Tech News

10

Former Meta PM Xiaoyin Qu predicts US and European enterprises will adopt Chinese AI models to self-host on private GPUsQu cites enterprise distrust of Anthropic's data retention policies

7h#1VIEWS191.6k#1LIKED1.6k#1BOOKMARKED628

#830

Top post: @zephyr_z9 “"American and European enterprises will ditch OpenAI and anthropic and adopt Chinese models." I love fireworks, Nebius et al But they don't have the scale or compute required to support this”

21

Kim Isenberg and Peter Wang argue the US is losing the AI race due to energy and data center constraintsChina uses free models to promote domestic Huawei hardware.

4h104.9k872409

#121

Top post: @garrytan “Build power and datacenters”

31

Startup 8090 raises $135 million in Series A funding led by Salesforce VenturesThe capital will fund hiring and infrastructure development

1h173.4k1.1k380

#347

Top post: @8090_Factory “We raised our Series A: $135M, led by Salesforce Ventures. We looked back at how we got here, because building this has been hard and worth it, and this is only step one.”

Watchlist

Video Signals

#347

8090 Raises $135M Series A Led by Salesforce Ventures

2h ago164K1.1K380

#1360

Chart Shows AI Uses Far Less Water Than Beef or Sugar

3h ago43K597303

#1360

Anthropic CEO Warns Open Source AI Moves Down Dangerous Path

7h ago236K18033

#684

DAIR AI Academy Launches Hands-On Lab on LLM-as-a-Judge

36m ago3.8K3144

#855

One Model Controls Multiple Robots in UMI-Style Mobile Manipulation Demo

1h ago2.2K279

#1853

Omni AI Video Tool Renders Text Accurately In Prompt Demo

1h ago4K3316

#1833

VC Praises Three-Minute AI-Generated Ad Featuring Suited Dogs

32m ago1.1K63

#403

DIY Builder Shares Sim2Real Quadruped And Biped Robot Walking Tests

2h ago35900

1,372 Humanoid Robots Set Guinness Record With Synchronized Dance

7h ago1.7K102

Google Builds Always-On AI Agent That Acts With Phone Locked

6h ago1.7K81

41

DeepSeek's upcoming V4 models will introduce peak-hour pricing that doubles API rates during peak Beijing hoursThe models support a 1M token context window.

4h103.4k86293

#501

Top post: @scaling01 “get ready for cheap Opus level intelligence”

53

Independent AI researcher Richard Ngo argues David Ricardo's 1817 framework shows capital owners remain vulnerable to expropriation under AGIThe critique challenges whether AGI lab jobs offer protection.

4h1.8k72

#121

Top comment: @alexolegimas “Link to the whole thing: https://www.marxists.org/reference/subject/economics/ricardo/tax/ch31.htm”

New github stars (48hrs)

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

xlang-ai/OSWorld-V27122h

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

1014

397

843

61

Chinese DRAM manufacturer CXMT signs $3 billion supply deal with Tencent amid talks with Alibaba, ByteDance, and XiaomiCXMT's DDR5 chip yields currently lag behind global competitors.

7h75.7k60069

#501

Top post: @zephyr_z9 “But a memory tourist told me CXMT will flood the market with cheap DRAM 😱😱😱”

72

Nomad List founder Pieter Levels criticizes Windows 11 for full-screen, multi-step Microsoft 365 upsell ads on a paid licenseEpic Games CEO Tim Sweeney joined the design criticism.

3h72.8k1.1k47#1COMMENTS182

RISING LIKES1.1k

#627

Top post: @levelsio “Okay so the solution is get LTSC version of Windows which is for enterprise and doesn't do any of this!”

82

Anthropic CEO Dario Amodei argues open-source AI cannot be monitored, revoked, or patched for safety after releaseThe clip reignited debate over open-source AI safety risks

7h45.3k53161

#403

Top post: @yacineMTB “For every software provider, your biggest competitor is the customer themselves -- building it for themselves and thereby reducing reliance on you. Instead consider making a net positive business that helps your customer, instead of trying to rent seek forever”

92

e/acc founder Beff argues employing genius physicists to work on incremental LLM mutations is a tragedyTeortaxes sparked the debate, noting many physicists join Anthropic.

6h22.6k26252

#501

Top post: @beffjezos “At Extropic, we have a whole team of physicists trying to understand the fundamentals of intelligence from first principles of thermodynamics and imbuing these principles in novel hardware. We are unapologetically anti-incrementalist and dissatisfied with "the current thing".”

102

Teortaxes argues export controls on Chinese memory manufacturers have reversed overcapacity concerns, leaving them unable to meet global demandThe supply gap fuels intense competition for critical fabrication tools

4h66.1k14877

#501

Top post: @zephyr_z9 “Everyone is fighting for the same DUVi, etch/deposition tools WFE sub systems are going to rip hard in the next 2 years”

112

X's Premium+ subscription provides up to $286 in retail API tokens for under $15 in backend compute costsThe package includes developer access to Composer 2.5 Fast.

5h43.7k26032

#501

Top post: @zephyr_z9 “"that's roughly $268-$286 in token prices" The production cost of those 70M tokens are less than $15”

122

Three labor economists debate whether AI will reverse the Industrial Revolution by making human labor abundantDavid Autor argues AI will create new human tasks.

3h5.6k6039

#1257

Top comment: @TheRocketMediaX “@rohanpaul_ai A clear evidence of nobody knows anything where this paranoia is heading towards. The only thing certain is that a lot is about to happen both positive and negative. And historically before a giant leap humankind has alwyas gone through some of the deepest troubles.”

132

Open-source medical AI project OpenMed reaches 340 million model downloads in its first yearOver 650 of the models run fully on-device.

27m4.6k153

#109

Top post: @BrianRoemmele “Wow! In one year… OpenMed has had 340M model downloads. 1,500+ open medical models, all Apache 2.0. 650+ run on your iPhone, fully on-device. Started by one person. Now built by a community. They want to demonize open source platforms. Open source AI changes the world.”

142

Teortaxes argues 3D printing drives rapid FPV drone iteration, countering skepticism over consumer printing utilityThe drone-building workflow is highly popular in Ukraine and Japan

5h3.2k306

#403

Top post: @yacineMTB “@teortaxesTex Super super popular hobby in Japan too. I've been absolutely shocked with what some of the twitter hobbyists are capable of”

152

Teortaxes argues GLM 5.2's 11-point benchmark jump brings open-source models close to parity with Claude OpusThe model's index score rose from 40 to 51.

4m1.8k164

#501

Top post: @teortaxesTex “V4-Flash equaling GLM-5.2's score would also be neat (though I think it's far closer to its ceiling vs Pro) – "almost frontier" on consumer hardware. One cheap area of improvement for both is atrocious hallucination discipline sounds a bit too good to be true”

1610

Meta AI releases Brain2Qwerty v2, a non-invasive decoder that translates raw neural signals into full sentences in real timeThe system is designed for clinical assistive communication.

34m39581

#437

Top comment: @AIatMeta “To help accelerate neuroscience breakthroughs, we're releasing the full training code for Brain2Qwerty v1 and v2, and our partner, @bcbl_, is releasing the v1 dataset. Learn more and explore the artifacts here: https://go.meta.me/42ed9c”

179

XLang NLP releases OSWorld 2.0 to test computer-use AI agents on complex, multi-hour workflowsTop-performing Claude Opus achieved only 20.6% accuracy.

4h1000

#503

Top comment: @XLangNLP “🧵[2/8] What does "long-horizon and real-world" mean here? In one reimbursement task example, the agent must: 📄 follow a tutorial PDF 🏢 operate a legacy ExpenseFlow portal 📬 extract amounts from noisy receipt emails 🏦 cross-check evidence across bank transactions and email 📨 notice a new email that changes the task mid-execution 🔎 recover hidden employee info from a prior report 🧾 prepare valid supporting documents ⚠️ detect inconsistent evidence 🙋 ask the user instead of guessing ✅ complete final review and submission This is the kind of workflow where current agents still break.”

180

Pleias CTO Pierre-Carl Langlais argues large corporations have the compute but lack the specialized engineering talent to scale LLMsServing GLM-5.2 concurrently requires specialized inference provider expertise.

3h585140

#830

Top post: @_xjdr “@Dorialexander there are probably not a whole lot of people i would imagine could serve glm5.2 to more than a handful of concurrent users who's full time job isn't currently at an inference provider”

190

Pleias CTO Pierre-Carl Langlais and builder @xlr8harder say standard Slurm schedulers struggle with synthetic data generation workloadsA custom orchestration tool is being built to bypass Slurm

4h348100

#1537

Top post: @xlr8harder “@Dorialexander HPC solution for this is much harder. Slurm is not shaped well for this workflow. I might have a useful tool for that environment, will DM you.”

Today's Highlights

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

Former Meta PM Xiaoyin Qu predicts US and European enterprises will adopt Chinese AI models to self-host on private GPUs

Nomad List founder Pieter Levels criticizes Windows 11 for full-screen, multi-step Microsoft 365 upsell ads on a paid license

Top Stories

Former Meta PM Xiaoyin Qu predicts US and European enterprises will adopt Chinese AI models to self-host on private GPUsQu cites enterprise distrust of Anthropic's data retention policies

Kim Isenberg and Peter Wang argue the US is losing the AI race due to energy and data center constraintsChina uses free models to promote domestic Huawei hardware.

Startup 8090 raises $135 million in Series A funding led by Salesforce VenturesThe capital will fund hiring and infrastructure development

Video Signals

DeepSeek's upcoming V4 models will introduce peak-hour pricing that doubles API rates during peak Beijing hoursThe models support a 1M token context window.

Independent AI researcher Richard Ngo argues David Ricardo's 1817 framework shows capital owners remain vulnerable to expropriation under AGIThe critique challenges whether AGI lab jobs offer protection.

Chinese DRAM manufacturer CXMT signs $3 billion supply deal with Tencent amid talks with Alibaba, ByteDance, and XiaomiCXMT's DDR5 chip yields currently lag behind global competitors.

Nomad List founder Pieter Levels criticizes Windows 11 for full-screen, multi-step Microsoft 365 upsell ads on a paid licenseEpic Games CEO Tim Sweeney joined the design criticism.

Anthropic CEO Dario Amodei argues open-source AI cannot be monitored, revoked, or patched for safety after releaseThe clip reignited debate over open-source AI safety risks

e/acc founder Beff argues employing genius physicists to work on incremental LLM mutations is a tragedyTeortaxes sparked the debate, noting many physicists join Anthropic.

Teortaxes argues export controls on Chinese memory manufacturers have reversed overcapacity concerns, leaving them unable to meet global demandThe supply gap fuels intense competition for critical fabrication tools

X's Premium+ subscription provides up to $286 in retail API tokens for under $15 in backend compute costsThe package includes developer access to Composer 2.5 Fast.

Three labor economists debate whether AI will reverse the Industrial Revolution by making human labor abundantDavid Autor argues AI will create new human tasks.

Open-source medical AI project OpenMed reaches 340 million model downloads in its first yearOver 650 of the models run fully on-device.

Teortaxes argues 3D printing drives rapid FPV drone iteration, countering skepticism over consumer printing utilityThe drone-building workflow is highly popular in Ukraine and Japan

Teortaxes argues GLM 5.2's 11-point benchmark jump brings open-source models close to parity with Claude OpusThe model's index score rose from 40 to 51.

Meta AI releases Brain2Qwerty v2, a non-invasive decoder that translates raw neural signals into full sentences in real timeThe system is designed for clinical assistive communication.

XLang NLP releases OSWorld 2.0 to test computer-use AI agents on complex, multi-hour workflowsTop-performing Claude Opus achieved only 20.6% accuracy.

Pleias CTO Pierre-Carl Langlais argues large corporations have the compute but lack the specialized engineering talent to scale LLMsServing GLM-5.2 concurrently requires specialized inference provider expertise.

Pleias CTO Pierre-Carl Langlais and builder @xlr8harder say standard Slurm schedulers struggle with synthetic data generation workloadsA custom orchestration tool is being built to bypass Slurm

Recent Stars

Github Stars

Yesterday's Top Stories, Jun 28, 2026.

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 defends University of Washington PhD thesis on decomposing large language modelsShi has made her complete defense slides publicly available.

Policy analyst Dean W. Ball sparks a debate over an AI labor essay that is actually David Ricardo's 1817 textAI safety researchers deemed the replacement premise highly plausible

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

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.

Next.js creator and Vercel CEO Guillermo Rauch argues professionals should replace LinkedIn with self-owned personal websitesCreator Sophie backed this, citing career success from Twitter.

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.

Boris Cherny, creator of Claude Code at Anthropic, proposes five product team archetypes for modern AI developmentThe framework defines prototypers, builders, sweepers, growers, and maintainers.

A resurfaced 2015 email thread reveals "AI Summer" was a top candidate name for what became OpenAITechEmails shared the document detailing early naming brainstorms

Noah Smith argues merit-based immigration policies should target candidates who possess greater talent than the selectors themselvesMike Maples Jr. compared the approach to corporate hiring.

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.

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

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.

OpenAI Codex engineering lead Thibault Sottiaux says his team is investigating unexplained token usage spikesAn engineer jokingly blamed his own high token consumption.

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.

Geoffrey Hinton recommends physicist Adam Brown's lecture on how AGI will transform physics researchThe lecture covers AI scaling to Olympiad-level mathematics.

Databricks' Yuchen Jin says GLM-5.2 matches Claude's impact as enterprises adopt open-source models to own their weightsGary Marcus argues the trend proves a lack of moats

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.

XPRIZE founder Peter Diamandis says data centers are being sent into space because Earth cannot generate energy fast enoughElon Musk agreed, calling the orbital transition inevitable.

Elon Musk confirms xAI will release a 1.5-trillion parameter model in July and a 2-trillion parameter model in AugustThe 1.5T model uses training data from Cursor

DeepSeek releases DeepSpec, a collection of accelerated open-weight models optimized for local deployment based on Qwen3 and Gemma4The four models range from 4B to 14B parameters

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.

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

WhatsApp Leads Global Apps in Addictiveness and Stickiness Metrics

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

Theia Vogel, repeng maintainer, argues that tool-like AI safety plans fail because future systems will inevitably become autonomous agentsSafety researcher j⧉nus called mainstream human-control predictions contradictory.

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.

DeepSeek-AI and Peking University open-source DSpark, using speculative decoding to boost LLM inference throughput by up to 400%The method also accelerates Gemma and Qwen model inference.

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.

AI safety researcher David Manheim proposes a consensus statement to standardize AI evaluations and require replication packagesElon Musk suggested publishing evaluation frameworks directly on X