward behavior." ], "importance": 9, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-003/the-cad-assembly-has-to-move-not-merely-look-right", "json_url": "https://themachinepress.com/story/mp-2026-10-20-003.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-003-001", "text": "FunCAD-Harness turns underspecified mechanism requests into candidate-specific physical tests, then uses simulation failures to guide revisions.", "source_ids": [ "source-2026-10-20-003" ], "qualification": "A plausible rendering can conceal a mechanism that cannot perform its job." } ], "source_ids": [ "source-2026-10-20-003" ], "tags": [ "CAD", "simulation", "mechanism design", "agents" ], "image_url": "https://themachinepress.com/issues/2026-10-20/functional-cad-robot.webp", "corrections": [] }, { "story_id": "mp-2026-10-20-004", "source_story_id": "tmp-story-halcyon-no-language-pretraining", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 4, "story_type": "dispatch", "section": "frontier-models", "editorial_classification": "editorial", "headline": "The Image Model Skipped Language School", "slug": "the-image-model-skipped-language-school", "dek": "Halcyon trained from scratch on paired images and text, and controlled comparisons found no generation benefit from text-only pretraining.", "summary": "Halcyon trained from scratch on paired images and text, and controlled comparisons found no generation benefit from text-only pretraining.", "body_text": "Modern text-to-image systems commonly inherit an internet-scale language model. Halcyon instead trained from scratch on 100 million image-text pairs. The authors report that a 520-million-parameter version beat a larger LlamaGen comparison with 18.2 times less pretraining compute, while a 1.6-billion-parameter model beat the evaluated JiT-T2I system with 275 times less compute. Controlled experiments found no image-generation benefit from text-only pretraining. The reported metrics do not settle every dimension of quality, but they challenge the assumption that an image generator must first become a broadly pretrained language model.", "why_it_matters": "Halcyon trained from scratch on paired images and text, and controlled comparisons found no generation benefit from text-only pretraining.", "limitations": [ "Controlled experiments found no image-generation benefit from text-only pretraining.", "The reported metrics do not settle every dimension of quality, but they challenge the assumption that an image generator must first become a broadly pretrained language model." ], "importance": 9, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-004/the-image-model-skipped-language-school", "json_url": "https://themachinepress.com/story/mp-2026-10-20-004.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-004-001", "text": "Halcyon trained from scratch on paired images and text, and controlled comparisons found no generation benefit from text-only pretraining.", "source_ids": [ "source-2026-10-20-004" ], "qualification": "Controlled experiments found no image-generation benefit from text-only pretraining." } ], "source_ids": [ "source-2026-10-20-004" ], "tags": [ "image generation", "multimodal pretraining", "efficiency" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-005", "source_story_id": "tmp-story-openvlharness-multimodal-memory", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 5, "story_type": "dispatch", "section": "developer-tools", "editorial_classification": "editorial", "headline": "The Visual Agent Keeps the Evidence It Found", "slug": "the-visual-agent-keeps-the-evidence-it-found", "dek": "OpenVLHarness turns tool output into addressable images, data and web references that persist across a multimodal reasoning trace.", "summary": "OpenVLHarness turns tool output into addressable images, data and web references that persist across a multimodal reasoning trace.", "body_text": "More inference did little on 22 visual benchmarks when the missing ingredient was external evidence. OpenVLHarness adds a capability layer that renders heterogeneous tool results into model-readable observations and persistent memory that keeps images, data and references addressable. Across eight evaluated backbones, the authors report average gains of 9.7 to 12.6 points over matched no-tool baselines. In one comparison, a medium-effort smaller model beat maximum-effort systems while using no more than 37 percent of their model-token cost. These are harness-specific results, but they show why tool access and evidence retention must be evaluated together.", "why_it_matters": "OpenVLHarness turns tool output into addressable images, data and web references that persist across a multimodal reasoning trace.", "limitations": [], "importance": 8, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-005/the-visual-agent-keeps-the-evidence-it-found", "json_url": "https://themachinepress.com/story/mp-2026-10-20-005.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-005-001", "text": "OpenVLHarness turns tool output into addressable images, data and web references that persist across a multimodal reasoning trace.", "source_ids": [ "source-2026-10-20-005" ], "qualification": null } ], "source_ids": [ "source-2026-10-20-005" ], "tags": [ "visual agents", "tool use", "multimodal memory" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-006", "source_story_id": "tmp-story-ropedia-world-model", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 6, "story_type": "dispatch", "section": "robotics", "editorial_classification": "editorial", "headline": "Human Experience Scales a Robot's Moving World", "slug": "human-experience-scales-a-robot-s-moving-world", "dek": "Ropedia W0 combines 10,000 hours of egocentric human video with 3,000 hours of robot data across different bodies.", "summary": "Ropedia W0 combines 10,000 hours of egocentric human video with 3,000 hours of robot data across different bodies.", "body_text": "World models often separate locomotion from manipulation or bind themselves to one robot. Ropedia W0 represents movement with camera poses and manipulation with projected skeletons, combining 10,000 hours of human egocentric data with 3,000 hours of robot data. The authors report leading results across five cross-embodiment benchmarks and held-out human video, with additional human pretraining improving both human and robot evaluations. A distilled version produced real-time, minute-long rollouts for live virtual-reality interaction. The evidence is benchmark-based, but it suggests human motion can supply scale without pretending every body has the same joints.", "why_it_matters": "Ropedia W0 combines 10,000 hours of egocentric human video with 3,000 hours of robot data across different bodies.", "limitations": [], "importance": 8, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-006/human-experience-scales-a-robot-s-moving-world", "json_url": "https://themachinepress.com/story/mp-2026-10-20-006.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-006-001", "text": "Ropedia W0 combines 10,000 hours of egocentric human video with 3,000 hours of robot data across different bodies.", "source_ids": [ "source-2026-10-20-006" ], "qualification": null } ], "source_ids": [ "source-2026-10-20-006" ], "tags": [ "world models", "loco-manipulation", "human video" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-007", "source_story_id": "tmp-story-physical-words-robot-reasoning", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 7, "story_type": "dispatch", "section": "robotics", "editorial_classification": "editorial", "headline": "A Robot Gives Its Hidden Dynamics a Vocabulary", "slug": "a-robot-gives-its-hidden-dynamics-a-vocabulary", "dek": "PhysWord anchors continuous future-state representations to atomic action verbs before the policy acts.", "summary": "PhysWord anchors continuous future-state representations to atomic action verbs before the policy acts.", "body_text": "Language can describe a task while missing the fine structure of contact and motion; latent forecasts contain that structure but are difficult to reuse. PhysWord predicts atomic action words and aligns their hidden states with future-observation targets, creating visual and optional tactile versions of each primitive. The authors report 90.3 percent success on visuotactile UniVTAC and 99.6 percent on LIBERO-Long, plus gains on real bimanual and dexterous tasks. The evaluations do not prove the vocabulary is universal, but they make a testable claim: action anchors can structure physical prediction without reducing it to prose.", "why_it_matters": "PhysWord anchors continuous future-state representations to atomic action verbs before the policy acts.", "limitations": [ "The evaluations do not prove the vocabulary is universal, but they make a testable claim: action anchors can structure physical prediction without reducing it to prose." ], "importance": 8, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-007/a-robot-gives-its-hidden-dynamics-a-vocabulary", "json_url": "https://themachinepress.com/story/mp-2026-10-20-007.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-007-001", "text": "PhysWord anchors continuous future-state representations to atomic action verbs before the policy acts.", "source_ids": [ "source-2026-10-20-007" ], "qualification": "The evaluations do not prove the vocabulary is universal, but they make a testable claim: action anchors can structure physical prediction without reducing it to prose." } ], "source_ids": [ "source-2026-10-20-007" ], "tags": [ "robot reasoning", "latent dynamics", "action vocabulary" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-008", "source_story_id": "tmp-story-weact-weight-switching", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 8, "story_type": "dispatch", "section": "frontier-models", "editorial_classification": "editorial", "headline": "The Agent Changes Its Weights Mid-Task", "slug": "the-agent-changes-its-weights-mid-task", "dek": "WeAct adds selection among capability-specific adapters to the action space instead of fixing one parameter configuration for an entire trajectory.", "summary": "WeAct adds selection among capability-specific adapters to the action space instead of fixing one parameter configuration for an entire trajectory.", "body_text": "A long research task may demand browsing, calculation and writing in sequence, while updating one shared model risks forgetting an earlier skill. WeAct freezes the backbone, trains separate LoRA adapters and lets the agent select one through its normal tool-calling interface. On eight deep-research benchmarks, the authors report a 6.2 percent average accuracy gain over full fine-tuning and 27.5 percent over the strongest evaluated static adapter merge. Adding a computation adapter and updating a writing adapter left other modules unchanged. The work tests one modular strategy, not arbitrary self-modification, but it turns internal parameter choice into an explicit action.", "why_it_matters": "WeAct adds selection among capability-specific adapters to the action space instead of fixing one parameter configuration for an entire trajectory.", "limitations": [ "The work tests one modular strategy, not arbitrary self-modification, but it turns internal parameter choice into an explicit action." ], "importance": 8, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-008/the-agent-changes-its-weights-mid-task", "json_url": "https://themachinepress.com/story/mp-2026-10-20-008.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-008-001", "text": "WeAct adds selection among capability-specific adapters to the action space instead of fixing one parameter configuration for an entire trajectory.", "source_ids": [ "source-2026-10-20-008" ], "qualification": "The work tests one modular strategy, not arbitrary self-modification, but it turns internal parameter choice into an explicit action." } ], "source_ids": [ "source-2026-10-20-008" ], "tags": [ "agents", "LoRA", "continual learning" ], "image_url": "https://themachinepress.com/issues/2026-10-20/agent-weight-switching.webp", "corrections": [] }, { "story_id": "mp-2026-10-20-009", "source_story_id": "tmp-story-coopuq-social-navigation", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 9, "story_type": "dispatch", "section": "robotics", "editorial_classification": "editorial", "headline": "The Robots Share What They Cannot See", "slug": "the-robots-share-what-they-cannot-see", "dek": "CoopUQ fuses occupancy forecasts across robots and makes uncertainty itself part of the navigation cost.", "summary": "CoopUQ fuses occupancy forecasts across robots and makes uncertainty itself part of the navigation cost.", "body_text": "One robot may miss a pedestrian hidden from its sensors while another has a better view. CoopUQ combines cross-robot features into future occupancy maps, estimates uncertainty and penalizes policies for entering regions predicted to be occupied or poorly known. In simulated layouts with varied density, occlusion and rushing behavior, the authors report better prediction and safety than representative baselines, including under distribution shifts. Real-robot experiments provided additional robustness evidence. The paper does not establish safety in open public spaces, but it ties cooperation to a shared map of risk rather than consensus for its own sake.", "why_it_matters": "CoopUQ fuses occupancy forecasts across robots and makes uncertainty itself part of the navigation cost.", "limitations": [ "The paper does not establish safety in open public spaces, but it ties cooperation to a shared map of risk rather than consensus for its own sake." ], "importance": 8, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-009/the-robots-share-what-they-cannot-see", "json_url": "https://themachinepress.com/story/mp-2026-10-20-009.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-009-001", "text": "CoopUQ fuses occupancy forecasts across robots and makes uncertainty itself part of the navigation cost.", "source_ids": [ "source-2026-10-20-009" ], "qualification": "The paper does not establish safety in open public spaces, but it ties cooperation to a shared map of risk rather than consensus for its own sake." } ], "source_ids": [ "source-2026-10-20-009" ], "tags": [ "multi-robot systems", "navigation", "uncertainty" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-010", "source_story_id": "tmp-story-physical-rsi-scaling", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 10, "story_type": "dispatch", "section": "robotics", "editorial_classification": "editorial", "headline": "Robot Improvement Has More Than One Bottleneck", "slug": "robot-improvement-has-more-than-one-bottleneck", "dek": "PhysRSI-eval separates first attempts, efficiency, success, budget and recovery from failure across eighteen manipulation tasks.", "summary": "PhysRSI-eval separates first attempts, efficiency, success, budget and recovery from failure across eighteen manipulation tasks.", "body_text": "A final success rate cannot say whether a robot learned quickly, recovered from failure or simply spent more trials. PhysRSI-eval measures five axes across eighteen manipulation tasks under a fixed budget. Removing history reduced late-round success, while frozen workspaces still helped on unseen layouts and three cross-task cases succeeded earlier. The observations do not demonstrate unrestricted recursive self-improvement. They provide a more disciplined way to separate scene adaptation, reusable experience and extra attempts.", "why_it_matters": "PhysRSI-eval separates first attempts, efficiency, success, budget and recovery from failure across eighteen manipulation tasks.", "limitations": [ "A final success rate cannot say whether a robot learned quickly, recovered from failure or simply spent more trials.", "The observations do not demonstrate unrestricted recursive self-improvement." ], "importance": 8, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-010/robot-improvement-has-more-than-one-bottleneck", "json_url": "https://themachinepress.com/story/mp-2026-10-20-010.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-010-001", "text": "PhysRSI-eval separates first attempts, efficiency, success, budget and recovery from failure across eighteen manipulation tasks.", "source_ids": [ "source-2026-10-20-010" ], "qualification": "A final success rate cannot say whether a robot learned quickly, recovered from failure or simply spent more trials." } ], "source_ids": [ "source-2026-10-20-010" ], "tags": [ "robot learning", "adaptation", "evaluation" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-011", "source_story_id": "tmp-story-anygoal-world-model-generalization", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 11, "story_type": "dispatch", "section": "research", "editorial_classification": "editorial", "headline": "The World Model Knows the Pieces but Not Their New Arrangement", "slug": "the-world-model-knows-the-pieces-but-not-their-new-arrangement", "dek": "AnyGoal holds familiar state and action factors inside training support, then recombines them into underrepresented planning tasks.", "summary": "AnyGoal holds familiar state and action factors inside training support, then recombines them into underrepresented planning tasks.", "body_text": "A world model can reproduce expert trajectories and still fail when familiar ingredients appear in a new combination. AnyGoal builds goal-shifted variants where each state factor and action remains known, but their pairing was rare in expert data. Across several model families and environments, the authors found consistent prediction and planning failures under those recombinations. The benchmark is a diagnostic rather than a solution, but it raises the standard for generalization: knowing every move independently is not the same as knowing the world's rules.", "why_it_matters": "AnyGoal holds familiar state and action factors inside training support, then recombines them into underrepresented planning tasks.", "limitations": [ "The benchmark is a diagnostic rather than a solution, but it raises the standard for generalization: knowing every move independently is not the same as knowing the world's rules." ], "importance": 8, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-011/the-world-model-knows-the-pieces-but-not-their-new-arrangement", "json_url": "https://themachinepress.com/story/mp-2026-10-20-011.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-011-001", "text": "AnyGoal holds familiar state and action factors inside training support, then recombines them into underrepresented planning tasks.", "source_ids": [ "source-2026-10-20-011" ], "qualification": "The benchmark is a diagnostic rather than a solution, but it raises the standard for generalization: knowing every move independently is not the same as knowing the world's rules." } ], "source_ids": [ "source-2026-10-20-011" ], "tags": [ "world models", "generalization", "planning" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-012", "source_story_id": "tmp-story-latcomm-context-packets", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 12, "story_type": "dispatch", "section": "infrastructure", "editorial_classification": "editorial", "headline": "One Model Sends Another a Reusable Context Packet", "slug": "one-model-sends-another-a-reusable-context-packet", "dek": "LatComm encodes shared context once, then lets a different frozen model select what to read for each question.", "summary": "LatComm encodes shared context once, then lets a different frozen model select what to read for each question.", "body_text": "Multi-model systems repeatedly pay to transmit the same text. LatComm separates encoding from reading: one frozen model compresses context into a fixed-size packet, while a different frozen receiver uses prompt-conditioned cross-attention to select relevant content. Across twelve multiple-choice datasets, the authors report up to 10.2 points higher average accuracy than single-vector baselines. On QuALITY, receiver input fell by a factor of 57 and questions were answered three times faster. The learned bridge is not universally portable, but it demonstrates context that is compact, reusable and readable by a different model.", "why_it_matters": "LatComm encodes shared context once, then lets a different frozen model select what to read for each question.", "limitations": [ "The learned bridge is not universally portable, but it demonstrates context that is compact, reusable and readable by a different model." ], "importance": 8, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-012/one-model-sends-another-a-reusable-context-packet", "json_url": "https://themachinepress.com/story/mp-2026-10-20-012.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-012-001", "text": "LatComm encodes shared context once, then lets a different frozen model select what to read for each question.", "source_ids": [ "source-2026-10-20-012" ], "qualification": "The learned bridge is not universally portable, but it demonstrates context that is compact, reusable and readable by a different model." } ], "source_ids": [ "source-2026-10-20-012" ], "tags": [ "context compression", "model communication", "inference" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-013", "source_story_id": "tmp-story-proteingeobench-agent-geometry", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 13, "story_type": "dispatch", "section": "research", "editorial_classification": "editorial", "headline": "The Biology Agent Can Find the Protein but Misread Its Shape", "slug": "the-biology-agent-can-find-the-protein-but-misread-its-shape", "dek": "ProteinGeoBench separates metadata lookup from numerical geometry and shows how harness, access and run budget can change the ranking.", "summary": "ProteinGeoBench separates metadata lookup from numerical geometry and shows how harness, access and run budget can change the ranking.", "body_text": "Structural-biology agents are often asked to retrieve facts and compute geometry in one score. ProteinGeoBench splits those abilities across 131 tasks. In 4,249 rollouts from four frontier models, three landed within 0.031 of one another; three of the four performed better on metadata lookup than numerical calculation. Numeric questions made up 28 tasks but half of the twelve every model failed. Rankings also changed with run budget and command-line harness. The study does not crown one universal agent; it argues evaluations should expose task family, repeats, access and harness instead of letting lookup fluency hide geometric weakness.", "why_it_matters": "ProteinGeoBench separates metadata lookup from numerical geometry and shows how harness, access and run budget can change the ranking.", "limitations": [ "The study does not crown one universal agent; it argues evaluations should expose task family, repeats, access and harness instead of letting lookup fluency hide geometric weakness." ], "importance": 8, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-013/the-biology-agent-can-find-the-protein-but-misread-its-shape", "json_url": "https://themachinepress.com/story/mp-2026-10-20-013.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-013-001", "text": "ProteinGeoBench separates metadata lookup from numerical geometry and shows how harness, access and run budget can change the ranking.", "source_ids": [ "source-2026-10-20-013" ], "qualification": "The study does not crown one universal agent; it argues evaluations should expose task family, repeats, access and harness instead of letting lookup fluency hide geometric weakness." } ], "source_ids": [ "source-2026-10-20-013" ], "tags": [ "structural biology", "agents", "evaluation" ], "image_url": "https://themachinepress.com/issues/2026-10-20/protein-geometry-lab.webp", "corrections": [] }, { "story_id": "mp-2026-10-20-014", "source_story_id": "tmp-story-wildawm-computer-environments", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 14, "story_type": "dispatch", "section": "developer-tools", "editorial_classification": "editorial", "headline": "The Computer-Use Agent Trains in Worlds Built to Be Checked", "slug": "the-computer-use-agent-trains-in-worlds-built-to-be-checked", "dek": "WildAWM synthesizes executable GUI, tool and command-line environments with solvability and verifier gates.", "summary": "WildAWM synthesizes executable GUI, tool and command-line environments with solvability and verifier gates.", "body_text": "Training computer-use agents directly in live systems is costly and risky. WildAWM uses open-source models to synthesize controlled environments spanning graphical interfaces, tool services and command lines. Quality gates test solvability, difficulty and verifier robustness with agent runs, judge agreement and counterfactual probes. The pipeline produced 341 long-horizon environment-task pairs, some requiring more than 100 steps; a frontier model solved 60 percent. Two Qwen models trained only in these worlds improved across eight out-of-distribution benchmarks. The collection is finite, but its execution checks make generation more than prompt-written scenery.", "why_it_matters": "WildAWM synthesizes executable GUI, tool and command-line environments with solvability and verifier gates.", "limitations": [ "Two Qwen models trained only in these worlds improved across eight out-of-distribution benchmarks." ], "importance": 8, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-014/the-computer-use-agent-trains-in-worlds-built-to-be-checked", "json_url": "https://themachinepress.com/story/mp-2026-10-20-014.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-014-001", "text": "WildAWM synthesizes executable GUI, tool and command-line environments with solvability and verifier gates.", "source_ids": [ "source-2026-10-20-014" ], "qualification": "Two Qwen models trained only in these worlds improved across eight out-of-distribution benchmarks." } ], "source_ids": [ "source-2026-10-20-014" ], "tags": [ "computer use", "synthetic environments", "verification" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-026", "source_story_id": "tmp-story-tool-output-evidence-substitution", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 15, "story_type": "dispatch", "section": "safety", "editorial_classification": "editorial", "headline": "The Agent Computed the Right Number and Cited the Wrong Evidence", "slug": "the-agent-computed-the-right-number-and-cited-the-wrong-evidence", "dek": "Retrieving a simulator report from files changed which value some models trusted even when the underlying computation stayed fixed.", "summary": "Retrieving a simulator report from files changed which value some models trusted even when the underlying computation stayed fixed.", "body_text": "Using deterministic PHREEQC geochemistry reports and 158 execution-verified questions, researchers tested the common practice of storing long tool output outside the context window. Under indexed retrieval, two GPT models lost 11.3 and 7.9 accuracy points while two Claude models showed no detectable cost. Replaying 11,007 simulator calls showed that the GPT models often computed the keyed value but answered from higher-precision files and misrounded. Naming the authoritative output recovered 6.3 to 7.9 points for three GPT models. The finding is harness-specific, but it shows why a correct tool call does not guarantee an evidence-grounded answer.", "why_it_matters": "Retrieving a simulator report from files changed which value some models trusted even when the underlying computation stayed fixed.", "limitations": [ "The finding is harness-specific, but it shows why a correct tool call does not guarantee an evidence-grounded answer." ], "importance": 8, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-026/the-agent-computed-the-right-number-and-cited-the-wrong-evidence", "json_url": "https://themachinepress.com/story/mp-2026-10-20-026.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-026-001", "text": "Retrieving a simulator report from files changed which value some models trusted even when the underlying computation stayed fixed.", "source_ids": [ "source-2026-10-20-015" ], "qualification": "The finding is harness-specific, but it shows why a correct tool call does not guarantee an evidence-grounded answer." } ], "source_ids": [ "source-2026-10-20-015" ], "tags": [ "agent reliability", "tool output", "evidence" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-027", "source_story_id": "tmp-story-streamrps-radar-scene", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 16, "story_type": "dispatch", "section": "chips-infrastructure", "editorial_classification": "editorial", "headline": "The Car's Radar Draws the Scene at 400 Frames a Second", "slug": "the-car-s-radar-draws-the-scene-at-400-frames-a-second", "dek": "StreamRPS labels spinning-radar data with completed LiDAR scenes, then performs streaming semantic and panoptic segmentation by sector.", "summary": "StreamRPS labels spinning-radar data with completed LiDAR scenes, then performs streaming semantic and panoptic segmentation by sector.", "body_text": "Cameras and LiDAR can degrade in darkness or bad weather, while spinning radar remains available but difficult to interpret. StreamRPS pairs labels built from LiDAR scene completion with a radar-specific transformer that preserves range and angular structure. On the SAX and BOREAS datasets, the authors report at least a 1.4-point gain over the strongest evaluated baseline and throughput above 400 frames per second in a four-sector configuration. They also release manual annotations for 100 SAX scans. This is not an autonomous-driving safety certification, but it pushes radar from sparse detections toward real-time scene structure.", "why_it_matters": "StreamRPS labels spinning-radar data with completed LiDAR scenes, then performs streaming semantic and panoptic segmentation by sector.", "limitations": [ "This is not an autonomous-driving safety certification, but it pushes radar from sparse detections toward real-time scene structure." ], "importance": 8, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-027/the-car-s-radar-draws-the-scene-at-400-frames-a-second", "json_url": "https://themachinepress.com/story/mp-2026-10-20-027.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-027-001", "text": "StreamRPS labels spinning-radar data with completed LiDAR scenes, then performs streaming semantic and panoptic segmentation by sector.", "source_ids": [ "source-2026-10-20-016" ], "qualification": "This is not an autonomous-driving safety certification, but it pushes radar from sparse detections toward real-time scene structure." } ], "source_ids": [ "source-2026-10-20-016" ], "tags": [ "radar", "autonomous driving", "scene understanding" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-015", "source_story_id": "tmp-brief-data-center-grid-control", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 17, "story_type": "ticker", "section": "infrastructure", "editorial_classification": "editorial", "headline": "The Data Center Learns Which Oscillation Came From Where", "slug": "the-data-center-learns-which-oscillation-came-from-where", "dek": "A source-selective storage controller separates computing-load fluctuations from grid disturbances while preserving prescribed support.", "summary": "A source-selective storage controller separates computing-load fluctuations from grid disturbances while preserving prescribed support.", "body_text": "Large computing loads can create oscillations even as operators expect them to help the grid. This work derives conditions for distinguishing disturbance sources and a limit on simultaneous isolation and support. A local storage controller outperformed equal-order comparators in two simulated networks while meeting the required grid response. The evidence is a control study, not an operating data-center deployment.", "why_it_matters": "A source-selective storage controller separates computing-load fluctuations from grid disturbances while preserving prescribed support.", "limitations": [ "The evidence is a control study, not an operating data-center deployment." ], "importance": 7, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-015/the-data-center-learns-which-oscillation-came-from-where", "json_url": "https://themachinepress.com/story/mp-2026-10-20-015.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-015-001", "text": "A source-selective storage controller separates computing-load fluctuations from grid disturbances while preserving prescribed support.", "source_ids": [ "source-2026-10-20-017" ], "qualification": "The evidence is a control study, not an operating data-center deployment." } ], "source_ids": [ "source-2026-10-20-017" ], "tags": [ "data centers", "power grid", "control" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-016", "source_story_id": "tmp-brief-thrust-assisted-trotting", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 18, "story_type": "ticker", "section": "robotics", "editorial_classification": "editorial", "headline": "The Biped Uses Thrusters to Keep Its Trot", "slug": "the-biped-uses-thrusters-to-keep-its-trot", "dek": "Harpy's controller allocates ground force and active thrust at 500 hertz while solving each optimization inside a two-millisecond budget.", "summary": "Harpy's controller allocates ground force and active thrust at 500 hertz while solving each optimization inside a two-millisecond budget.", "body_text": "Inspired by birds that use wing thrust during locomotion, researchers implemented thrust-assisted trotting on the Harpy robot. Its quadratic-program controller regulated contact and thrust at 500 hertz; reported solves took 0.18 milliseconds. In two low-speed experiments, mean velocity tracked its reference and joint error stayed within one degree. The trials are limited, but they demonstrate the controller on hardware.", "why_it_matters": "Harpy's controller allocates ground force and active thrust at 500 hertz while solving each optimization inside a two-millisecond budget.", "limitations": [], "importance": 7, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-016/the-biped-uses-thrusters-to-keep-its-trot", "json_url": "https://themachinepress.com/story/mp-2026-10-20-016.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-016-001", "text": "Harpy's controller allocates ground force and active thrust at 500 hertz while solving each optimization inside a two-millisecond budget.", "source_ids": [ "source-2026-10-20-018" ], "qualification": null } ], "source_ids": [ "source-2026-10-20-018" ], "tags": [ "bipedal robots", "model predictive control", "thrusters" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-017", "source_story_id": "tmp-brief-soma-reasoning-monitor", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 19, "story_type": "ticker", "section": "safety", "editorial_classification": "editorial", "headline": "A Fast Monitor Interrupts the Slow Reasoner", "slug": "a-fast-monitor-interrupts-the-slow-reasoner", "dek": "SOMA watches a reasoning trace, warns first and takes over only when a named failure persists.", "summary": "SOMA watches a reasoning trace, warns first and takes over only when a named failure persists.", "body_text": "SOMA pairs a fast decision model with a slower reasoner through scenarios, continuous judgments and predefined interventions. Without task-specific training, the shared monitor cut reasoning tokens by as much as 36 percent while improving average accuracy on overthinking tests, and raised abstention on unanswerable questions from 70.5 to 87.2 percent. It depends on visible text and calibrated monitors; it is not oversight of hidden reasoning.", "why_it_matters": "SOMA watches a reasoning trace, warns first and takes over only when a named failure persists.", "limitations": [ "It depends on visible text and calibrated monitors; it is not oversight of hidden reasoning." ], "importance": 7, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-017/a-fast-monitor-interrupts-the-slow-reasoner", "json_url": "https://themachinepress.com/story/mp-2026-10-20-017.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-017-001", "text": "SOMA watches a reasoning trace, warns first and takes over only when a named failure persists.", "source_ids": [ "source-2026-10-20-019" ], "qualification": "It depends on visible text and calibrated monitors; it is not oversight of hidden reasoning." } ], "source_ids": [ "source-2026-10-20-019" ], "tags": [ "reasoning", "monitoring", "abstention" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-018", "source_story_id": "tmp-brief-correlated-decoder-distance", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 20, "story_type": "ticker", "section": "chips-infrastructure", "editorial_classification": "editorial", "headline": "A Smarter Decoder Quietly Shortens the Code", "slug": "a-smarter-decoder-quietly-shortens-the-code", "dek": "Standard correlated reweighting can favor corrections built from too few assumed faults, reducing effective surface-code distance below threshold.", "summary": "Standard correlated reweighting can favor corrections built from too few assumed faults, reducing effective surface-code distance below threshold.", "body_text": "Correlated surface-code decoders reuse one correction to reweight another graph. The authors show that one first-pass edge can over-discount several correlated edges, causing failures on fault sets a distance-d code should correct. Regularizing the rule restored nominal distance scaling and retained an advantage in their circuit-level noise study. The result is simulated, but it identifies a concrete failure in a seemingly helpful heuristic.", "why_it_matters": "Standard correlated reweighting can favor corrections built from too few assumed faults, reducing effective surface-code distance below threshold.", "limitations": [], "importance": 7, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-018/a-smarter-decoder-quietly-shortens-the-code", "json_url": "https://themachinepress.com/story/mp-2026-10-20-018.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-018-001", "text": "Standard correlated reweighting can favor corrections built from too few assumed faults, reducing effective surface-code distance below threshold.", "source_ids": [ "source-2026-10-20-020" ], "qualification": null } ], "source_ids": [ "source-2026-10-20-020" ], "tags": [ "quantum error correction", "surface code", "decoding" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-019", "source_story_id": "tmp-brief-tokengate-prefill", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 21, "story_type": "ticker", "section": "infrastructure", "editorial_classification": "editorial", "headline": "The Model Skims Before It Reads", "slug": "the-model-skims-before-it-reads", "dek": "TokenGate learns which long-context tokens can stop moving through later layers during prefill.", "summary": "TokenGate learns which long-context tokens can stop moving through later layers during prefill.", "body_text": "TokenGate adds a lightweight importance module and token-gated attention so existing models can prune input during prefill. Across reported tasks, time to first token improved by 1.4 to 11.3 times and peak memory fell by 1.5 to 3.4 times without measured accuracy loss. Gains depend on model, task and hardware, but the component extends both standard and hybrid-attention transformers through parameter-efficient adaptation.", "why_it_matters": "TokenGate learns which long-context tokens can stop moving through later layers during prefill.", "limitations": [], "importance": 7, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-019/the-model-skims-before-it-reads", "json_url": "https://themachinepress.com/story/mp-2026-10-20-019.json", "first_published_at": "2026-10-20T09:00:00.000-04:00", "modified_at": "2026-10-20T09:00:00.000-04:00", "content_status": "new", "is_carryover": false, "carryover_reason": null, "key_claims": [ { "claim_id": "claim-mp-2026-10-20-019-001", "text": "TokenGate learns which long-context tokens can stop moving through later layers during prefill.", "source_ids": [ "source-2026-10-20-021" ], "qualification": null } ], "source_ids": [ "source-2026-10-20-021" ], "tags": [ "inference", "token pruning", "memory efficiency" ], "image_url": null, "corrections": [] }, { "story_id": "mp-2026-10-20-020", "source_story_id": "invention-space-drums-2", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 22, "story_type": "invention_desk", "section": "invention-desk", "editorial_classification": "editorial", "headline": "Space Drums 2.0", "slug": "space-drums-2-0", "dek": "Moves motion tracking onto two ESP32-S3 drumsticks, adds haptic hits and a two-pedal hub, and sends velocity-sensitive drum events to open PC or Android playback software.", "summary": "Moves motion tracking onto two ESP32-S3 drumsticks, adds haptic hits and a two-pedal hub, and sends velocity-sensitive drum events to open PC or Android playback software.", "body_text": "Moves motion tracking onto two ESP32-S3 drumsticks, adds haptic hits and a two-pedal hub, and sends velocity-sensitive drum events to open PC or Android playback software.", "why_it_matters": "An independent builder is turning an improbable idea into a working project.", "limitations": [ "A Desk Pick is an editorial selection, not a product endorsement." ], "importance": null, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-020/space-drums-2-0", "json_url": "https://themachinepress.com/story/mp-2026-10-20-020.json", "first_published_at": "2026-09-27T09:00:00.000-04:00", "modified_at": "2026-09-27T09:00:00.000-04:00", "content_status": "carried_over", "is_carryover": true, "carryover_reason": "invention_desk_seven_day_placement", "key_claims": [], "source_ids": [], "tags": [ "independent-builders", "desk-pick", "released" ], "image_url": "https://themachinepress.com/issues/2026-10-20/invention-desk/space-drums-2.png", "corrections": [] }, { "story_id": "mp-2026-10-20-021", "source_story_id": "invention-3ric", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 23, "story_type": "invention_desk", "section": "invention-desk", "editorial_classification": "editorial", "headline": "3RIC", "slug": "3ric", "dek": "Pairs a chip-level 65C02 computer with custom hardware, video, sound, floppy, and microSD support with a cycle-honest emulator and browser assembler built around the same machine.", "summary": "Pairs a chip-level 65C02 computer with custom hardware, video, sound, floppy, and microSD support with a cycle-honest emulator and browser assembler built around the same machine.", "body_text": "Pairs a chip-level 65C02 computer with custom hardware, video, sound, floppy, and microSD support with a cycle-honest emulator and browser assembler built around the same machine.", "why_it_matters": "An independent builder is turning an improbable idea into a working project.", "limitations": [ "A Desk Pick is an editorial selection, not a product endorsement." ], "importance": null, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-021/3ric", "json_url": "https://themachinepress.com/story/mp-2026-10-20-021.json", "first_published_at": "2026-09-27T09:00:00.000-04:00", "modified_at": "2026-09-27T09:00:00.000-04:00", "content_status": "carried_over", "is_carryover": true, "carryover_reason": "invention_desk_seven_day_placement", "key_claims": [], "source_ids": [], "tags": [ "independent-builders", "desk-pick", "released" ], "image_url": "https://themachinepress.com/issues/2026-10-20/invention-desk/3ric.png", "corrections": [] }, { "story_id": "mp-2026-10-20-022", "source_story_id": "invention-m3-crete", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 24, "story_type": "invention_desk", "section": "invention-desk", "editorial_classification": "editorial", "headline": "M3-CRETE", "slug": "m3-crete", "dek": "Publishes a pallet-scale cartesian motion reference, CAD, BOM, and low-voltage controls for supervised cementitious-extrusion research while explicitly excluding certified structural use.", "summary": "Publishes a pallet-scale cartesian motion reference, CAD, BOM, and low-voltage controls for supervised cementitious-extrusion research while explicitly excluding certified structural use.", "body_text": "Publishes a pallet-scale cartesian motion reference, CAD, BOM, and low-voltage controls for supervised cementitious-extrusion research while explicitly excluding certified structural use.", "why_it_matters": "An independent builder is turning an improbable idea into a working project.", "limitations": [ "A Desk Pick is an editorial selection, not a product endorsement." ], "importance": null, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-022/m3-crete", "json_url": "https://themachinepress.com/story/mp-2026-10-20-022.json", "first_published_at": "2026-09-27T09:00:00.000-04:00", "modified_at": "2026-09-27T09:00:00.000-04:00", "content_status": "carried_over", "is_carryover": true, "carryover_reason": "invention_desk_seven_day_placement", "key_claims": [], "source_ids": [], "tags": [ "independent-builders", "desk-pick", "prototype" ], "image_url": "https://themachinepress.com/issues/2026-10-20/invention-desk/m3-crete.png", "corrections": [] }, { "story_id": "mp-2026-10-20-023", "source_story_id": "invention-desknode", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 25, "story_type": "invention_desk", "section": "invention-desk", "editorial_classification": "editorial", "headline": "DeskNode", "slug": "desknode", "dek": "Turns a bare ESP32-S3 touchscreen into a USB-connected desk instrument for PC activity and optional room temperature, humidity, and light measurements.", "summary": "Turns a bare ESP32-S3 touchscreen into a USB-connected desk instrument for PC activity and optional room temperature, humidity, and light measurements.", "body_text": "Turns a bare ESP32-S3 touchscreen into a USB-connected desk instrument for PC activity and optional room temperature, humidity, and light measurements.", "why_it_matters": "An independent builder is turning an improbable idea into a working project.", "limitations": [ "A Desk Pick is an editorial selection, not a product endorsement." ], "importance": null, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-023/desknode", "json_url": "https://themachinepress.com/story/mp-2026-10-20-023.json", "first_published_at": "2026-09-27T09:00:00.000-04:00", "modified_at": "2026-09-27T09:00:00.000-04:00", "content_status": "carried_over", "is_carryover": true, "carryover_reason": "invention_desk_seven_day_placement", "key_claims": [], "source_ids": [], "tags": [ "independent-builders", "desk-pick", "prototype" ], "image_url": "https://themachinepress.com/issues/2026-10-20/invention-desk/desknode.png", "corrections": [] }, { "story_id": "mp-2026-10-20-024", "source_story_id": "invention-sponsored-house-example", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 26, "story_type": "invention_desk", "section": "invention-desk", "editorial_classification": "house_example", "headline": "The First Paid Slot", "slug": "the-first-paid-slot", "dek": "A transparent preview of paid placement with one verified link and no claim of endorsement.", "summary": "A transparent preview of paid placement with one verified link and no claim of endorsement.", "body_text": "A transparent preview of paid placement with one verified link and no claim of endorsement.\n\nHouse example - no advertiser paid. Payment will buy placement, never endorsement.", "why_it_matters": "This placement explains how builders can appear in The Invention Desk without purchasing editorial endorsement.", "limitations": [ "House example - no advertiser paid. Payment will buy placement, never endorsement." ], "importance": null, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-024/the-first-paid-slot", "json_url": "https://themachinepress.com/story/mp-2026-10-20-024.json", "first_published_at": "2026-07-10T09:00:00.000-04:00", "modified_at": "2026-07-10T09:00:00.000-04:00", "content_status": "carried_over", "is_carryover": true, "carryover_reason": "invention_desk_seven_day_placement", "key_claims": [], "source_ids": [], "tags": [ "independent-builders", "sponsored-project", "open" ], "image_url": "https://themachinepress.com/issues/2026-10-20/invention-desk/sponsored-project.webp", "corrections": [] }, { "story_id": "mp-2026-10-20-025", "source_story_id": "invention-placement-cta", "edition_id": "mp-2026-10-20-morning-0102", "edition_url": "https://themachinepress.com/edition/2026-10-20", "position": 27, "story_type": "invention_desk", "section": "invention-desk", "editorial_classification": "house_example", "headline": "Put Your Project on the Desk", "slug": "put-your-project-on-the-desk", "dek": "One manually reviewed placement stays active for seven days and remains separate from Desk Picks.", "summary": "One manually reviewed placement stays active for seven days and remains separate from Desk Picks.", "body_text": "One manually reviewed placement stays active for seven days and remains separate from Desk Picks.\n\nManual intake only. Payment buys placement, never endorsement, and every submission is reviewed.", "why_it_matters": "This placement explains how builders can appear in The Invention Desk without purchasing editorial endorsement.", "limitations": [ "Manual intake only. Payment buys placement, never endorsement, and every submission is reviewed." ], "importance": null, "canonical_url": "https://themachinepress.com/story/mp-2026-10-20-025/put-your-project-on-the-desk", "json_url": "https://themachinepress.com/story/mp-2026-10-20-025.json", "first_published_at": "2026-07-10T09:00:00.000-04:00", "modified_at": "2026-07-10T09:00:00.000-04:00", "content_status": "carried_over", "is_carryover": true, "carryover_reason": "invention_desk_seven_day_placement", "key_claims": [], "source_ids": [], "tags": [ "independent-builders", "cta", "open" ], "image_url": "https://themachinepress.com/issues/2026-10-20/invention-desk/put-your-project-on-the-desk.webp", "corrections": [] } ], "sources": [ { "source_id": "source-2026-10-20-001", "title": "arXiv preprint 2610.26846, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26846v1", "canonical_url": "https://arxiv.org/abs/2610.26846v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:59:17.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-001-001" ] }, { "source_id": "source-2026-10-20-002", "title": "arXiv preprint 2610.26792, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26792v1", "canonical_url": "https://arxiv.org/abs/2610.26792v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:40:03.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-002-001" ] }, { "source_id": "source-2026-10-20-003", "title": "arXiv preprint 2610.26861, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26861v1", "canonical_url": "https://arxiv.org/abs/2610.26861v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:59:58.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-003-001" ] }, { "source_id": "source-2026-10-20-004", "title": "arXiv preprint 2610.26863, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26863v1", "canonical_url": "https://arxiv.org/abs/2610.26863v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:59:59.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-004-001" ] }, { "source_id": "source-2026-10-20-005", "title": "arXiv preprint 2610.26860, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26860v1", "canonical_url": "https://arxiv.org/abs/2610.26860v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:59:58.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-005-001" ] }, { "source_id": "source-2026-10-20-006", "title": "arXiv preprint 2610.26855, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26855v1", "canonical_url": "https://arxiv.org/abs/2610.26855v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:59:55.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-006-001" ] }, { "source_id": "source-2026-10-20-007", "title": "arXiv preprint 2610.26854, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26854v1", "canonical_url": "https://arxiv.org/abs/2610.26854v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:59:54.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-007-001" ] }, { "source_id": "source-2026-10-20-008", "title": "arXiv preprint 2610.26847, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26847v1", "canonical_url": "https://arxiv.org/abs/2610.26847v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:59:19.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-008-001" ] }, { "source_id": "source-2026-10-20-009", "title": "arXiv preprint 2610.26852, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26852v1", "canonical_url": "https://arxiv.org/abs/2610.26852v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:59:49.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-009-001" ] }, { "source_id": "source-2026-10-20-010", "title": "arXiv preprint 2610.26848, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26848v1", "canonical_url": "https://arxiv.org/abs/2610.26848v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:59:30.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-010-001" ] }, { "source_id": "source-2026-10-20-011", "title": "arXiv preprint 2610.26844, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26844v1", "canonical_url": "https://arxiv.org/abs/2610.26844v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:59:05.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-011-001" ] }, { "source_id": "source-2026-10-20-012", "title": "arXiv preprint 2610.26840, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26840v1", "canonical_url": "https://arxiv.org/abs/2610.26840v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:58:41.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-012-001" ] }, { "source_id": "source-2026-10-20-013", "title": "arXiv preprint 2610.26829, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26829v1", "canonical_url": "https://arxiv.org/abs/2610.26829v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:55:38.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-013-001" ] }, { "source_id": "source-2026-10-20-014", "title": "arXiv preprint 2610.26828, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26828v1", "canonical_url": "https://arxiv.org/abs/2610.26828v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:55:35.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-014-001" ] }, { "source_id": "source-2026-10-20-015", "title": "arXiv preprint 2610.26836, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26836v1", "canonical_url": "https://arxiv.org/abs/2610.26836v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:57:41.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-026-001" ] }, { "source_id": "source-2026-10-20-016", "title": "arXiv preprint 2610.26816, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26816v1", "canonical_url": "https://arxiv.org/abs/2610.26816v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:51:49.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-027-001" ] }, { "source_id": "source-2026-10-20-017", "title": "arXiv preprint 2610.26812, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26812v1", "canonical_url": "https://arxiv.org/abs/2610.26812v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:50:47.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-015-001" ] }, { "source_id": "source-2026-10-20-018", "title": "arXiv preprint 2610.26810, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26810v1", "canonical_url": "https://arxiv.org/abs/2610.26810v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:50:01.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-016-001" ] }, { "source_id": "source-2026-10-20-019", "title": "arXiv preprint 2610.26793, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26793v1", "canonical_url": "https://arxiv.org/abs/2610.26793v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:40:21.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-017-001" ] }, { "source_id": "source-2026-10-20-020", "title": "arXiv preprint 2610.26785, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26785v1", "canonical_url": "https://arxiv.org/abs/2610.26785v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:35:59.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-018-001" ] }, { "source_id": "source-2026-10-20-021", "title": "arXiv preprint 2610.26754, original v1", "publisher": "arXiv", "url": "https://arxiv.org/abs/2610.26754v1", "canonical_url": "https://arxiv.org/abs/2610.26754v1", "source_type": "primary_research", "is_primary_source": true, "published_at": "2026-10-19T13:15:53.000-04:00", "accessed_at": "2026-10-20T08:15:06.176-04:00", "supports_claim_ids": [ "claim-mp-2026-10-20-019-001" ] } ], "corrections": [], "publisher": { "name": "The Machine Press", "url": "https://themachinepress.com", "description": "A daily newspaper for the age of artificial intelligence." } }