https://megadose.ai/i/1dc7b6c3-c9c0-43c2-ac1b-5b9906baa7cf Megadose en 2026-09-02T12:40:01+00:00 Valencia-based iPronics raised $125M co-led by Maverick Silicon and Light Street, with participation from Nvidia, to develop optical circuit switch tech (Stephen Nellis/Reuters) https://megadose.ai/i/a035cf29-3fb6-4c13-9964-ec410bb146b6 Megadose en 2026-09-02T10:15:15+00:00 Robotics and AI startup Lyte, founded by ex-Face ID engineers, raised $165M led by Maverick Silicon at a $1.6B valuation, taking its total funding to $272M (Samantha Kelly/Bloomberg) https://megadose.ai/i/713e0c73-e55b-4a63-84c8-5eb9f783ac1b Megadose en 2026-09-02T10:00:00+00:00 Fable 5.1, Enterprise Frontier Safeguards https://megadose.ai/i/14cd976c-53a1-4193-9181-5e4ba73fc36d Megadose en 2026-09-02T07:55:00+00:00 OpenAI launches a ChatGPT Health integration with Epic's EHR system and a new Healthcare Public Data plug-in that can fetch info from sources including PubMed (Ivan Mehta/TechCrunch) https://megadose.ai/i/86e014c4-441b-42f2-aaa3-c04045465f0a Megadose en 2026-09-02T07:46:08+00:00 [AINews] Claude Fable/Mythos 5.1: new SOTA model, 75% cache price cut but 70% more output tokens https://megadose.ai/i/713148d4-29ec-41bb-aef7-3632c02f5aa0 Megadose en 2026-09-02T06:30:00+00:00 A look at South Korea's sovereign AI efforts, including its planned $919B AI infrastructure investment and Nvidia's expanded ties with Samsung and SK Hynix (SemiAnalysis) https://megadose.ai/i/21c055bf-af0f-47b9-ab80-b33abde53346 Megadose en 2026-09-02T04:15:56+00:00 The Emergent Symbolic Structure of Artificial Neural Networks https://megadose.ai/i/afbb073e-5efe-4a12-810b-0fd6221f1114 Megadose en 2026-09-02T03:25:03+00:00 SciFin, which helps revenue teams converge fragmented information to understand business needs, raised a $44M seed co-led by Altimeter and Madrona (Kyt Dotson/SiliconANGLE) https://megadose.ai/i/722d57ef-d3f1-471e-964b-7c66f52129ba Megadose en 2026-09-02T00:55:01+00:00 Sources: Cognition is raising around $1B at a ~$47B valuation, up from $26B in May; Cognition received nearly $10B in investor interest (Bloomberg) https://megadose.ai/i/ccc81bf5-c3e1-432f-b65d-ca121f405d4b Megadose en 2026-09-01T23:59:59+00:00 Anthropic launches Claude Fable 5.1 and Mythos 5.1 https://megadose.ai/i/901a0b3a-7e36-4a00-acc1-74c2a65213f9 Megadose en 2026-09-01T23:52:37+00:00 Anthropic says Fable 5.1 will cost an estimated 25% less than Fable 5 for typical workloads and up to 45% less for highly agentic work (Zac Hall/9to5Mac) https://megadose.ai/i/d67596eb-10ea-40a7-a4e1-2a50127d3e2a Megadose en 2026-09-01T23:45:01+00:00 Anthropic prices Fable 5.1 at $10/1M input tokens and $50/1M output tokens, unchanged from Fable 5, and cuts cache read pricing by 75% to $0.25/1M tokens (Frederic Lardinois/The New Stack) https://megadose.ai/i/6cbcab93-e019-405d-b11a-e9c1b3523624 Megadose en 2026-09-01T23:20:00+00:00 IPO filing: SB Energy says it has 8.8GW of data center capacity contracted or under construction across Texas and Ohio and a $430B data center segment backlog (Monique Mulima/Bloomberg) https://megadose.ai/i/a149f4e2-a7ac-464a-9a5c-7c8e0c861bb0 Megadose en 2026-09-01T22:08:24+00:00 AfterQuery reportedly becomes Y Combinator’s fastest-ever unicorn, now valued at $3.2B https://megadose.ai/i/1a0c59d1-ecaa-43e6-96bf-1031fa3744b8 Megadose en 2026-09-01T21:06:24+00:00 Open AI’s Astra model is on the way—and very good at breaking into computer systems https://megadose.ai/i/d63f8163-8be7-4252-a31e-81d5f691d528 Megadose en 2026-09-01T21:05:01+00:00 Palo Alto Networks reports Q4 revenue up 34% YoY to $3.41B, vs. $3.35B est., and acquires Console, which provides AI-powered IT service management (Samantha Subin/CNBC) https://megadose.ai/i/e3475f7f-f2cf-4820-b0d4-450148d6dd03 Megadose en 2026-09-01T20:55:02+00:00 Claude Fable 5.1 and Mythos 5.1 are Anthropic's first models to watermark text outputs; a detection API is available to eligible groups as required under EU law (Ben Patterson/PCWorld) https://megadose.ai/i/e604b606-144f-46e5-bbcd-827da977001b Megadose en 2026-09-01T20:20:00+00:00 Dell reports Q2 revenue up 58% YoY to $46.97B, vs. $44.92B est., and forecasts $192B in FY 2027 revenue, vs. $172.67B est.; DELL jumps 9%+ after hours (Jordan Novet/CNBC) https://megadose.ai/i/b10d6030-49ba-43f8-bd10-5a607c1b52f9 Megadose en 2026-09-01T20:00:50+00:00 Sources: AfterQuery, which sells coding and finance training data to AI labs, has hit a valuation of $3.2B, up from $300M in April, and is profitable (Anna Tong/Forbes) https://megadose.ai/i/3a114b11-b2d3-40be-952f-11fe70bce11a Megadose en 2026-09-01T19:50:01+00:00 Meta launches Muse Voice Transcribe, MSL's first real-time audio perception model, with streaming automatic speech recognition, trained with 70+ languages (Meta AI Research) https://megadose.ai/i/bb3f52a9-76c7-4d5d-963a-7a37fb23c8c0 Megadose en 2026-09-01T19:39:22+00:00 Anthropic’s new Fable release is cheaper, less restrictive https://megadose.ai/i/9f73033d-9aff-4b59-84a3-b83c5a99e167 Megadose en 2026-09-01T19:30:02+00:00 Wafer, which makes AI agents that optimize open-source models for a business's workload, raised a $40M Series A, a source says at a $200M+ valuation (Stephanie Palazzolo/The Information) https://megadose.ai/i/876db8a0-4c39-4031-ba1d-fafa2c2017ac Megadose en 2026-09-01T19:15:03+00:00 Fei-Fei Li's World Labs unveils Atlas, a multimodal world model that generates image/video frames with pixel-perfect camera control and reconstructs them in 3D (World Labs) https://megadose.ai/i/9516b16a-fba6-45e8-9a5e-3d232a9bd69e Megadose en 2026-09-01T19:00:36+00:00 Anthropic says Fable 5.1 sets new standards on coding, knowledge work, and long-running problem-solving tasks, and can fix the root causes of software issues (Carl Franzen/VentureBeat) https://megadose.ai/i/18da9520-596d-4ac0-8e6d-71e599648d6d Megadose en 2026-09-01T18:40:03+00:00 Aslan, which offers AI agents for the FBI and wider intelligence community that can pose as analysts and undercover spies in online forums, raised $20.8M (Sam Sabin/Axios) https://megadose.ai/i/338bacef-910a-44fc-a1c8-0549bafe57b3 Megadose en 2026-09-01T18:31:05+00:00 Anthropic launches Enterprise Frontier Safeguards to let businesses control how their data is reviewed, stored, and managed, after pushback from customers (Ashley Capoot/CNBC) https://megadose.ai/i/20b27de8-9c8b-44d9-bedd-a38fb91c4479 Megadose en 2026-09-01T18:23:45+00:00 Anthropic unveils Claude Fable 5.1, which is generally available, and Mythos 5.1, for trusted partners and with cybersecurity and life sciences safeguards (Anthropic) https://megadose.ai/i/babf73b6-ae2a-4dad-9f91-5d24321996ea Megadose en 2026-09-01T18:10:00+00:00 Félix, an AI-powered WhatsApp remittance platform focused on Latinos in the US, raised a $200M Series C co-led by a16z and GC, including $113M in debt (Mary Ann Azevedo/Crunchbase News) https://megadose.ai/i/74e86087-e93b-4711-aac1-e4f2a52c21c5 Megadose en 2026-09-01T17:59:49+00:00 Beyond Scores: Understanding LLM-as-a-Judge Mechanisms in Summarization Evaluation https://megadose.ai/i/7c00abaa-2f2e-4200-93b4-9463edf9960c Megadose en 2026-09-01T17:59:46+00:00 Efficient SWE Agent Benchmarking via Trajectory-Aware Evaluation https://megadose.ai/i/67b5ffce-ebf8-408b-995b-52fe483d4fde Megadose en 2026-09-01T17:59:39+00:00 Adaptive Critical Token-Aware Retrieval for Repository-Level Code Generation https://megadose.ai/i/ce8b994f-7304-4499-b72c-2d6ed855091e Megadose en 2026-09-01T17:59:13+00:00 CordisBench: Can Language Models Reason About Component Lifecycles in Dynamic Agent Harnesses? https://megadose.ai/i/f1f565d1-0689-49a9-b89b-2f0c15e85548 Megadose en 2026-09-01T17:58:07+00:00 Facet-0: A Robotic Foundation Model for Contact-Rich Precise Manipulation https://megadose.ai/i/0c72f81a-500a-4f21-a83b-c66b3436227b Megadose en 2026-09-01T17:57:50+00:00 Mechanism Design for Alignment and Control https://megadose.ai/i/658645e7-a08e-482c-9502-277ccac1383a Megadose en 2026-09-01T17:53:41+00:00 The Structure of Quantization Damage in LLMs: Why the Next Bit Should Be Spent Globally https://megadose.ai/i/71562303-8f78-4761-91f7-4619cbe1f088 Megadose en 2026-09-01T17:40:29+00:00 Closing Cost-Quality Gap in Document VLMs: Difficulty-Aware Data Curation and Quality-Adjusted Deployment Economics https://megadose.ai/i/6cdeaab8-c7fd-4be9-af8d-37b7798de7de Megadose en 2026-09-01T17:39:55+00:00 Scaling Near-Optimal SFT-RL Annotation Budget Allocation from Small to Large LLMs https://megadose.ai/i/9cdbde85-5dd8-4663-a195-ec4829d14abb Megadose en 2026-09-01T17:39:26+00:00 From Production Traffic to Post-Training: Building a Self-Hosted LLM That Covers the Corporate Request Mix https://megadose.ai/i/54d8d7db-6512-48bc-a552-da351553d3ad Megadose en 2026-09-01T17:35:37+00:00 Google’s answer to Canva is an AI tool where you prompt instead of design https://megadose.ai/i/69655d9e-b3c8-46fb-b2aa-2d50e9ad2309 Megadose en 2026-09-01T17:33:41+00:00 Selective Agent Guidance via Entropy: Learning Autonomous Policies from Imperfect VLM Teachers https://megadose.ai/i/93a08bee-96b0-4176-bfb0-53e8a18286ae Megadose en 2026-09-01T17:31:02+00:00 From Confusion to Clarity: Confusion-Aware Retrieval and Knowledge Injection for Text Classification https://megadose.ai/i/e734c14d-59c4-4f4c-8aef-8528e5e8efb4 Megadose en 2026-09-01T17:21:40+00:00 H3-World: Turning Language Understanding into World Control https://megadose.ai/i/3141fe5c-c16d-4084-98e7-2054bac953db Megadose en 2026-09-01T17:20:13+00:00 Gradient-Update Mismatch: Rethinking Conflict-Free Training of Physics-Informed Neural Networks https://megadose.ai/i/4c5fc3a4-ac90-4f5c-a3c4-fdb896f7eef3 Megadose en 2026-09-01T17:19:57+00:00 Retrieved but not ranked: surface-form bias in structural retrieval, from mathematics to agent trajectories https://megadose.ai/i/bb11b9e7-aa67-4860-b224-b1bad938284b Megadose en 2026-09-01T17:17:59+00:00 Can LLMs Discover Scientific Laws in Real and Parallel Worlds? https://megadose.ai/i/050536e8-c3e7-44e8-8fdd-93bd80ccb6b2 Megadose en 2026-09-01T17:16:21+00:00 A Mathematical Theory of Reusable Neural Bases for Network Compression https://megadose.ai/i/191c1e1b-c491-46b4-9df8-25b2bfba7668 Megadose en 2026-09-01T17:15:41+00:00 NashDreamer: Model-Based Reinforcement Learning for Zero-Sum Imperfect-Information Games https://megadose.ai/i/f66b9734-c0a5-410f-8cf1-521e7e8c262b Megadose en 2026-09-01T17:10:02+00:00 Empirik, which uses AI to predict and prevent IT outages by analyzing system changes and their effects, raised a $21M seed and spins out from Sequoia Capital (Marina Temkin/TechCrunch) https://megadose.ai/i/6174b93e-21de-498f-951a-f902df5741bf Megadose en 2026-09-01T17:08:51+00:00 Introducing agentic video understanding with Gemini https://megadose.ai/i/1a28f9cc-897f-4409-b427-9c8e2b444191 Megadose en 2026-09-01T17:04:47+00:00 Can LLMs Design Video Coding Tools? A Case Study on Planar Mode https://megadose.ai/i/3eb9416b-7338-48d3-b1f4-d0fe9589e725 Megadose en 2026-09-01T17:00:30+00:00 Knowledge Distillation During Mid-Training Favors Reasoning over Factual Recall https://megadose.ai/i/c75cd1d3-a312-480a-91da-1cf9d047f2b7 Megadose en 2026-09-01T17:00:00+00:00 ChatGPT Health adds Epic integration for clinicians to import patient data https://megadose.ai/i/f599cdb4-14ad-47ea-af3b-84f37e4f15bb Megadose en 2026-09-01T16:55:56+00:00 EvoSCM: Scientific Belief Revision Through Causal Model Evolution and Experimentation https://megadose.ai/i/2d20deb8-0a95-438e-974b-b0ba7bc65794 Megadose en 2026-09-01T16:55:11+00:00 Relational-Core Graph Analytics Querying graphs at SQL scale, and why the node/edge model is a performance tax, not a truer picture of connected data https://megadose.ai/i/62917d34-0dc6-47d8-90f8-515bee25cc2d Megadose en 2026-09-01T16:48:13+00:00 When Guardrails Look Effective: Construct Validity Failures in LLM Agent Commerce Evaluation https://megadose.ai/i/4743c455-ba89-4ecf-adad-64c0e93ee209 Megadose en 2026-09-01T16:45:02+00:00 TempCloze: Can Video-LLMs Identify the Missing Middle? https://megadose.ai/i/cd1f2e5f-caa3-41c8-a39a-aec5438117c2 Megadose en 2026-09-01T16:38:15+00:00 LatentPress: Context Compression Beyond Text and Vision https://megadose.ai/i/14349dbf-90c3-445c-8fc2-02b1eb9a0d4a Megadose en 2026-09-01T16:31:04+00:00 Sequoia-incubated Empirik launches with $21M to predict outages before they happen https://megadose.ai/i/6a1fda81-074b-4751-8965-62b2703d31ae Megadose en 2026-09-01T16:24:24+00:00 Visko launches Orbis, a new real-time AI video model https://megadose.ai/i/333f1f69-3f95-4d74-9bb8-8f5f94cadeb5 Megadose en 2026-09-01T16:22:07+00:00 GlossoGen: Emergent Language in Complex Multi-Agent LLM Interactions https://megadose.ai/i/5602b80c-65fe-4b71-a183-1cb024c4ae74 Megadose en 2026-09-01T16:20:00+00:00 AI power infrastructure company SB Energy, which is backed by SoftBank, OpenAI, and Nvidia, files for an IPO, and says it is "substantially dependent" on OpenAI (CJ Haddad/CNBC) https://megadose.ai/i/4d20871e-0424-4927-a2a0-fca25b1741f1 Megadose en 2026-09-01T16:19:31+00:00 Defense-as-Skill: Evolving Runtime Guard Skill for Skill-Augmented Agents https://megadose.ai/i/da07854f-330e-490e-8b01-45983fa20b19 Megadose en 2026-09-01T16:17:18+00:00 Harness-of-Harness: Multi-Day Autonomous Software Development with Continual Improvement https://megadose.ai/i/c06a6265-3149-47de-bb20-f903dce37171 Megadose en 2026-09-01T16:10:01+00:00 Google launches Google Pics, a creative design suite powered by Gemini and Nano Banana to compete with Canva and Adobe Express, available as a Workspace app (Jess Weatherbed/The Verge) https://megadose.ai/i/587bd9ab-b59a-4130-9110-5b8eb4c297a5 Megadose en 2026-09-01T16:03:54+00:00 Parsing the Stream: A Live Trace Model for Long-Horizon Agents and Their Observers https://megadose.ai/i/b3ca0aae-389c-427f-9a4b-ebcbf679f2b4 Megadose en 2026-09-01T15:59:32+00:00 When Safety Routing Breaks: Understanding Alignment Fragility under Benign Fine-Tuning https://megadose.ai/i/fa8b4d37-c6c8-4801-b36a-a9a7798eb16c Megadose en 2026-09-01T15:55:03+00:00 Air Security, which builds a security service for extensions and other tools installed on AI agents, emerges from stealth with $50M led by Sequoia and Greenoaks (Meir Orbach/CTech) https://megadose.ai/i/75171f67-9dc4-4912-85f7-df9ed569ad37 Megadose en 2026-09-01T15:55:01+00:00 Diffusion as a Training Curriculum for Timestep-Free Iterative Reasoning https://megadose.ai/i/75ca975e-73c9-4d73-a6e7-fbf51c7812c3 Megadose en 2026-09-01T15:45:51+00:00 AIR raises $50M to help companies vet the skills and add-ons AI agents use https://megadose.ai/i/2b7b5f3a-d201-4b27-a435-7472d846bc2b Megadose en 2026-09-01T15:45:33+00:00 HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness? https://megadose.ai/i/ce308709-1922-4447-a747-56df1eb74d68 Megadose en 2026-09-01T15:41:59+00:00 Efficiently Estimating Optimal Hyperparameter Scaling Laws through Power-Law Entropy Search https://megadose.ai/i/9d2b97c8-d81f-434e-b773-8d62d632ac9a Megadose en 2026-09-01T15:40:09+00:00 TRIAGE: Three-level Routing and Intelligent Agent Guidance for Efficient Execution https://megadose.ai/i/c81deebb-83eb-4e33-9d6e-7b6442605d32 Megadose en 2026-09-01T15:40:03+00:00 DataAgent, which is developing AI agents that autonomously fix failures inside companies' own cloud infrastructure, emerges from stealth with a $10M pre-seed (CTech) https://megadose.ai/i/1432974d-119c-4850-9280-05981c4048d2 Megadose en 2026-09-01T15:20:01+00:00 Perplexity launches Hybrid Compute, which splits workloads between frontier cloud models like Opus 5 and local LLMs, for all users of its Mac app (Igor Bonifacic/Engadget) https://megadose.ai/i/6de8b9d1-649d-4b66-a736-880beb82110e Megadose en 2026-09-01T15:10:58+00:00 Fambot introduces an ‘AI chief of staff’ for families https://megadose.ai/i/30e25b92-6215-4cbd-bd0c-5f8bab64bccf Megadose en 2026-09-01T14:55:00+00:00 Sonos opens its platform to third-party AI assistants, upgrades Sonos 27voice assistant with an in-house LLM, and plans to let users create their own AI agents (Chris Welch/Bloomberg) https://megadose.ai/i/6ab57808-4b7f-482b-af49-9aae724d9d2e Megadose en 2026-09-01T14:45:37+00:00 PSI, which aims to use AI "to industrialize the discovery of new physics" including cracking interstellar travel, emerges from stealth with a $58M seed (Sabrina Ortiz/The Deep View) https://megadose.ai/i/41b92434-b9b4-4fc5-a781-4f2ee0d6e982 Megadose en 2026-09-01T14:35:00+00:00 Waymo starts offering driverless rides to customers in San Diego, Tampa, Florida, and Denver; Amazon's Zoox plans to start testing in Houston and San Diego (Lora Kolodny/CNBC) https://megadose.ai/i/7e8f3e5b-4f47-4094-8aae-6cee8d4b3b7a Megadose en 2026-09-01T13:10:06+00:00 Nvidia plans to launch DLSS 5 on September 3 on RTX 50-series desktop and laptop GPUs and through GeForce Now, following a divisive announcement in March (Jay Peters/The Verge) https://megadose.ai/i/8920523b-e212-4949-98fa-96fb530ff74e Megadose en 2026-09-01T13:00:00+00:00 Path to Astra: critical capabilities and frontier safeguards https://megadose.ai/i/6fda77db-e6d6-44bd-83e9-e9cf1851f47b Megadose en 2026-09-01T12:20:19+00:00 On the Design Fundamentals of Pixel Text Representation Learning https://megadose.ai/i/46af381c-b9ac-49e8-ac8e-ee843de01469 Megadose en 2026-09-01T12:20:05+00:00 StainPresetNet: Stain Preset Network for Fast Multi-to-Multi Stain Normalization https://megadose.ai/i/33539cbc-9e45-4f11-ab91-2ec44fa928a0 Megadose en 2026-09-01T12:17:43+00:00 Revisiting Face Recognition for Monozygotic Twins: The Celeb Twins Test Set https://megadose.ai/i/a26d8046-6d4a-4c89-a1cc-d773f49b6fa8 Megadose en 2026-09-01T12:15:53+00:00 Does task decomposition improve automatic NLG evaluation? https://megadose.ai/i/c86a8831-ad34-4318-a06e-36ffd93d3cb2 Megadose en 2026-09-01T12:11:29+00:00 Overfitting Mitigation via Singular Value Decomposition in Minimum Bayes Risk Decoding https://megadose.ai/i/04ed4efd-b0d0-42ba-a61c-0cef95baf085 Megadose en 2026-09-01T12:05:11+00:00 Scaled Idempotence in Transformer Attention: Paired OV Geometry and Shared-Value Algebras https://megadose.ai/i/c9c35f59-430d-496f-b6ef-59b0902d65ff Megadose en 2026-09-01T12:04:08+00:00 When Does Online Adaptation Pay on the Edge? A Leakage-Free Evaluation of Warmup, Learning-Rate Selection, and Resource Trade-offs for Time-Series Forecasting https://megadose.ai/i/20f41053-19b7-423a-b49c-a4fcee022a61 Megadose en 2026-09-01T12:00:00+00:00 Healthcare organizations can now connect EHR and additional industry data to ChatGPT https://megadose.ai/i/0aaf8216-f70d-4c4f-bcc0-ecc28f8ed9bd Megadose en 2026-09-01T11:59:00+00:00 DNC-IMM: Early Lane-Change Intention Recognition via Neural Calibration Based on Driving Context Information https://megadose.ai/i/98ab8126-86a9-4c8c-bf16-3d30fbed775f Megadose en 2026-09-01T11:56:17+00:00 Latent Recurrent Thoughts: Recurrent Refinement of Proposed Latents for Reasoning with Frozen LLMs https://megadose.ai/i/0cf2a9e6-a839-457b-8019-65ba0194ba03 Megadose en 2026-09-01T11:51:49+00:00 EDRAC: Benchmarking Arabic Dialect Reading Comprehension https://megadose.ai/i/03a93054-91c1-4a9e-9723-6f076d77b60c Megadose en 2026-09-01T11:51:37+00:00 ClinTraceBench: Source-Verifiable Longitudinal Clinical Reasoning over EHR-Derived Dialogues https://megadose.ai/i/f4d9210f-1db9-47cb-9804-f91fda9f7501 Megadose en 2026-09-01T11:48:16+00:00 Replicating TRACE: A Practitioner's Guide to Its Threshold and Particle Budget https://megadose.ai/i/dc54a560-0cd4-4b14-90a5-92002e389aea Megadose en 2026-09-01T11:47:39+00:00 Hints Help But Do They Teach? Evaluating Skills Transfer in Code Generation https://megadose.ai/i/4c426af8-f7be-4c21-b622-86af270a3a59 Megadose en 2026-09-01T11:46:56+00:00 When Modality Gap Reduction Fails: Prediction-Level Hubness in CLIP https://megadose.ai/i/551bcf20-5ee3-4c4c-844b-c6199aacf8c7 Megadose en 2026-09-01T11:46:28+00:00 Neural Symbollic Regression Using Deep Learning and Sparse Modelling https://megadose.ai/i/69ccdf7e-d6cd-4873-9d2d-6ea926154752 Megadose en 2026-09-01T11:43:41+00:00 Beyond Magnitude: Contrastive Routing for Modular Mixture-of-Experts https://megadose.ai/i/1885dd12-2d70-4c2d-95b0-cecec5e6169b Megadose en 2026-09-01T11:32:53+00:00 Subliminal Learning as Trait-Direction Drift: A Mechanism and Targeted Control under SFT Distillation https://megadose.ai/i/ff60fd4e-1517-4525-a0fd-d8a6ec6ca600 Megadose en 2026-09-01T11:29:03+00:00 Modelpedia: A Catalog of Model Findings for the Meta-Science of AI https://megadose.ai/i/798b67ed-87f8-4e2f-a5be-f9bd8c5994b4 Megadose en 2026-09-01T11:13:26+00:00 StateSwap: Probing Support-Elimination Hidden States in Multiple-Choice Questions https://megadose.ai/i/8bc26e7b-73db-4e17-9b01-cde68ad13e26 Megadose en 2026-09-01T11:07:18+00:00 Text-guided flow matching enables sample-efficient crystal structure generation https://megadose.ai/i/3abf3020-0bd7-4739-8bd3-bd802e1a3b0b Megadose en 2026-09-01T11:03:56+00:00 Post-hoc Alignment of LLM-judges to Human Judgment Distribution https://megadose.ai/i/0b3e1112-4817-444d-b2d8-903bba69a6e1 Megadose en 2026-09-01T11:03:30+00:00 Let Confidence Change, Not the Prediction: Prediction-Preserving Repair for Post-hoc Calibration https://megadose.ai/i/def26efa-8b94-447d-a2b6-23de0c93d48b Megadose en 2026-09-01T11:00:39+00:00 OUTLETS: Output-Length Prediction from Speculative Decoding Backbones https://megadose.ai/i/5d9eedc9-b1f6-4cef-9735-60cd6fc242ea Megadose en 2026-09-01T10:54:13+00:00 ARISE-RL: Agentic Rubric-Grounded Iterative Self-Evolution with Reinforcement Learning https://megadose.ai/i/7e869bed-347b-43c8-861d-06f99ee110a6 Megadose en 2026-09-01T10:53:07+00:00 WorldBench: Culturally Grounded Benchmark for Multilingual Agents https://megadose.ai/i/deea9e8c-fffa-4049-9984-5ca7739a50ad Megadose en 2026-09-01T10:47:11+00:00 SAGE: Subpopulation-Aware Generative Enhancement for Mitigating Spurious Correlations https://megadose.ai/i/ee8b2661-144d-4d60-b503-605f1ce53806 Megadose en 2026-09-01T10:46:42+00:00 Lagged Coupling: Internal Representations Become Readable Before They Become Causal https://megadose.ai/i/43d16aee-4d10-4dc5-a882-9ce0b6840f8e Megadose en 2026-09-01T10:45:11+00:00 AgentFactory: Towards Automated Agentic System Design and Optimization https://megadose.ai/i/50dd5f10-ca7a-4d1e-8c90-015fe571740f Megadose en 2026-09-01T10:43:24+00:00 From Truncation to Commitment: Persistent Context in Uniform Discrete Diffusion https://megadose.ai/i/da266033-91fd-4d9b-b4a9-13be78f3beed Megadose en 2026-09-01T10:39:25+00:00 ViTAMINS: An Empirical Study of Training Self-Supervised Vision Transformers with Synthetic Hard Negatives https://megadose.ai/i/6749ab01-aacb-4b11-b121-0358bb7cea90 Megadose en 2026-09-01T10:32:48+00:00 The Multiple Timescales of Gradient Descent on the Edge of Stability: A Perturbative Derivation of the Central Flow https://megadose.ai/i/54ee7d3f-d0c9-4efc-9ef0-7f9b920f3ce0 Megadose en 2026-09-01T10:21:10+00:00 PCoMoE: Shifting MoE Inference from Monolithic Expert Selection to Fine-Grained Path Composition https://megadose.ai/i/a24b9d38-dc22-4370-bf1d-4e637384ecf6 Megadose en 2026-09-01T10:07:00+00:00 Phrase-Localized Language-Contrastive Guidance: Training-Free Localized Accent Control for Code-Switching Text-to-Speech https://megadose.ai/i/ffa66266-8069-4b69-a343-c60a4cbc127f Megadose en 2026-09-01T09:53:47+00:00 Figures as Programs: Recursive Generation of Editable Scientific Figures https://megadose.ai/i/f405bec2-ddc6-45ac-a2fe-8327cd357ec5 Megadose en 2026-09-01T09:52:45+00:00 I trained a small transformer in 1.5hrs and it beats many LLMs https://megadose.ai/i/888bdf51-92c1-42ca-89d7-9f6698190b3f Megadose en 2026-09-01T09:52:10+00:00 SinkPruner: Sink-Free Visual Token Pruning for Multimodal Large Language Models https://megadose.ai/i/c6191297-0fb9-4d16-889e-0691658f73df Megadose en 2026-09-01T09:40:01+00:00 Hong Kong-listed Z.ai reports H1 2026 revenue up 5x YoY to ~$142M, with "open platform and API revenue" up 28x to ~$122M, and net loss down 12% YoY to ~$308M (Juro Osawa/The Information) https://megadose.ai/i/f1ec7bfc-7939-4c55-a5f9-87520c11946d Megadose en 2026-09-01T04:36:54+00:00 [AINews] Fal’s H3 Max Live breaks the infinite videogen barrier https://megadose.ai/i/404b4240-3513-4713-ac47-1d0b0f44bba4 Megadose en 2026-09-01T00:00:00+00:00 EM^2Mem: Event-Centric Multimodal Memory for Large Language Models https://megadose.ai/i/d46f6e1b-f5ed-4171-95a7-5030ab8a936e Megadose en 2026-09-01T00:00:00+00:00 ReFlowSET: Representation-Aligned Latent Flow Matching for SAR-to-EO Image Translation https://megadose.ai/i/1239cf26-5540-4c8b-b53a-e441608c763d Megadose en 2026-09-01T00:00:00+00:00 The Safeguard Worked. Is the LLM System Safer? https://megadose.ai/i/bac37b05-6214-477a-a4a9-b1c4e27f7c42 Megadose en 2026-09-01T00:00:00+00:00 DiagEvo: Diagnosis-Guided Self-Evolution via Hierarchical Error Memory https://megadose.ai/i/8aefd2c2-b182-445e-acb9-23e055b5a5ab Megadose en 2026-09-01T00:00:00+00:00 SMELT: Scaling Laws for Compute-Matched MoE Looped Transformers https://megadose.ai/i/8006dcb8-33dd-49c9-9e8b-d7b01c1b5e29 Megadose en 2026-09-01T00:00:00+00:00 Harness-of-Harness: Multi-Day Autonomous Software Development with Continual Improvement https://megadose.ai/i/08ecb857-3b5f-46ba-ad76-86613ca16fe8 Megadose en 2026-09-01T00:00:00+00:00 H3-World: Turning Language Understanding into World Control https://megadose.ai/i/23f4ae3e-9c03-46ec-8588-912cc5ae0689 Megadose en 2026-09-01T00:00:00+00:00 StudentSim: Training LLM-based Student Simulators https://megadose.ai/i/3fa8f1d7-adfc-48f9-b183-d04def3dca6e Megadose en 2026-09-01T00:00:00+00:00 Evaluating Multimodal LLMs as Generalist Vision-Language-Action Agents for Drone Control: Commanding, Approaching, Tracking and Searching https://megadose.ai/i/0c0014d8-516e-4f11-91fa-aeb55e33c00f Megadose en 2026-09-01T00:00:00+00:00 Control-Data Flow Separation: Stable Prompt Optimization in Multi-Agent LLMs https://megadose.ai/i/1c749ea0-4b1e-4af7-8836-01f8057212b1 Megadose en 2026-09-01T00:00:00+00:00 Uncovering Understanding-Generation Synergy in Native Unified Multimodal Models: From Representation, Task to System https://megadose.ai/i/9daddac7-12cd-44a7-882b-9221fffe3079 Megadose en 2026-08-31T23:40:03+00:00 Sources: Anthropic has signed a $35B cloud deal with Nvidia-backed Lambda; Nvidia will hold the lease on and supply chips to a Texas data center built by Hut 8 (Anissa Gardizy/Wall Street Journal) https://megadose.ai/i/0bc734e6-7acd-41be-88b8-07b839336345 Megadose en 2026-08-31T21:25:01+00:00 Sources: AI sales and marketing startup Clay is raising a round led by Wellington at a $7B pre-money valuation, up from $5B via an employee tender in January (Lucinda Shen/Axios) https://megadose.ai/i/a031b988-4e2d-465d-82dc-08a0cde063c5 Megadose en 2026-08-31T21:00:46+00:00 The Pentagon launches ChatGPT Mil and Grok for Government on its GenAI.mil platform, giving its 3M personnel access to AI tools "tailored to warfighter needs" (Kirsten Korosec/TechCrunch) https://megadose.ai/i/61c0f007-051d-4c8d-85b5-bb5a0ee0a239 Megadose en 2026-08-31T17:59:46+00:00 Context-Aware Interleaved Batching for WhisperX https://megadose.ai/i/7241a757-7cf7-4d84-a529-20a079b502e6 Megadose en 2026-08-31T17:59:16+00:00 SUN: Persistent Programs For Language-Grounded Control-to-Learning-to-Real Policies https://megadose.ai/i/b3c223d8-0b3c-4d33-8d80-35243a73662c Megadose en 2026-08-31T17:48:24+00:00 Auditing Anonymous AI Models: A Four-Stage Protocol for Black-Box Identity Verification https://megadose.ai/i/166467f6-9aed-4bf8-9413-e0dbe735c9fe Megadose en 2026-08-31T17:45:01+00:00 Reframe, which uses AI and industrial robot arms to make modular homes in a factory, raised a $40M Series A extension round led by Energy Impact Partners (Rocket Drew/The Information) https://megadose.ai/i/487d8580-d58a-4c72-a0e2-b3458569868a Megadose en 2026-08-31T17:31:18+00:00 PaperGym: Rubric-Centered Evolution for Research-Plan Generation https://megadose.ai/i/db782a39-a7e0-4f84-a70c-673c87c8275d Megadose en 2026-08-31T17:19:40+00:00 TimesFM-3: A zero-shot foundation model for multivariate forecasting https://megadose.ai/i/e36b4f04-c5d6-4a06-bf55-59e9242e5272 Megadose en 2026-08-31T17:14:59+00:00 Aspire: Can Models Self-Evolve from Vague Goals? https://megadose.ai/i/aae9745d-6e58-46ed-843c-286ceafcba45 Megadose en 2026-08-31T17:10:59+00:00 BLOOM-WILT: Logit Tilting for Behaviour Elicitation in Automated LLM Auditing https://megadose.ai/i/78ced18d-0224-49ea-b0cb-b1d17f5910f0 Megadose en 2026-08-31T17:08:41+00:00 LLM Post-Training as Brownfield Maintenance: An Industrial Perspective on Dataware Engineering https://megadose.ai/i/0791a971-63d7-4bb7-876f-73d715f48e0e Megadose en 2026-08-31T17:05:41+00:00 S3Gym: Can LLMs Turn Self-Testing and Self-Judging into Self-Improvement? https://megadose.ai/i/295b24cb-c0e7-464d-9a71-b6cc074f3157 Megadose en 2026-08-31T17:03:45+00:00 One Adapter, Many Tasks: Task-Conditioned Feature Transformations for Continual Learning https://megadose.ai/i/16b39ab7-6a05-412b-808b-ea940afee326 Megadose en 2026-08-31T17:00:09+00:00 Minimax bounds for watermarked and masked recursive discrete distribution estimation https://megadose.ai/i/1a85e9d9-c5a7-47cd-8074-5c667bdd2d86 Megadose en 2026-08-31T16:56:20+00:00 The First Token Is a Clue: Verbalizing Multi-Token Concepts from the J-lens https://megadose.ai/i/26d272ba-7ffc-4741-aa05-dc8c8b129667 Megadose en 2026-08-31T16:53:45+00:00 Token-Efficient Data Reasoning Agents via Adaptive Structuring of Unstructured Data https://megadose.ai/i/2115553c-bb0d-40e8-8646-327443631522 Megadose en 2026-08-31T16:52:57+00:00 Sycophantic Agreement Transfers with Neutral Data via Contrastive Preference Optimization https://megadose.ai/i/027065cb-e171-4f7d-be26-c3bb53d641f1 Megadose en 2026-08-31T16:51:50+00:00 Reconciling Process Supervision with Outcome-Based Credit in Agentic Policy Optimization https://megadose.ai/i/938f4379-5a13-4be8-937f-3266a5407df0 Megadose en 2026-08-31T16:48:51+00:00 Learning to Evaluate Before Improving: Automatic Rubric Induction for Automatic Research Agents https://megadose.ai/i/fefcd69e-1264-43db-80f2-8b04270054c5 Megadose en 2026-08-31T16:45:41+00:00 A Model with No Head and Many Thoughts https://megadose.ai/i/ca8e04cb-8237-4694-b9bd-ccdeb0477cee Megadose en 2026-08-31T16:43:55+00:00 Wrong Prediction, Right Answer: Recovering Evidence from Collapsed LLM Sequence Scores https://megadose.ai/i/3b5d3c95-bc1c-4ee8-aea1-a815efa18c6a Megadose en 2026-08-31T16:42:11+00:00 Universal Transformers for Circuit Computations: Perfect Length Generalization in Tiny Transformers https://megadose.ai/i/ce28857b-c167-4000-8dc4-391baca3428a Megadose en 2026-08-31T16:41:14+00:00 Every Token Leaves a Ripple in the Stream of Thought: Eliciting Model-Internal Token Saliency for Chain-of-Thought Compression https://megadose.ai/i/ed0902a4-7335-49c5-8b0c-82fe8dbb66b7 Megadose en 2026-08-31T16:35:31+00:00 Improving Information Extraction with Learned Queries https://megadose.ai/i/a24145c0-2fe4-4da0-b6e6-05a8b7f768dc Megadose en 2026-08-31T16:34:51+00:00 Measure Before You Manage: Evaluating Agent Working Memory in Coding Agents https://megadose.ai/i/85a463d2-3cd8-4428-88c3-8d31d1264b61 Megadose en 2026-08-31T16:30:01+00:00 a16z says its fifth growth fund has closed additional capital, bringing its total to $8.5B (Andreessen Horowitz) https://megadose.ai/i/3799caa1-30ee-427b-9f0c-8465863adf5b Megadose en 2026-08-31T16:22:10+00:00 Does On-Policy Distillation Really Distill? From Noisy Teacher to Self-Improvement https://megadose.ai/i/6540480f-a755-4343-85a7-b4cafc2e8222 Megadose en 2026-08-31T16:15:36+00:00 Normalized Low-Rank Adaptation https://megadose.ai/i/acceb424-de04-4ca6-845a-9495b3697076 Megadose en 2026-08-31T16:10:04+00:00 Driving on Memory https://megadose.ai/i/6a05265f-726f-41f3-9425-ee1cad4704c5 Megadose en 2026-08-31T16:05:39+00:00 MNIST-PRO: MNIST is Back as a Partially Observable World for AI Agents https://megadose.ai/i/6bc4c85c-03fb-4470-b98a-d4ee5ccd4df5 Megadose en 2026-08-31T16:00:27+00:00 One note in three: a verified census of three deployed AI scribes, and the instrument that counted it https://megadose.ai/i/6138484e-2e0a-4a39-a09a-ba60c7a6137e Megadose en 2026-08-31T16:00:00+00:00 Clipto uses AI to search terabytes of video and is now valued at $250M https://megadose.ai/i/758e644a-c1a9-43bc-baf8-d58432d9ed6f Megadose en 2026-08-31T16:00:00+00:00 GigaPath-Flash and GigaTIME-Flash: Toward population-scale discovery with efficient pathology foundation models https://megadose.ai/i/aee0d9a2-cf05-457f-acf7-5940432285e4 Megadose en 2026-08-31T15:59:54+00:00 LLM Judges Verify Presence, Not Absence: Omission Blindness in AI Clinical Notes and What Recovers It https://megadose.ai/i/2339d6e7-788b-49bc-a792-d000204a5987 Megadose en 2026-08-31T15:15:25+00:00 Nvidia’s $3.5B MediaTek bet reveals its plan for tackling Big Tech’s AI chip buildout https://megadose.ai/i/5a7fe90a-245b-40d7-8b00-341c3206ba7d Megadose en 2026-08-31T15:10:01+00:00 Together AI, which serves open models, announces a deal to use compute from Humain in Saudi Arabia, where it can bypass US backlash over data centers (Niko Gallogly/New York Times) https://megadose.ai/i/f478e18c-950e-41d2-8750-52e719063ab5 Megadose en 2026-08-31T14:05:58+00:00 Sources: CXMT has begun producing HBM3E memory chips in small quantities and plans to expand production in 2027, a major advance for China's chip industry (The Information) https://megadose.ai/i/bde2a58a-35a4-4a66-b2bb-ff2b8d0997c2 Megadose en 2026-08-31T13:12:08+00:00 OpenAI says its ad business has hit $1B in annualized revenue run rate and is expanding globally, as it touts its "diversified business model" ahead of an IPO (Ashley Capoot/CNBC)