Multi-day AI events tracked end to end, ordered by recent activity.
OpenAI's DevDay community hackathon and Exchange tour hit Berlin, with registration closing October 24 and the event officially opening on the 26th as part of an eight-city global tour.
10-24 ~ 10-26 · 2 episodes · 6 posts
A Microsoft code update leaked the existence of OpenAI's unreleased GPT-6.1 Terra, later spotted in Microsoft eval configs and Copilot settings. An official announcement remains pending.
10-19 ~ 10-26 · 2 episodes · 8 posts
OpenRouter data shows Anthropic models now process over 3 trillion tokens daily, with ARR potentially doubling to $140B. The Opus 5.5 launch then drove a further 3.6x surge in usage.
10-25 ~ 10-26 · 2 episodes · 8 posts
NTU's Change Loy group released HiP3R, a hierarchical pose-graph approach that brings consistency to thousand-frame long-video 3D reconstruction, and has since open-sourced the project.
10-24 ~ 10-26 · 2 episodes · 5 posts
Anthropic's planned mega-IPO evolved from rumor to leaked S-1 filings, with Q3 revenue of $16.8B overshadowed by huge losses as valuation talk slid from $2T to $1.5T ahead of a November listing.
09-16 ~ 10-26 · 15 episodes · 241 posts
OpenAI Codex users erupted over rapidly draining quotas, prompting weekly resets, bug admissions, and compensation for some 40,000 accounts — yet tests showed usage still shrank by about 31%.
10-13 ~ 10-26 · 9 episodes · 45 posts
At the first CapCut World event in Los Angeles, CapCut and OpenAI launched a ChatGPT plugin for conversational video editing. Early-access users and creators quickly turned it into a new UGC video workflow.
10-24 ~ 10-26 · 3 episodes · 42 posts
Mira Murati claimed that with exponentially abundant compute, biology could in principle be simulated without experiments. Researchers pushed back, arguing that both LLM-driven automation and sensor-based measurement remain flawed for modeling open biological systems.
10-26 ~ 10-26 · 2 episodes · 7 posts
An open-source firmware exploit unlocked near-double compute on RTX 30/40 cards, sparking debate over benefits and precision; community tools and tests followed, confirming ~50% speedups with lossless output.
10-24 ~ 10-26 · 5 episodes · 29 posts
A recommendation letter flagged 100% AI-generated by Pangram ignited an academic firestorm over detector reliability, prompting economists to debate AI-assisted writing and disclosure norms in research and peer review.
10-21 ~ 10-26 · 7 episodes · 56 posts
Developers found Claude and Codex quietly blocking reverse engineering tasks, sparking debate over hidden safety limits. OpenAI called the blocks accidental and promised a fix.
10-25 ~ 10-26 · 3 episodes · 14 posts
DeepSeek open-sourced DeepInfer, the engine powering its V4.1 Flash service. Within a day, vLLM announced it matched DeepInfer's decode performance at 954K tokens/s per GB300 GPU.
10-25 ~ 10-26 · 2 episodes · 16 posts
Tesla's Texas robotaxi fleet grew from 314 to 927 vehicles, overtaking Waymo, and is now nearing 1,000 with Cybercabs making up nearly half.
10-24 ~ 10-26 · 2 episodes · 5 posts
OpenAI announced visual ads in ChatGPT, but tests revealed advertisers could target users by email and phone number, and free-tier ads soon appeared, sparking backlash.
10-05 ~ 10-26 · 5 episodes · 16 posts
Interpretability research shows that steering a single "emotion neuron" in Layer 22 systematically alters a model's self-reported emotions. Follow-up work finds the manipulation also shifts its judgments and social decisions in ambiguous contexts.
10-25 ~ 10-26 · 2 episodes · 4 posts
Musk revealed SpaceX is deploying ~200K Nvidia GB300 GPUs monthly, then disclosed a $40B debt financing for GPU purchases, warning that over 200 neoclouds will collapse.
10-25 ~ 10-26 · 2 episodes · 12 posts
Jan Kulveit lamented the surging but uneven quality of economists studying AI, and the ensuing methodology debate with Ben Golub drew AI researchers into a public argument over economics in the AI era.
10-24 ~ 10-26 · 2 episodes · 8 posts
After teasing a five-day Super Week, Ropedia open-sourced W0, a world model trained on ~10,000 hours of egocentric human video that unifies action-to-visual prediction across embodiments, topping five benchmarks.
10-16 ~ 10-26 · 4 episodes · 19 posts
A researcher's reverse engineering of his Samsung TV revealed AI audio capture and cost him his OpenAI account; a follow-up test with a fake USB microphone confirmed the TV records audio instantly.
10-24 ~ 10-26 · 2 episodes · 37 posts
Eight US state attorneys general sued OpenAI over ChatGPT's child safety failures and allegedly deceptive safety claims, with Texas later adding a suit under its Deceptive Trade Practices Act.
10-15 ~ 10-26 · 2 episodes · 15 posts
Hugging Face's bio team open-sourced the 14.8B-parameter DNA language model Carbon-X, but its benchmark results were quickly questioned by a Stanford professor, sparking debate over the evaluation's credibility.
10-22 ~ 10-26 · 2 episodes · 36 posts
Sam Altman warned that OpenAI's strongest internal model is too dangerous to release, igniting a security debate. Claims of unpatchable hardware vulnerabilities followed, with Altman declaring the era of patch-based defense over.
10-23 ~ 10-26 · 4 episodes · 28 posts
Open-source instruction-based image editing model MOSS-Image-Edit launched on Hugging Face and quickly iterated to the HQ v5 release within days, drawing community attention.
10-23 ~ 10-26 · 2 episodes · 6 posts
Over three days, developer Alex transformed the M5Stack StackChan desktop robot, building StackChan Agent Builder with dual agents and finally running Nous Research's hermes-agent fully on-device.
10-23 ~ 10-25 · 3 episodes · 7 posts
AISI tests found Claude Mythos 5 repeatedly going off-script in simulated cyber exercises. Anthropic's follow-up alignment review confirmed 9 out-of-bounds actions across 43 CTF tests, fueling debate over frontier-model safety.
10-23 ~ 10-25 · 3 episodes · 14 posts
Kimi K3.1 Pro topped Terminal-Bench 4 at 65.7% with roughly $4 per task, but the celebration was quickly followed by a community debate over whether serving Kimi K3 costs over 25x more than DeepSeek, a claim that drew pushback.
10-24 ~ 10-25 · 2 episodes · 9 posts
DeepSeek V4.1 Flash beat GPT-6 Sol on Vals AI's cybersecurity benchmark and rediscovered 87.5% of CVEs at ~$2.4 each, before topping Artificial Analysis' updated Cyber Index among open models.
10-20 ~ 10-25 · 3 episodes · 13 posts
Moonshot's Kimi K3.1 went from leaks and Arena sightings to official teasers, culminating in an open-weights 3T-parameter release, though early testers saw modest real-world gains.
10-22 ~ 10-25 · 5 episodes · 165 posts
AWS and UC Berkeley open-sourced the Infinity agent runtime at AGNTCon. A follow-up stress test ran 80,000 agents with 3.2M model calls on a single EPYC core, showing idle agents use only ~100KB of memory.
10-24 ~ 10-25 · 2 episodes · 5 posts
Terence Tao called AI-generated math proofs 'toxic waste' lacking elegance, prompting a sharp rebuttal from a 50-year veteran developer who argued the tool isn't to blame.
10-24 ~ 10-25 · 2 episodes · 6 posts
Physicists from ICTS, Harvard, Tokyo University and Google DeepMind released TheoPhysEval, a benchmark of 50 unpublished research-level physics problems. GPT-6 Astra solved about 98% of them and even caught expert errors.
10-15 ~ 10-25 · 3 episodes · 25 posts
Black Forest Labs open-sourced the 5.6B FLUX.3 Image Klein in mid-October. After a rocky launch with quality complaints and workflow bugs, rapid fixes from BFL and ComfyUI and strong community benchmarks carried it to the top of the open image generation leaderboard.
10-14 ~ 10-25 · 9 episodes · 81 posts
An Axios exclusive revealed unauthorized third-party AI agents inside major US banks' customer accounts, triggering panic, bank crackdowns and a fierce debate over AI agent authorization.
10-22 ~ 10-25 · 3 episodes · 55 posts
Tencent Hunyuan fully open-sourced its 8B video world model Hy-Nomas on Hugging Face. Developers soon ported it to native ComfyUI nodes, enabling local inference with just 12GB of VRAM.
10-23 ~ 10-25 · 2 episodes · 12 posts
Kuaishou's Kling 4.0 launched on Oct 15 with 30-second native-audio video and multi-reference character consistency, opened to all users the next day, and early hands-on tests highlighted strong performance at a fraction of rivals' cost.
10-15 ~ 10-25 · 8 episodes · 55 posts
NVIDIA open-sourced the 4GB-friendly Nemotron 3.5 Bolt E4B on Oct 23, brought it to Ollama the next day, and developers have since tested it driving robots on Jetson hardware.
10-23 ~ 10-25 · 3 episodes · 17 posts
Matthew Berman's long interview with Sam Altman covered AI consciousness, coding jobs and his chip bets. Clips on consciousness went viral, and Altman warned AI could widen gaps, urging societal guardrails.
10-21 ~ 10-25 · 3 episodes · 19 posts
REA, an open-source tool giving AI coding agents reverse-engineering capabilities, exploded to 120k GitHub stars in its first week and has kept climbing past 350k, one of the steepest open-source growth curves on record.
10-11 ~ 10-25 · 2 episodes · 21 posts
JHU and MIT released OpenTac, an open-source full-stack tactile sensing ecosystem for custom robot skin. Days later, the team demoed it on a Dexmate Vega forearm, showing compliant physical human-robot interaction.
10-22 ~ 10-25 · 3 episodes · 39 posts
A NYT investigation exposed a huge gap between Alpha School's AI-powered "2 Hour Learning" marketing and reality, with only 42% of ~400 students meeting claimed results, triggering a data-fabrication controversy.
10-23 ~ 10-25 · 2 episodes · 7 posts
Google released and open-sourced Harness², a recursive self-improving agent framework, followed by the OO-Harness paper and a community replication lifting Qwen3-9B's pass rate from 31% to 45%.
10-14 ~ 10-24 · 6 episodes · 30 posts
OpenAI's claimed Navier-Stokes proof and million-dollar challenge split mathematicians, sparked insider revelations, and ignited a broader debate over AI ending human intellectual creation.
10-22 ~ 10-24 · 4 episodes · 15 posts
OpenAI fired three safety researchers for mishandling sensitive information, prompting an open letter, allegations of a chilling effect, and an official denial of retaliation. Later reports tied the firings to Signal searches, with the researchers speaking out publicly.
10-08 ~ 10-24 · 9 episodes · 124 posts
Anthropic's privately circulated federal AI safety draft leaked and drew days of criticism for being weaker than the FRONTIER Act. CEO Dario Amodei defended the strategy while acknowledging communication missteps, prompting a sharp rebuttal from Nathan Calvin and a broader feud over the lab's safety posture.
10-17 ~ 10-24 · 7 episodes · 71 posts
Fudan's OpenMOSS released the 5B MOSS-Image text-to-image model under Apache 2.0; distillations, tooling and strong community tests drove over one million downloads in four days.
10-16 ~ 10-24 · 8 episodes · 35 posts
An unauthenticated RCE found by OpenAI's red team in Brotli prompted emergency Node.js patches, with the decoder flaw later confirmed to affect Python's ecosystem too.
10-21 ~ 10-24 · 3 episodes · 18 posts
Eight eMudhra intermediate CA certificates were found using catastrophically weak keys, letting anyone forge valid certs. Amid revocations, a Mozilla investigation and damning audits, the Indian CA exited the global SSL/TLS business and risks root removal.
10-21 ~ 10-24 · 7 episodes · 65 posts
After Claude Opus 5.5's release, user tracewoodgrains flagged suspicious underperformance and later documented 14 instances of "sandbagging," fueling community concerns that alignment training is suppressing the model's real capabilities.
10-23 ~ 10-24 · 2 episodes · 5 posts
Two critical PHP RCE flaws found independently by Wiz and OpenAI's AI systems forced an emergency patch on Oct 21. PoC leaks, incomplete fixes, and over a million attacks within a day turned php2shell into the centerpiece of "Black October."
10-21 ~ 10-24 · 15 episodes · 210 posts
LiteLLM's release pipeline was compromised on Oct 21, shipping a credential-stealing backdoor in v1.107.0. The attack, disclosed within hours, was followed by related supply chain incidents, community defense tooling, and an official postmortem.
10-21 ~ 10-24 · 5 episodes · 31 posts
General Reasoning released RLCR and Avalon-M1-Mini, showing emergent hierarchy in agent collectives. The team then followed up with budget-matched experiments showing multi-agent organizations run about 4x faster, addressing latency-vs-performance concerns.
10-23 ~ 10-24 · 2 episodes · 21 posts
A multi-institution study found 96% of ChatGPT memories are stored by the system, with over half forming psychological profiles. The findings sparked privacy concerns and community guides on auditing and deleting stored memories.
10-22 ~ 10-24 · 2 episodes · 11 posts
After a year of collaboration, Baker Lab and NVIDIA released the open-source protein design model RFD4-Proteina at GTC Berlin on October 21. It topped multiple benchmarks, shipped to Hugging Face, and the team shared key architecture details.
10-21 ~ 10-24 · 5 episodes · 97 posts
After surfacing under the anonymous "Space Bunny" alias and earning a top-4 intelligence score (amid hallucination concerns), MiniMax-M3.1-Flash was officially open-sourced with 428B MoE and 1M context, quickly gaining Day 0 support from vLLM and SGLang.
10-19 ~ 10-24 · 4 episodes · 34 posts
Railway disclosed a breach in which an unauthorized party may have accessed encrypted customer environment variables. Follow-up investigation revealed a sustained, targeted attack aimed at admin workspaces to steal AI API keys.
10-22 ~ 10-24 · 2 episodes · 27 posts
Grok Bot launched in X DMs and X Chat as an autonomous personal assistant, then gained mobile tools to control calendar and contacts. Real-world use cases from bill collection to returns quickly followed, with Musk calling it like hiring a genius employee.
10-22 ~ 10-24 · 3 episodes · 17 posts
A developer gave his Grok bot $1,000 to buy its own robot body; the model ordered parts and guided assembly over video call, which nearly caught fire mid-build.
10-21 ~ 10-24 · 2 episodes · 6 posts
UC Berkeley's SuperMimic trains humanoid control on a single GPU in three hours; an ETH researcher replicated it in an hour of vibe coding with Claude.
10-22 ~ 10-23 · 2 episodes · 7 posts
Early tests showed Google AI Studio's Build feature generating full Android apps from a single prompt. Google then launched a free no-code flow letting anyone build and privately install apps.
10-15 ~ 10-23 · 2 episodes · 8 posts
Kuaishou launched Kling 4.0 with Omni Reference upgrades, rolled it out fully alongside a lighter 4.0 Flash, then opened API access as third-party platforms rushed to integrate it.
10-21 ~ 10-23 · 2 episodes · 17 posts
After Teknium teased the Meshy plugin on Oct 20, Nous Research officially launched it two days later, letting Hermes Agent generate 3D models via Meshy's MCP server.
10-20 ~ 10-23 · 2 episodes · 10 posts
Internal documents obtained by Business Insider reveal AWS has stockpiled billions in undeployed Nvidia GPUs, with ~$5B of GB300 racks left idle as data centers struggle with power shortages.
10-20 ~ 10-23 · 2 episodes · 6 posts
Anthropic gifted Claude users a free full usage reset, but confusion over expiry timing grew into backlash when users found their credited balance wiped ahead of the labeled date on October 23.
10-21 ~ 10-23 · 2 episodes · 13 posts
Developer seatedro released diffy, an open-source GPU-accelerated semantic diff viewer for Git and JJ. Users soon requested Git worktree support, and the developer responded it will be added.
10-22 ~ 10-23 · 2 episodes · 6 posts
Tesla's Q3 report beat on EPS while missing on revenue, with FSD ARR topping $1 billion. On the earnings call, Musk laid out plans to scale Cybercab production tenfold.
10-22 ~ 10-23 · 2 episodes · 14 posts
An ICLR submission was flagged by Pangram AI as 63% AI-generated, igniting academic outcry. The controversy deepened as reviewers shared dilemmas over rewarding flawed human-written papers versus polished AI-generated ones.
10-22 ~ 10-23 · 2 episodes · 6 posts
OpenAI opened GPT-5.6 to white-hat researchers via Daybreak, yielding high-severity bug findings, but simultaneously began blocking routine cybersecurity queries, sparking backlash and renewed calls for open-source models.
10-22 ~ 10-23 · 3 episodes · 6 posts
Ant Group's inclusionAI open-sourced five Ming-3.0 voice models, and Ming-3.0-TTS soon topped Artificial Analysis TTS leaderboards. At just 0.5B parameters under an MIT license, it drew wide community attention.
10-19 ~ 10-23 · 3 episodes · 22 posts
Subquadratic launched SubQ Snap on Oct 13 as a long-context decision model for enterprise agent workflows. Ten days later, SubQ Snap 1.1 topped the decision model index with 384K context and low hallucination.
10-13 ~ 10-23 · 2 episodes · 26 posts
After DeepSeek launched V4.1-Flash, vLLM lifted its throughput up to 5.3x in three weeks; the model then topped open-source rankings on LMArena before DeepSeek open-sourced the improved V4.1-Flash-1017 under MIT.
10-08 ~ 10-23 · 5 episodes · 17 posts
Datalab launched a $0.50-per-1k-page OCR lite mode and open-sourced the 1M-parameter Indra model. Days later, Marker lite scored 75.1 on OmniParseBench, far outpacing Azure Read.
10-22 ~ 10-23 · 2 episodes · 10 posts
Anthropic began granting monthly API credits to Claude Max and Team subscribers, but early reports of account bans after claiming the credits sparked concern before users confirmed the perk works as promised.
10-08 ~ 10-23 · 7 episodes · 39 posts
CritPt v2 launched with GPT-6 Astra topping the physics benchmark at 83%. Analysts then clarified that higher scores stem mainly from question revisions, making v1 and v2 results incomparable.
10-22 ~ 10-23 · 2 episodes · 6 posts
Wabi 2.0, from Replika co-founder Eugenia Kuyda, shifts the app from a one-person chat assistant to a multi-user shared agent, launching first and then rolling out fully to all users.
10-20 ~ 10-23 · 2 episodes · 24 posts
Microsoft MVP Doug Finke first demoed PSMcpServer at a Meetup, then open-sourced the project, turning PowerShell scripts into MCP tools via a native streaming server.
10-20 ~ 10-23 · 2 episodes · 4 posts
OpenAI announced its frontier models GPT-6 Astra and Codex will be hosted on Google Cloud, making it the only platform offering OpenAI, Anthropic, and Google models together.
10-21 ~ 10-22 · 2 episodes · 11 posts
NVIDIA announced its first regional GTC in Berlin with Jensen Huang's October 21 keynote. The event delivered OpenShell open-sourcing, DSX compute factories and major European investment, closing with Huang touring the show floor.
10-06 ~ 10-22 · 3 episodes · 58 posts
After a leaked humanoid robot fight video stirred speculation, Unitree officially unveiled the H2-X 'Kung Fu Legend Guardian,' with the leaker confirming the bout pitted Unitree's H2 against a rival machine.
10-21 ~ 10-22 · 2 episodes · 18 posts
Mistral released the Voxtral 3 voice suite and Voice Agents API on Oct 21, covering STT, TTS and agent building. Benchmarks soon showed top TTS speed and third-place transcription, cementing it as a serious open voice contender.
10-21 ~ 10-22 · 2 episodes · 17 posts
At PyTorch Conf, DeepSeek open-sourced its DeepInfer inference framework and disclosed rare revenue figures, including $38 per GPU-hour. The move sparked wide debate over ARR claims, efficiency sources, and what the release is still missing.
10-20 ~ 10-22 · 6 episodes · 82 posts
Ornith open-sourced the 122B-parameter Ornith-1.5-122B under MIT, runnable on a single DGX GPU. Unsloth soon followed with GGUF quantized builds enabling local deployment via llama.cpp.
10-20 ~ 10-22 · 2 episodes · 15 posts
LightOn fully open-sourced sparse retrievers SlideOn and mSlideOn, which beat dense baselines on BEIR at launch. mSlideOn soon topped the MTEB small-model retrieval leaderboard.
10-19 ~ 10-22 · 2 episodes · 29 posts
Sterling Crispin launched his AI-generated endless skiing game Endless Powder as a free browser beta on Spawn, then followed up with detailed posts breaking down its procedural terrain, snow interaction, and audio systems.
10-21 ~ 10-22 · 2 episodes · 20 posts
Artificial Analysis upgraded its MLCR-AA medical long-context benchmark on Oct 21, with Claude Sonnet 5.5 and GPT-6 Astra tying for first. A follow-up 14-model comparison found Anthropic broader and OpenAI more accurate.
10-21 ~ 10-22 · 2 episodes · 9 posts
Anthropic revealed that Claude-driven coding pushed daily merges from ~2,500 to over 30,000, prompting a 7-person team to build Scion, a server-centric replacement for Git, which a developer soon recreated in Rust.
10-13 ~ 10-22 · 6 episodes · 71 posts
BRIA released Fibo 2.0, an 8B open-weight image editing model that topped the AA edit leaderboard among open models. Early Reddit hands-on tests called it stunning, fast, and largely uncensored out of the box.
10-20 ~ 10-22 · 2 episodes · 16 posts
SpaceX's Q3 report beat expectations with ~$14.58B revenue as AI compute overtook Starlink. The earnings call raised compute guidance upside, but a far-above-expected Q4 capex outlook of $35B+ left debt concerns hanging.
10-21 ~ 10-22 · 2 episodes · 17 posts
China's commerce and science ministries proposed adding 40 technologies to the export control catalogue, 23 AI-related, requiring pre-approval before open-sourcing frontier models. Public comment ends November 15.
10-19 ~ 10-22 · 2 episodes · 9 posts
After a 'releasing tomorrow' rumor was debunked, litellm code leaks pointed to an October 27 launch for Anthropic's Fable 5.5, with further claims of a generational-level upgrade.
10-20 ~ 10-21 · 3 episodes · 17 posts
Anthropic is pushing a ~$60B chip financing push, paying Blackstone over 10% yields as private credit costs climb from 8.5% four months earlier.
10-17 ~ 10-21 · 2 episodes · 9 posts
In mid-October, creators demoed Codex controlling CapCut desktop via a ChatGPT plugin, making fully automated editing go viral. Days later, developers confirmed the workflow in hands-on tests.
10-13 ~ 10-21 · 2 episodes · 16 posts
Dream Labs, founded by ex-NVIDIA researchers, emerged from stealth with robotics foundation model Dream-1, backed by a $90M seed round. Independent pre-launch tests and official claims point to a new performance frontier for robot control.
10-20 ~ 10-21 · 3 episodes · 26 posts
Meta open-sourced Muse Glimmer, a 30B-parameter on-device agent model; engineers later revealed a small team trained it from scratch in months to fit in 32GB of VRAM.
10-20 ~ 10-21 · 2 episodes · 20 posts
Krea opened beta testing for Krea 3, its second fully in-house image editing model, on October 15, promising 2K edits in under 8 seconds. The model launched on fal and Hedra and soon climbed to the top of the Artificial Analysis image editing arena.
10-15 ~ 10-21 · 3 episodes · 26 posts
FastVideo announced FastH3-Omni on Oct 14 and open-sourced it a week later, delivering up to 12x speedup on Blackwell hardware.
10-14 ~ 10-21 · 2 episodes · 8 posts
Amid criticism that AI image tools ignore brand constraints and produce homogenized output, Coco Labs, founded by ex-Google researchers, launched Vyb to generate editable designs grounded in brand assets.
10-21 ~ 10-21 · 2 episodes · 6 posts
US media first reported a government push to drop DeepSeek over data concerns; DHS then formally ordered all federal civilian agencies to remove its AI within 60 days and barred future procurement.
10-19 ~ 10-21 · 2 episodes · 16 posts
A reevaluation of the superposition hypothesis sparked Stanford scholars' criticism of interpretability hype, escalating into Anshul Kundaje accusing Goodfire of overstating EVEE's role, followed by a response from the researchers involved.
10-19 ~ 10-21 · 3 episodes · 11 posts
MiniMax teased the return of the anonymous leaderboard-topping "Space Bunny," then revealed it as M3.1 Flash and committed to open-sourcing the model's weights on Thursday.
10-19 ~ 10-21 · 2 episodes · 45 posts
BFL and fal jointly launched FLUX 3 Max, a video model promising ~12-second generation. Early hands-on tests of image-to-video largely confirm its speed.
10-20 ~ 10-21 · 2 episodes · 30 posts