Tech News

10

Elon Musk says xAI's 1.5-trillion-parameter Grok 4.5 has entered private beta testing at Tesla and SpaceXEarly evaluations indicate performance approaches or exceeds Opus.

9h#1VIEWS4.2M#1LIKED31.1k#1BOOKMARKED1.9k#1COMMENTS5.5k

#52

Top post: @kimmonismus “I hope Elon is referring to opus 4.8. would be a welcome release if grok 4.5 outperforms opus 4.8 on key benchmarks”

20

30

Austria lobbies the EU to host Anthropic on European infrastructure to bypass US regulatory limitsSkeptics question if EuroHPC supercomputers can run Claude Opus.

5h133.3k1.1k134

#353

Top post: @AndrewCurran_ “From the letter: 'Let us jointly explore the strategic establishment and participation of Anthropic within the European Union. With legal certainty, market access, capital and a set of values that suits this company.' 'Anthropic fits us particularly well. A company that understands the ethical use of AI not as marketing, but as a core conviction. That places safety over speed. That is a deeply European attitude. This company would not be constrained in Europe; it would be unleashed.' The UK also made similar overtures a few months ago. It won't happen. Anthropic will stay where the compute is, and where the supply is guaranteed. They can't risk getting cut off, and from here on out the compute will increasingly be concentrated within US borders.”

Watchlist

Video Signals

#1360

Anthropic Overtakes OpenAI in US Corporate AI Paid Adoption

7h ago137K908250

#398

Coinbase Positions As Financial Account For Intelligence Age

2h ago61K75675

#1861

Critics Accuse Anthropic CEO Dario Amodei Of Urging AI Model Bans

5h ago29K18629

#1360

Anthropic CEO Calls Open Source A Distraction For AI Transparency

1h ago284K27640

#1257

Austria Urges EU To Host Anthropic Amid US AI Export Curbs

1h ago3.8K113

#543

OpenAI Codex App Reaches 5M Weekly Users With Focus on Taste

34m ago4K1811

#577

LLM Generates Working Terminal Window Manager in Python in One Minute

2h ago1.8K255

#225

Swyx Teams Up With NVIDIA To Make Local AI The Default

55m ago16K00

#1029

Writer CEO May Habib Urges Survival Mantra Amid AI Lab Rivalry

1h ago1.5K41

Video Shows Shenzhen Restaurant Robots Claiming Global Hub Status

12h ago4.7K2514

40

50

PCAST Co-Chair David Sacks argues US regulations failed to prevent China from matching Anthropic in cybersecurityInvestor Sheel Mohnot questioned the origin of current export controls.

4h207k2.4k228

#192

Top post: @pitdesi “@DavidSacks So who do we blame for the current export controls and centralized permissioning? Is there going to be a new AI Czar?”

New github stars (48hrs)

OSWorld 2.0 benchmark and environment for long-horizon computer use agent evaluation

xlang-ai/OSWorld-V2713h

Benchmarks computer-use agents on long-horizon real-world desktop tasks via VM environments and gated task sets.

1014

397

843

Full-stack Python codebase for training and evaluating draft models used in speculative decoding

deepseek-ai/DeepSpec1.5k1d

Provides full-stack code for training/evaluating speculative decoding draft models with data prep and eval scripts.

484

1194

1136

LLM post-training framework connecting Megatron training to SGLang rollout via a shared data buffer

THUDM/slime7k1d

Integrates Megatron training with SGLang rollouts for LLM post-training and RL scaling.

458

1014

484

60

Resurfaced 2015 emails published by TechEmails show OpenAI's founders brainstorming and debating the organization's nameThe correspondence details branding ideas prior to the official launch.

3h40.6k36038

#113

Top post: @rohanpaul_ai “Interesting thread. on naming OpenAI.”

73

Khosla Ventures' Keith Rabois endorses Jaya Gupta's guide to building iconic companies in the AI eraExecutive Josh Elman also agreed, calling the guide timeless.

6h315.4k6791.7k

#384

Top post: @joshelman “The advice here is timeless. And a great reminder that in this exciting moment where everything seems different, core principles still hold true”

80

AVERI's Miles Brundage offers a 100-to-1 bet disputing WSJ's claim that Zhipu AI's GLM-5.2 matches Anthropic's Mythos in cybersecurityThe wager hinges on UK AISI cyber range tests.

12h17.1k19915

#57

Top post: @Miles_Brundage “Cosigned, and it's also ironic given WSJ's China hawk demeanor that the only source here on the claim in the headline is a Chinese company”

92

OpenAI's Andrew Ambrosino says the Codex desktop app has grown sixfold since February to 5 million weekly active usersNearly all OpenAI employees, including non-engineers, use it daily.

1h59.5k194131

#543

Top post: @jxnlco “This is the hot codex guy?!”

104

Wharton's Ethan Mollick argues AI model routers systematically underestimate the difficulty of qualitative and non-coding tasksThis routing mistake deprives creative tasks of advanced models

3h29.7k43658

#184

Top post: @emollick “It is worth being very, very careful about how you are approaching routing, especially when the systems are primarily tested on verifiable IT benchmarks, which may lead you to overestimate the ability of weaker models.”

112

Investor Vijay Pande argues distributed AI inference will succeed on consumer edge devices rather than decentralized computing networksApple M-series chips could run local workloads for free.

7h294.1k1k396

#347

Top post: @vijaypande “I bet this distributed inference will look pretty simple — everything on the edge. That’s Apple’s play and the M silicon does great. Combine with OS models and you’re only six months out of date (which is just fine for a ton of people) at zero marginal cost”

123

Ramp transaction data shows Anthropic overtook OpenAI in its share of paying US business customers by late 2025Holiday coding side projects drove the corporate adoption shift.

7h138.8k921251

#1360

Top post: @NickADobos “Holiday vacation winter 2025 was a huge inflection point for Claude code & Anthropic The perfect timing window of having the best model + harness at the time, when corporate decision makers had time to play on side projects, won Claude trillions $$$ in future enterprise tokens sales. I expect model labs to push huge releases in December from now on, the same way retailers & video game launches try to get things out around that time for the Christmas cycle.”

133

Chinese "transfer stations" bypass Anthropic's KYC to resell Claude API access at 5% to 10% of retail pricesProxy operators harvest user prompts and swap models.

6h9.3k9253

#113

Top post: @rohanpaul_ai “A crazy blog. Chinese developers are buying Claude access through gray-market API transfer stations that can sell tokens at 5% to 10% of official prices while hiding the real user from Anthropic. A transfer station is a middle server that takes a user’s prompt, sends it to Claude through overseas accounts, returns the answer, and collects payment through WeChat or Alipay. The transfer station collects many Claude accounts through free credits, discounted accounts, shared subscriptions, overseas payment workarounds, fake verification, or sometimes stolen-card accounts. It connects all those accounts behind one proxy, so Chinese users do not talk to Anthropic directly and only pay the proxy in RMB. The cheap price comes from account farming, free-credit abuse, resale of unused quota, subscription splitting, possible stolen cards, and a darker trade where user prompts and outputs become training data. So the price hugely cheap not because Anthropic is giving a discount; it is cheap because the transfer station lowers its own cost and creates extra hidden revenue. The user thinks they are buying cheap inference, but the proxy may swap Opus for weaker models, inflate token use, or store private code, tool calls, reasoning traces, and business data. The proxy may store user prompts, code, outputs, and tool traces, then sell or reuse that data for model training. This breaks a core assumption behind KYC, account bans, and abuse monitoring: the AI company sees the proxy, not the real person, so banning one account leaves the upstream supply chain alive.”

144

Hugging Face CEO Clément Delangue says regulatory scrutiny of risky AI models has become an effective enterprise marketing tacticNathan Lambert blamed the trend on subjective, vibe-based regulation.

5h43.3k43127

#80

Top post: @natolambert “This is real and a horrible consequence of vibe regulation of frontier models.”

154

Marathon Management Partners' Gokul Rajaram backs Coach to replace traditional PRDs with structured, four-part product specsThe new spec format is optimized for AI agents.

2h31.4k265502

#974

Top post: @gokulr “Coach: http://www.usecoach.ai”

169

Resurfaced claims that Dario Amodei called GPT-2 too dangerous to release spark debate over OpenAI's 2019 strategyCritics argue sensationalized media distorted OpenAI's original 2019 announcement.

11h14.8k11011

#861

Top post: @scaling01 “Dario Amodei actually never made such statements what you are showing is the framing of the press, which is of course clickbaity af you should read what OpenAI actually said: https://openai.com/index/better-language-models/”

178

Nathan Lambert reports backlash for criticizing AI regulatory capture, choosing independence over lucrative big-tech rolesAVERI Executive Director Miles Brundage backed Lambert's independent stance

1h6.6k641

#57

Top comment: @natolambert “I have a lot of discussions with people who literally want open models banned. Happened in 2023. Happened in 2024. Happens today.”

185

Google delists Pragmatic Engineer creator Gergely Orosz's investigation into Pollen's collapse following a fraudulent DMCA takedownThe fraudulent filer claimed residency on uninhabited Bouvet Islands.

2h23.7k49556

FASTEST CLIMBING+5

#1155

Top post: @xeophon “@GergelyOrosz Someone about to find out what a Streisand effect is”

192

Nathan Lambert, AI2 post-training lead, tracks 30 open models and argues that banning open-source AI is technically futileHe warns restrictions concentrate power and undermine safety.

2h8.3k8417

#80

Top post: @xeophon “Another artifact with many new (and recurring!) friends from @NVIDIAAI, @poolsideai, @cohere, @jetbrains and ofc @Zai_org! An interesting one, given the recent developments. We argue that banning open models is not only unsafe, but ultimately anti-freedom.”

204

Independent ML consultant Hamel Husain says Codex's cross-device remote sessions are superior to ClaudeThe system synchronizes local and remote projects via mobile.

3h30.7k22639

#766

Top post: @HamelHusain “@jxnlco Goodwill Hunting is goated for memes”

Today's Highlights

DeepSeek and Peking University release DSpark, a speculative decoding framework boosting DeepSeek-V4 speed by up to 85%

Elon Musk says xAI's 1.5-trillion-parameter Grok 4.5 has entered private beta testing at Tesla and SpaceX

Weijia Shi, University of Washington NLP researcher, releases PhD defense slides on modular architectures to break monolithic language models

Google delists Pragmatic Engineer creator Gergely Orosz's investigation into Pollen's collapse following a fraudulent DMCA takedown

Top Stories

Elon Musk says xAI's 1.5-trillion-parameter Grok 4.5 has entered private beta testing at Tesla and SpaceXEarly evaluations indicate performance approaches or exceeds Opus.

Weijia Shi, University of Washington NLP researcher, releases PhD defense slides on modular architectures to break monolithic language modelsLuke Zettlemoyer and Noah Smith advised her doctoral thesis.

Austria lobbies the EU to host Anthropic on European infrastructure to bypass US regulatory limitsSkeptics question if EuroHPC supercomputers can run Claude Opus.

Video Signals

Steven Sinofsky and Guillaume Verdon argue breaking up Apple would raise consumer costs, disputing Ocasio-Cortez's antitrust proposalThe proposal followed reports of anticipated hardware price increases.

PCAST Co-Chair David Sacks argues US regulations failed to prevent China from matching Anthropic in cybersecurityInvestor Sheel Mohnot questioned the origin of current export controls.

Resurfaced 2015 emails published by TechEmails show OpenAI's founders brainstorming and debating the organization's nameThe correspondence details branding ideas prior to the official launch.

Khosla Ventures' Keith Rabois endorses Jaya Gupta's guide to building iconic companies in the AI eraExecutive Josh Elman also agreed, calling the guide timeless.

AVERI's Miles Brundage offers a 100-to-1 bet disputing WSJ's claim that Zhipu AI's GLM-5.2 matches Anthropic's Mythos in cybersecurityThe wager hinges on UK AISI cyber range tests.

OpenAI's Andrew Ambrosino says the Codex desktop app has grown sixfold since February to 5 million weekly active usersNearly all OpenAI employees, including non-engineers, use it daily.

Wharton's Ethan Mollick argues AI model routers systematically underestimate the difficulty of qualitative and non-coding tasksThis routing mistake deprives creative tasks of advanced models

Investor Vijay Pande argues distributed AI inference will succeed on consumer edge devices rather than decentralized computing networksApple M-series chips could run local workloads for free.

Ramp transaction data shows Anthropic overtook OpenAI in its share of paying US business customers by late 2025Holiday coding side projects drove the corporate adoption shift.

Chinese "transfer stations" bypass Anthropic's KYC to resell Claude API access at 5% to 10% of retail pricesProxy operators harvest user prompts and swap models.

Hugging Face CEO Clément Delangue says regulatory scrutiny of risky AI models has become an effective enterprise marketing tacticNathan Lambert blamed the trend on subjective, vibe-based regulation.

Marathon Management Partners' Gokul Rajaram backs Coach to replace traditional PRDs with structured, four-part product specsThe new spec format is optimized for AI agents.

Resurfaced claims that Dario Amodei called GPT-2 too dangerous to release spark debate over OpenAI's 2019 strategyCritics argue sensationalized media distorted OpenAI's original 2019 announcement.

Nathan Lambert reports backlash for criticizing AI regulatory capture, choosing independence over lucrative big-tech rolesAVERI Executive Director Miles Brundage backed Lambert's independent stance

Google delists Pragmatic Engineer creator Gergely Orosz's investigation into Pollen's collapse following a fraudulent DMCA takedownThe fraudulent filer claimed residency on uninhabited Bouvet Islands.

Nathan Lambert, AI2 post-training lead, tracks 30 open models and argues that banning open-source AI is technically futileHe warns restrictions concentrate power and undermine safety.

Independent ML consultant Hamel Husain says Codex's cross-device remote sessions are superior to ClaudeThe system synchronizes local and remote projects via mobile.

Recent Stars

Github Stars

Yesterday's Top Stories, Jun 27, 2026.

DeepSeek and Peking University release DSpark, a speculative decoding framework boosting DeepSeek-V4 speed by up to 85%They also open-sourced DeepSpec, an MIT-licensed draft-model codebase

Coinbase cuts AI spending by nearly 50% using dynamic routing, open-weight models, and prompt cachingPrompt cache hit rates increased from 5% to 60%.

US Commerce Department partially lifts export restrictions on Anthropic's Mythos 5 model for 100 authorized institutionsAnthropic's Fable 5 model remains fully banned.

Systems architect Daniel Jeffries warns that sudden restrictions on frontier AI models could spark an open-source rebellionHigh-level model access was reportedly blocked for two weeks.

Anthropic CEO Dario Amodei argues open-source AI is a distraction because model weights do not grant true transparencyCritics argue open weights allow local deployment and fine-tuning

Trump administration reportedly close to lifting restrictions on Anthropic's Fable 5 after 15-day security outageThe Pentagon and NSA must still grant final clearance.

Eric Topol says GPT-5.5 Pro scored 79/100 on medical benchmark but remains below clinical deployment thresholdSeparately, Sophont released Medmarks v1.0 to automate medical evaluations.

Suhail Doshi, Playground AI founder, details his AI workflow for digesting queued research papers and articles to avoid doomscrollingA custom prompt translates academic papers into simplified summaries.

UnconvAI releases Un-0, an image generation model built on a backbone of coupled oscillatorsThe architecture leverages the Winfree Oscillatory Neural Network framework

Pseudonymous AI researcher Janus argues that halting AI development increases existential risk and invites poor government oversightAndrew Curran agreed, arguing that slowing down is counterproductive.

Zhipu AI model reportedly matches Claude Mythos in cybersecurity vulnerability detection, sparking debate over dual-use risksCritics question if the benchmark evaluated Zhipu's GLM 5.2-Cyber.

Policy analyst Dean W. Ball argues the US has shifted to a restrictive, de facto AI licensing regimeTrump reportedly asked OpenAI to stagger its upcoming model

Founder Beff argues US restrictions on leading AI models backfired due to flawed comparisons to nuclear weaponsT3 Stack creator Theo Browne initiated the policy debate.

OpenAI updates its Codex application with faster thread switching and position-preserved scrollingThe updates add deeper local history and archive management.

Study finds compressing Chain-of-Thought traces during knowledge distillation degrades downstream student model accuracyEvaluations used teacher-student pairs like Llama-70B and Ministral-14B.

Olivia Moore, a16z partner, argues consumer LLM apps lack collaborative features beyond static link sharingTechnical and design hurdles currently prevent multi-user chat environments.

BINEVAL framework evaluates LLMs using atomic binary questions to outperform G-Eval and UniEvalHamel Husain says binary metrics simplify LLM error analysis

Independent researcher Douglas Yao synthesizes novel Alzheimer's compound PAC-832 in his garage using Claude Code and OpenTronsThe candidate is the first selective GalR1 antagonist.

Perplexity's Aravind Srinivas argues enterprises will run in-house development loops to optimize token value per wattProprietary domain knowledge prevents replication by generalized external models.

AI2 post-training lead Nathan Lambert accuses Anthropic of using political pressure on model distillation for regulatory captureLambert claims safety focus blinds Anthropic's staff to competition.

DeepSeek recruits deep learning researchers in Beijing to optimize pre-training on Ascend NPUs and NVIDIA GPUsThe roles focus on hardware-software co-design to pursue AGI.

AI developer Teortaxes questions the validity of the prinzbench AI benchmark after GLM-5.2 scores a low 30 out of 99The benchmark placed Grok 4.20 ahead of Opus 4.8.

Yaroslav Bulatov says physical data movement, not computation, drives GPU energy consumption because wires cannot shrink as fast as transistorsTraditional computing models fail to account for physical data location

Andrew Curran predicts Fable 5 and GPT-5.6 will receive regulatory release approval next weekTeknium noted that compliance with these guidelines is voluntary.

Daniel Kokotajlo, AI Futures Project leader, shares a forecasting workflow urging practitioners to confidently extrapolate trends despite social biasesHe recommends using quantitative tools like the Bio Anchors framework

Policy scholar Dean W. Ball argues that establishing AI development as First Amendment-protected expression is the next critical legal challengeSeth Lazar warns this constitutional framing is philosophically problematic.

Midjourney founder David Holz shares a speculative future timeline predicting 10 billion humanoid robots and the dissolution of MercuryThe timeline also projects 10 billion brain-computer interfaces

Y Combinator co-founder Paul Graham says a tool to bypass the UK's proposed social media ban would see rapid startup growthMichael Nielsen likened the ban to restricting bedtime reading

Reid Hannaford, filmmaker, demonstrates an AI workflow using Midjourney, Blender, and Seedance for precise 3D camera choreographySimple 3D shapes replace unpredictable text prompts for scene blocking

Sam Altman teases a frontier AI model reaching 750 tokens per second with a planned July releaseThe performance update will run on 5.6 sol hardware