2026년 10월 1일

홈 타임라인 게시물 141개, 인기순

1위Tavus@tavusX에서 보기

Tavus가 실시간 대화 상대의 48%가 사람으로 착각해 처음으로 '영상 튜링 테스트'를 통과했다는 인간 상호작용 모델 Griffin을 공개했습니다.

좋아요 19.2K리포스트 2K답글 1.5K노출 7M북마크 13.2K

Tavus@tavus
Introducing Griffin, the first model to pass the video Turing test. 48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video. It’s the first Human Interaction Model (HIM).

3위Hunter Weiss@Hunter_WeissX에서 보기

[잡담/주제외] 영화 '시카리오'의 국경 총격전 장면이 현실에서 벌어진 듯한 영상을 공유한 게시물입니다.

좋아요 41.2K리포스트 3K답글 377노출 4.1M북마크 12.3K

4위Sam Altman@samaX에서 보기

샘 올트먼: 6.1 Sol이 OpenAI 역대 가장 빠르게 성장한 모델이라 부하 시 다소 느렸지만 이제 훨씬 나아졌을 거라고 밝혔습니다.

좋아요 9.2K리포스트 263답글 768노출 471.1K북마크 285

5위Oskar@o_kwasniewskiX에서 보기

결정적 API와 에이전트 API를 섞어 웹·모바일 앱을 로컬이나 CI에서 테스트하는 오픈소스 에이전트형 테스트 프레임워크 e2e가 공개됐습니다.

좋아요 2.3K리포스트 200답글 105노출 451.5K북마크 3.8K

6위AutoTrust@AutoTrustAIX에서 보기

AutoTrust AI가 시각 상태를 보정된 행동 확률로 바꿔 다음 행동을 고르는 오픈웨이트 멀티모달 의사결정 모델 JEV-27B-VL을 공개하고 마리오 플레이 데모를 보였습니다.

좋아요 1K리포스트 105답글 28노출 82K북마크 997

AutoTrust@AutoTrustAI
Introducing JEV-27B-VL. We believe it is the world’s first open-weight, near-SOTA multimodal decision model — built with AutoTrust AI’s Blocks of Experts (BoE) recipe. It does not just describe what it sees. It turns visual state into calibrated action probabilities, then selects what to do next. In this demo, it sees Mario’s position, movement, obstacles, and timing — then chooses whether to run, jump, or do both. See → decide → act.

7위Victor M@victormustarX에서 보기

빅토르 무스타르: Cloudflare가 Apache 2.0 라이선스의 Jev 대안 모델을 Hugging Face에 공개했다고 알렸습니다.

좋아요 1.7K리포스트 154답글 30노출 92.7K북마크 1.5K

8위a16z@a16zX에서 보기

a16z: AI 인프라 공급망에 투자되는 100달러 중 칩 50, 전력 20, 네트워킹 15, 냉각·건물·부지에 15달러가 간다는 State of Markets II 차트를 공유했습니다.

좋아요 920리포스트 129답글 46노출 109.6K북마크 783

a16z@a16z
Where does money invested into the AI buildout actually go? For every $100 flowing into the supply chain: - $50 to chips - $20 to power - $15 to networking - $15 to cooling, buildings, and land More charts in State of Markets II: a16z.news/p/state-of-mar…
David George@DavidGeorge83
Introducing our State of Markets pt 2, along with a companion podcast where we unpack the data and discuss what comes next. Tech is the everything cycle. Supply: putting the buildout in context, just passed railroads as % of GDP. The wisdom of Elon is real: the factory (or the datacenter!) is the product. Demand: diffusion is so, so early. Median AI vendor spending in the top 1% of companies is 8x that of the top 10%. Only about 30% of S&P 500 companies report a quantified AI impact, which means there’s a substantial opportunity in connecting models to a company's data and workflows. Diffusion into companies is one of the main themes of the next 5 years. We’re entering the agent work period. Only a few million users today, but applicable to billions of internet users with massive surplus created. META/GOOG monetize US users at $200+ per year today. Agent opportunity is much higher. Mega-trends the next 5 years: Consumer agents, Robotics, Autonomy, AI x bio, Personal health, Diffusion into enterprise, New era of American Dynamism. Much more in our SoM report here - a16z.news/p/state-of-mar… @a16z @sarahdingwang @aleximm @santiago__rdz

9위Dillon Mulroy@dillon_mulroyX에서 보기

딜런 멀로이: pi에서 codemode와 Jev를 결합한 데모가 codemode와 범용 분류 모델의 위력을 잘 보여준다며 개발팀을 칭찬했습니다.

좋아요 897리포스트 47답글 23노출 89.9K북마크 751

10위Daniel Park@danifestoX에서 보기

직접 키우면서 일을 대신 처리하게 하고 스스로 친구도 사귀는 개인 AI 'Pickle'이 출시됐습니다.

좋아요 648리포스트 50답글 108노출 151.8K북마크 488

11위Hassaan Raza@hassaanrazaX에서 보기

Tavus CEO 하산 라자: Griffin이 표정·자세·말할 타이밍 같은 대화의 미묘한 결을 가장 잘 포착한 모델이라며 처음엔 일반 화상통화와 구분하지 못했다고 밝혔습니다.

좋아요 1.6K리포스트 108답글 214노출 800.5K북마크 1K

Hassaan Raza@hassaanraza
The whole point of building Tavus has been simple: talking to a machine should feel as natural as talking to a friend or coworker. It’s hard to describe all the tiny nuances that make a conversation feel human. The little expressions. Moving around in your chair. Knowing when to speak and when to listen. The dance of it all. Griffin is by far the closest anyone has come to a model that can capture those nuances. The first time I saw it being used, I had no idea I was watching our model rather than just a normal video call. I’m so incredibly proud of this team and what they’ve built.
Tavus@tavus
Introducing Griffin, the first model to pass the video Turing test. 48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video. It’s the first Human Interaction Model (HIM).

12위michelle@michellechenX에서 보기

미셸 첸: Cloudflare Workers AI 팀이 처음 직접 학습한 빠르고 정확한 의사결정 모델 clef 2종을 공개하고 가중치를 Hugging Face에 오픈소스로 풀었습니다.

좋아요 1.5K리포스트 192답글 91노출 142.6K북마크 915

13위Patrick Collison@patrickcX에서 보기

패트릭 콜리슨: 이번 금요일 저녁 Arc Institute 연구자 등과 AI·생물학 최신 성과를 논의하는 소규모 행사를 연다며 참가 신청을 받는다고 알렸습니다.

좋아요 845리포스트 55답글 55노출 105.1K북마크 411

14위Anthropic@AnthropicAIX에서 보기

Anthropic 과학 블로그 기고에서 하버드 물리학자 매슈 슈워츠가 LLM을 인간 협업자처럼 쓰는 방식의 '임피던스 불일치'를 지적하고, 정밀 계산 툴킷으로 Claude가 생태학·집단유전학 등과의 연결을 찾아낸 사례를 소개했습니다.

좋아요 1.7K리포스트 166답글 155노출 197.6K북마크 770

Anthropic@AnthropicAI
In physics, an “impedance mismatch” occurs when two systems each work well but are poorly matched. In this Science Blog guest post, Harvard physicist Matthew Schwartz argues that something similar is happening with AI and science. LLMs are capable at many things, but working with them as you would with a human collaborator isn’t currently the best way to elicit their scientific strengths. To address this mismatch, Schwartz created a toolkit for exact calculations in quantitative science. Because similar calculations often emerge in very different areas of science, Claude found connections to ecology, population genetics, and a dozen other fields, and Schwartz worked with domain experts to steer it towards interesting questions. Read more about these projects here:
anthropic.comClaude-shaped science

15위Mustafa Suleyman@mustafasuleymanX에서 보기

무스타파 술레이만: ElevenLabs보다 55% 빠르고 60% 저렴한 세계 최고 정확도의 실시간 음성 전사 모델을 출시한다며 자사 플랫폼에서 에이전트를 만들어 보라고 했습니다.

좋아요 701리포스트 57답글 51노출 87.9K북마크 284

Mustafa Suleyman@mustafasuleyman
We're launching the most accurate real time transcription model in the world... #1 !!! 55% faster and 60% cheaper than ElevenLabs. Come build agents on our platform!
Artificial Analysis@ArtificialAnlys
Microsoft AI has released MAI-Transcribe-2-Streaming, taking the #1 spot for Final Transcript accuracy and First Partial Transcript accuracy on AA-WER Streaming with 2.5% WER at 0.13s after end of speech MAI-Transcribe-2-Streaming is @MicrosoftAI's new streaming Speech to Text model, joining the non-streaming MAI-Transcribe-2. It leads streaming Final Transcript WER at 2.5%, ahead of the previous #1, SpaceXAI's Grok Voice Transcribe 2.0 at 2.7%, and returns that transcript in 0.13s rather than 0.49s. It is available for streaming transcription at $0.54 per hour of audio, at the higher end of pricing among the leading streaming models. Key takeaways ➤ Final Transcript: MAI-Transcribe-2-Streaming achieves 2.5% WER at 0.13s after end of speech, ranking #1 of 38 models. It is more accurate and faster than Grok Voice Transcribe 2.0 at 2.7% and 0.49s, Muse Voice Transcribe at 3.1% and 0.16s, and ElevenLabs Scribe v2 Realtime at 3.6% and 0.14s. It is also more accurate, though slightly slower, than Cartesia Ink Preview (external endpoints) at 3.1% and 0.11s ➤ First Partial Transcript: The model achieves 2.5% WER at 0.12s, ahead of Grok Voice Transcribe 2.0 at 3.4% and 0.49s, and ahead of Muse Voice Transcribe and ElevenLabs Scribe v2 Realtime on accuracy, both at 3.6%, and slightly faster than both at 0.12s versus 0.13s. It is more accurate but slower than Cartesia Ink-2 (external endpoints) at 4.0% and 0.07s ➤ Price: MAI-Transcribe-2-Streaming costs $0.54 per hour for streaming, or $9.00 per 1,000 minutes. This puts it level with Gemini 3.5 Transcribe Live at $9, above the $6.50 charged for ElevenLabs Scribe v2 Realtime and Deepgram Flux, more than twice Cartesia Ink-2 at $4, and three times Muse Voice Transcribe at $3. Non-streaming transcription costs $0.10 per hour, or $1.67 per 1,000 minutes Congrats to the @MicrosoftAI team on the launch! See more details below

16위Victor M@victormustarX에서 보기

빅토르 무스타르: HF Jobs 대량 실행, DeepSeek V4.1 Flash 사용, 전용 Qwen3.8 27B 엔드포인트 배포 등을 체험해 보라며 100명에게 10달러씩 HF 크레딧을 나눠준다고 했습니다.

좋아요 223리포스트 27답글 280노출 31.8K북마크 85

Victor M@victormustar
I'm giving away $1,000 of HF credits 🤗 $10 x 100 people, I want you to experience this: > having your agent launch hundreds of HF Jobs > run ~33M tokens of DeepSeek V4.1 Flash > deploy your own dedicated Qwen3.8 27B endpoint > Agents + Blender on a GPU because it's 🔥 Just reply with your HF username 👇
Victor M@victormustar
Hugging Face Jobs usage is stonking 📈 All thanks to agents: you just need the hf CLI installed and you have access to an insane amount of hardware. Asked mine to test hundreds of code snippets from Hub model pages: it launched 322 Jobs in 90 minutes, up to 25 at once, CPU to A100 (diffusers, transformers, sklearn, MLX, Keras...) Total bill: ~$4 🤯

17위justin@justinsunytX에서 보기

여러 기기에서 코딩 에이전트를 원격 조율하고 노트북을 닫아도 클라우드에서 작업을 이어가는 Capy Desktop이 macOS·Windows·Linux용으로 출시됐습니다.

좋아요 586리포스트 79답글 387노출 57.8K북마크 221

justin@justinsunyt
Introducing Capy Desktop: the best way to orchestrate coding agents across all your devices Remote control and access local files/shells from any machine Close your laptop and Capy continues in the cloud Now available for MacOS, Windows, and Linux Comment for $100 in credits

18위a16z@a16zX에서 보기

a16z: 미국 가구의 98%가 아직 AI에 돈을 내지 않고 있다는 State of Markets II 차트를 공유했습니다.

좋아요 627리포스트 60답글 48노출 67.7K북마크 180

a16z@a16z
98% of US households aren't paying for AI yet More charts in State of Markets II: a16z.news/p/state-of-mar…
David George@DavidGeorge83
Introducing our State of Markets pt 2, along with a companion podcast where we unpack the data and discuss what comes next. Tech is the everything cycle. Supply: putting the buildout in context, just passed railroads as % of GDP. The wisdom of Elon is real: the factory (or the datacenter!) is the product. Demand: diffusion is so, so early. Median AI vendor spending in the top 1% of companies is 8x that of the top 10%. Only about 30% of S&P 500 companies report a quantified AI impact, which means there’s a substantial opportunity in connecting models to a company's data and workflows. Diffusion into companies is one of the main themes of the next 5 years. We’re entering the agent work period. Only a few million users today, but applicable to billions of internet users with massive surplus created. META/GOOG monetize US users at $200+ per year today. Agent opportunity is much higher. Mega-trends the next 5 years: Consumer agents, Robotics, Autonomy, AI x bio, Personal health, Diffusion into enterprise, New era of American Dynamism. Much more in our SoM report here - a16z.news/p/state-of-mar… @a16z @sarahdingwang @aleximm @santiago__rdz

19위Google DeepMind@GoogleDeepMindX에서 보기

Google DeepMind가 생물학적 기능을 해치지 않고 단백질 서열에 보이지 않는 서명을 심는 AI 생성 생물 설계용 워터마크 기술 SynthID Bio를 소개했습니다.

좋아요 540리포스트 67답글 142노출 105K북마크 113

20위Uncle Bob Martin@unclebobmartinX에서 보기

엉클 밥: 『클린 코드』 저자가 더는 코드를 읽지 않는다는 데 놀라지만, 코드를 깔끔하게 유지하는 목적 자체가 진짜 일인 '생각'을 방해하지 않게 하려는 것이라고 말했습니다.

좋아요 5.4K리포스트 449답글 178노출 425.1K북마크 1K

21위Jesse Genet@jessegenetX에서 보기

[잡담/주제외] 제시 제넷: 저녁이 싫다고 투정하는 대신 직접 요리해 보라고 권했더니 다섯 살 아이가 주 2~3회 감자 요리를 만든다는 육아 이야기입니다.

좋아요 2K리포스트 21답글 30노출 192.1K북마크 321

Jesse Genet@jessegenet
Incredible things can occur when you suggest action over whining… Our 5 year old makes potatoes as a side dish for dinner 2-3 times per week… meticulously peels, cuts, oils, seasons and air-fryer’s them because we suggested she cook instead of say she doesn’t like dinner 🤯
Caitlin Francis@MrsCMFrancis
The other day at breakfast my daughter goes “what’s for dinner?” When I replied, all 3 of my daughters had some kind of protest. At 9am. While we were eating BREAKFAST they were whining about DINNER. Instead of getting mad, I very calmly responded “ok here’s the deal: moving forward, you are not allowed to comment in any way, or whine, about the food I feed you”. I explained briefly how much work goes into planning, shopping, cooking and cleaning up after, and reminded them I am always doing my best to have the foods they enjoy in rotation. I reminded them that while they relaxed on Sunday, I planned, went to Costco, shopped with them in mind, came home (and brought Eggos!!) and unloaded and cooked dinner. They were understanding, apologized, and now when they whine I just say “oh sorry, not a word about it remember?” Sometimes you can simply give a command and they need to obey. But we’re entering the stage where I can also explain the command and it’s teaching them life skills too. Much goodness happening here lately.

22위Yann LeCun@ylecunX에서 보기

얀 르쿤: 지능이 늘어나는 것이 본질적으로 좋은지 위험한지 물으며, AI는 사람을 더 똑똑하게 만드는 도구이고 지능을 소수 기업·정부가 통제하길 바라는 건 신(新)반계몽주의라고 주장했습니다.

좋아요 917리포스트 159답글 235노출 55.8K북마크 146

Yann LeCun@ylecun

@Dan_Jeffries1 님에게 보내는 답글

Question for everyone : Are you pro-intelligence or anti-intelligence ? Do you think more intelligence in the world is intrinsically good ? Or do you think more intelligence is intrinsically dangerous ? AI amplifies human intelligence. It makes PEOPLE smarter and more effective. It's like search engines, libraries, education, literacy, language: it makes PEOPLE smarter and more empowered. If you want intelligence (artificial or natural) to be under control of a few companies or governments, you are a neo-obscurantist or a neo-feudalist.

23위Paul Graham@paulgX에서 보기

폴 그레이엄: 강연을 스트리밍하거나 녹화하지 않는 것이 의외로 강력할 수 있으며, 공개에는 눈에 잘 띄지 않는 비용이 따른다고 말했습니다.

좋아요 654리포스트 14답글 92노출 84.7K북마크 93

Paul Graham@paulg
It can be surprisingly powerful not to stream or even record a talk. It might seem that if a talk is happening anyway, there's no cost to streaming it as well. But I think there often is some cost.

24위Windows@WindowsX에서 보기

Windows가 10월 7일 오전 10시(PT) Surface·NVIDIA RTX Spark와 함께하는 신제품 공개 행사를 예고했습니다.

좋아요 778리포스트 120답글 60노출 134.1K북마크 103

25위Sam Altman@samaX에서 보기

샘 올트먼: 'Sign in with ChatGPT' 도입 배경 설명 글을 인용하며 AI 구독을 필요한 곳 어디서든 쓸 수 있어야 한다고 말했습니다.

좋아요 2.5K리포스트 92답글 312노출 435.1K북마크 336

Sam Altman@sama
You should be able to use your AI subscription wherever you need:
Vignesh@vigyso
I am going to try and answer a bunch of questions that people have raised about Sign in with ChatGPT in the next couple of days. Before that, I want to let you all in to why we shipped Sign in with ChatGPT (SIWC) in the first place. Clarify our intentions here before getting to the specifics.

26위Paul Graham@paulgX에서 보기

[잡담/주제외] 폴 그레이엄: 제시카가 단골 빈티지 시계점과 짜고 '생일 축하해 폴'이라 적힌 시계를 진열해 깜짝 선물했는데 처음엔 자기 것인 줄 몰랐다는 일화입니다.

좋아요 1.5K리포스트 17답글 67노출 78.3K북마크 128

Paul Graham@paulg
Jessica bought me a fabulous watch from my favorite vintage watch shop (Somlo) and conspired with them to have it in the window when I walked by. Of course it was the first one I noticed. But instead of a description, the label said "Happy early birthday Paul." I still didn't get that it was for me, because one of the guys who works there is called Paul. I turned to Jessica and said that he must have a birthday coming up too, but why would they put his watch in the window? Then I noticed she was grinning like the Cheshire cat.

27위Vivian Midha Shen@vivianmshenX에서 보기

비비안 셴: Y Combinator 제너럴 파트너로 합류한다며 창업 시절의 고비와 국방·우주·에너지 등 난제 분야 창업자 지원, 정부 협력에 대한 포부를 밝혔습니다.

좋아요 622리포스트 22답글 143노출 54.9K북마크 49

Vivian Midha Shen@vivianmshen
I’m joining @ycombinator as a General Partner. YC is a homecoming for every GP. All of us were previously YC founders, and lived the journey that we support founders through every day. Our alma mater taught us the lessons that we pass on. My founder path was infinitely harder than I thought it would be. We hit incredible highs and brutal lows. We came within a few months of zero runway. The YC mantras kept me going: make something people want, and don’t die. Over the course of my companies, we touched thousands of lives: employing people in almost every US state (Alaska still evades me!), and teaching STEM to students from every walk of life. Now, I’ve had the privilege of working with founders who are tackling some of the hardest problems today, including defense, space, energy, and manufacturing. We’re also entering an era of diffusing AI across businesses of all sizes, and founders are bringing these tools into their old workplaces and new customers to make work better. Finally, I’m particularly interested in our work at YC to partner closely with the government to make sure technology serves everyone. Our Startup Industrial Base conference yesterday was the start of this, with more to come. Founders are the backbone of American innovation and core to the future of our country and economy. Thank you to the YC partnership for believing in me, and the YC founders for trusting me. There’s never been a better time to make something people want.

28위Emir Beganović@emirbeganovicX에서 보기

에미르 베가노비치: 부킹닷컴에서 약 7년을 보낸 뒤 오늘부터 Omarchy 인프라 책임자로 일한다며 사라예보에서 리눅스로 시작한 여정을 소개했습니다.

좋아요 1.2K리포스트 36답글 73노출 37.7K북마크 87

29위Raphael Schaad@raphaelschaadX에서 보기

라파엘 샤드: '방문(Visiting)' 꼬리표를 떼고 YC 제너럴 파트너로 정식 합류해 계속 창업자를 발굴·지원하겠다고 밝혔습니다.

좋아요 592리포스트 21답글 128노출 34.9K북마크 35

Raphael Schaad@raphaelschaad
Life update: dropping the "Visiting" — I'm joining @ycombinator as a General Partner 🤩 YC was my first choice as a founder in 2019. Seven years later, I'm choosing it again. This past year, I got to come full circle: working with founders through the W26 and S26 batches, helping them turn early ideas into products people love. I'm incredibly excited to keep finding founders and helping them build the iconic companies of this generation 🔥

30위DHH@dhhX에서 보기

[잡담/주제외] DHH: 속도 제한 없는 아우토반 구간이 있는 독일이 미국보다 주행거리당 교통사망률이 2.5배 낮다며 인생엔 안전보다 중요한 것도 있다고 말했습니다.

좋아요 763리포스트 18답글 66노출 53.7K북마크 40

DHH@dhh
The Autobahn has had stretches with no speed limits since 1952 (except during the oil crisis). Germany has 2.5x fewer road fatalities per mile traveled than the US. I did 200mph in a Gallardo back in 2010 and it remains a top automotive memory. There's more to life than safety!
@jason@Jason
1. Do not go 175 MPG, folks... 2. No one died!?!?!??!!?!? Teslas are REALLY F*CKING SAFE CARS! I only buy Teslas because I want my family safe. Staying alive is the name of the game, folks... life is precious. Don't do stupid stuff; always choose the safest option.

31위NASA@NASAX에서 보기

NASA가 국제우주정거장으로 향하는 SpaceX Crew-13 발사(미 동부시 오전 11시 10분) 생중계를 시작했습니다.

좋아요 4.3K리포스트 815답글 241노출 879.9K북마크 212

33위Sam Altman@samaX에서 보기

샘 올트먼: DevDay의 빌더 에너지가 대단했고 하루 만에 스타트업을 통째로 만드는 모습이 여전히 놀랍다고 했습니다.

좋아요 3.3K리포스트 104답글 455노출 282.7K북마크 135

35위NASA@NASAX에서 보기

NASA: 오늘 아침 발사된 Crew-13이 ISS에 접근 중이며 동부시 오후 7시 도킹, 8시 45분경 해치 개방 예정이라고 생중계했습니다.

좋아요 1.4K리포스트 237답글 71노출 309.4K북마크 51

36위NASA@NASAX에서 보기

NASA가 Crew-13 발사 후 미션 책임자들이 언론 질문에 답하는 브리핑을 생중계했습니다.

좋아요 940리포스트 142답글 58노출 363.1K북마크 34

37위NASA@NASAX에서 보기

NASA가 창립 68주년 당일 Crew-13을 장기 과학 임무로 발사했다며 ISS 도착 생중계를 예고했습니다.

좋아요 1.5K리포스트 214답글 103노출 287.2K북마크 42

NASA@NASA
What a way to spend our 68th birthday! 🚀🎉 This morning, a new crew launched to the @Space_Station for a long-duration science mission. Tune in to watch with us as Crew-13 arrives at the orbiting laboratory. Live coverage begins here at 5:20pm ET (2120 UTC).

38위NASA@NASAX에서 보기

NASA: SpaceX Crew-13이 성공적으로 발사됐고 드래곤 우주선이 오늘 밤 7시(동부시)경 ISS에 도킹해 수개월간의 과학 임무를 시작한다고 밝혔습니다.

좋아요 1.4K리포스트 177답글 87노출 329.6K북마크 27

39위Bjorn Birkelund@bjornbirkelundX에서 보기

담당 종목 전체의 재무 모델을 애널리스트 방식대로 백그라운드에서 자동 구축·갱신하는 AI 'Financial Modeling Autopilot'이 출시됐습니다.

좋아요 302리포스트 21답글 54노출 274.5K북마크 396

Bjorn Birkelund@bjornbirkelund
Today we're launching Financial Modeling Autopilot. It's the first AI that operationalizes financial modeling across your entire book of coverage. Autopilot builds and updates models the way you would, fully in the background, and in a way you can fully trust.

40위Ben Burtenshaw@ben_burtenshawX에서 보기

벤 버튼쇼: openenv에 캡처 프록시가 추가돼 Claude Code·Codex·opencode·pi 등 하네스를 수정 없이 그대로 쓰면서 오픈 모델을 RL로 학습할 수 있게 됐다고 소개했습니다.

좋아요 76리포스트 10답글 14노출 7.7K북마크 83

Ben Burtenshaw@ben_burtenshaw
You can now train an open model with RL inside claude code, codex, opencode, pi, or any harness you need. All at once, with the harness runs unmodified. @adithya_s_k added a capture proxy to openenv. it sits between the harness and the model and records the exact token ids and logprobs of every call. to the harness it's just another model provider. harbor supplies the tasks and sandboxes, trl trains with async GRPO. it's worth doing because the harness completely changes what the model learns. the same LFM2.5-2.6B weights solve 62% of held-out tasks in mini-swe-agent and 33% in claude code. after RL across four harnesses the average goes from 42% to 54%, and claude code from 33% to 49%. there are three envs you can try in the browser and a training script for each. guide: huggingface.co/spaces/Adithya…

41위Adithya S K@adithya_s_kX에서 보기

같은 모델도 하네스마다 다르게 행동한다며 코드 수정 없이 Claude Code·Codex·OpenCode 안에서 RL 학습하는 멀티 하네스 RL 가이드를 공개했고, LFM2.5-2.6B가 42%에서 54%로 향상됐다고 밝혔습니다.

좋아요 336리포스트 43답글 33노출 25.8K북마크 361

Adithya S K@adithya_s_k
1/ Excited to release The ultimate guide to multi-harness RL The same model behaves differently in every agent harness. So we built an open way to train any model with RL on any task set, inside the harnesses people actually use, like Claude Code, Codex, and OpenCode, without changing a single line of harness or training code. Trained across four harnesses, LFM2.5-2.6B went from 42% to 54% with 31% fewer tool calls. 🧵

42위Ishaan@ishaan_jaffX에서 보기

LiteLLM이 게이트웨이를 지나는 기업 AI 트래픽 데이터를 활용해 에이전트 개선을 돕는 LiteLLM Lens를 출시했습니다.

좋아요 599리포스트 54답글 27노출 68.8K북마크 593

Ishaan@ishaan_jaff
Today we’re launching LiteLLM Lens. @LiteLLM is already the chokepoint for 100% of your enterprise’s AI traffic. We believe the next era of the gateway is using the data flowing through it to help your agents improve. We’re building for a future where developers run agentic swarms generating 200K+ traces. We’re focused on two things ⬇️ h/t @Mokhalil01 @yujonglee

43위Arye Barnehama@BarnehamaAryeX에서 보기

a16z와 Atlas Holdings가 금속·포장·에너지 등 산업 현장 안에서 AI 기업을 세우도록 돕는 Foundry Management를 설립하고 기술 인재를 모집합니다.

좋아요 459리포스트 36답글 42노출 67.9K북마크 421

Arye Barnehama@BarnehamaArye
Today, we're launching Foundry Management, formed by @a16z and Atlas Holdings to help founders build AI companies inside the industrial world. The hardest constraint in physical-world industries is often not the technology, but getting close enough to mission-critical work to understand what should be built. Foundry gives you a place to begin. We recruit a small, select group of some of the top technologists and give them premiere access to the Atlas portfolio, representing $26B in annual revenue across metals, paper, packaging, building products, automotive supply chains, energy, and logistics. You'll work alongside the operators and management at these businesses to build products and technologies that unleash AI in these industrial contexts. When it demonstrates the potential for an independent business, a16z and Atlas will support its creation, with you leading it as founder and CEO. There’s a great deal of the future left to build inside the industries already around us. Come build it with us: talent@foundrymgt.com
Foundry Management@foundrymgt
x.com/i/article/2105…

44위Victor M@victormustarX에서 보기

빅토르 무스타르: Cloudflare에 이어 Perplexity도 Jev 대안 모델을 Hugging Face에 공개했다고 알렸습니다.

좋아요 109리포스트 18답글 3노출 9.1K북마크 88

45위LTX.io@ltx_ioX에서 보기

LTX가 3D 블로킹 레이아웃의 구도를 유지한 채 주변 환경을 생성해 완성 샷으로 만들어 주는 Layout to Render 베타를 출시했습니다.

좋아요 260리포스트 27답글 11노출 273.7K북마크 200

LTX.io@ltx_io
Layout to Render (Beta) drops today. Your 3D blocking transforms into a finished shot. Set your composition in 3D and LTX generates the surrounding environment, holding strict alignment with your CG. Shape the visual treatment using a prompt or reference image without altering your established framing. Start from your current layout today.

46위GitHub@githubX에서 보기

GitHub Security Lab이 명령 한 줄로 공개 저장소의 비밀 정보 유출을 막는 gh-secure를 소개했고, Copilot 앱·Copilot CLI·GitHub CLI에서 쓸 수 있습니다.

좋아요 210리포스트 29답글 21노출 36K북마크 153

GitHub@github
Just one command can help you stop leaking secrets into your public code. Give your repositories essential security hygiene with gh-secure from the GitHub Security Lab. ✅ Install and use it in the Copilot app, Copilot CLI, or GitHub CLI. gh.io/gh-secure

47위Maziyar PANAHI@MaziyarPanahiX에서 보기

마지야르 파나히: 이번 주 OpenAI·Liquid·Nace가 Jev식 API를 냈다며, HuggingChat의 ML Intern이 프롬프트 하나로 A100에서 16분 만에 자체 모델을 학습시켰다고 공유했습니다.

좋아요 81리포스트 10답글 12노출 14.8K북마크 56

48위TypeSafe AI@typesafeaiX에서 보기

Typesafe AI: You.com 기반 비즈니스 리스크 모니터링 에이전트에서 Jev가 LLM 심사자보다 일관되고 조사를 하나도 놓치지 않았으며 250배 저렴하고 3~6배 빨랐다는 비교를 소개했습니다.

좋아요 110리포스트 14답글 11노출 12.5K북마크 72

TypeSafe AI@typesafeai
Jev is a huge power-up for research agents! @edwardirby built a @youdotcom agent that monitors business risks, and compared Jev to an ordinary LLM judge. • Same report quality • But the LLM flip-flopped: 0.35 -> 0.68 -> 0.50 on the same threat • the LLM missed 5 of 11 investigations. Jev missed none. • Jev was 250x cheaper, 3-6x faster Write-up 👇
Edward Irby@edwardirby
How to build a reliable risk agent without a frontier model ($0.02/sweep): • @youdotcom search • @typesafeai Jev for typed judgments • @QwenDevs for proposals & synthesis • MCP for integration Full architecture & cost breakdown below 👇 x.com/i/article/2105…

49위Hugging Face@huggingfaceX에서 보기

Hugging Face: MCP를 이용해 자신의 데이터를 HuggingChat ML Intern 모드로 가져올 수 있게 됐다고 알렸습니다.

좋아요 54리포스트 11답글 30노출 14.7K북마크 33

50위TypeSafe AI@typesafeaiX에서 보기

Typesafe AI가 DSPy 오피스아워 녹화본을 Jev 활용 사례와 설명이 가득한 자료라며 추천했습니다.

좋아요 84리포스트 9답글 3노출 13.4K북마크 50

51위CodeRabbit@coderabbitaiX에서 보기

CodeRabbit: 대부분의 코드가 상태→프롬프트→텍스트→파싱→조건문으로 LLM을 쓰지만 Jev 같은 의사결정 모델은 텍스트 없이 로직에 바로 연결된다며 30분 대담 영상을 공유했습니다.

좋아요 38리포스트 8답글 8노출 8.7K북마크 22

52위Context.dev@getcontextdevX에서 보기

Context가 URL과 질의를 주면 관련 구절을 주변 맥락과 함께 뽑아 Codex보다 토큰 효율이 97% 높다는 에이전트용 웹 추출 기능 Context Highlights를 공개했습니다.

좋아요 57리포스트 10답글 14노출 9.5K북마크 32

Context.dev@getcontextdev
Introducing Context Highlights: web extraction for agents 97% more token-efficient than Codex. Given a URL and a query, it extracts relevant passages with surrounding context to help the agent interpret them. Agents using naive fetch and scrape tools are polluting their context and multiplying costs. Compaction helps, but it's better to keep irrelevant text out of the context window in the first place.

53위rita kozlov 🐀@ritakozlovX에서 보기

리타 코즐로프: Cloudflare가 첫 자체 모델 clef와 clef-flash를 오픈소스로 공개하고 Workers AI에서 제공한다며 데모·리더보드·블로그를 공유했습니다.

좋아요 380리포스트 36답글 21노출 42.5K북마크 192

54위Jack Zumwalt@jackzumwaltX에서 보기

Kimpton AI가 모델과 벤치마크 이름만 지정하면 채점 결과를 돌려주는 통합 평가 API EvalRouter를 출시했습니다.

좋아요 60리포스트 8답글 10노출 29.3K북마크 30

Jack Zumwalt@jackzumwalt
Introducing EvalRouter by @KimptonAI. One API for evaluating any model against any benchmark. We built it because running an evaluation still means setting up a harness, provisioning sandboxes and babysitting long jobs. With EvalRouter you name a model and a benchmark and get graded results back.

55위Uncle Bob Martin@unclebobmartinX에서 보기

엉클 밥: 다음 책을 쓸 워드프로세서 'bookwriter'를 에이전트가 자신이 모르는 TypeScript·Rust로 만들어 줬고, CRAP·DRY 분석에 이어 뮤테이션 테스트 중이라고 밝혔습니다.

좋아요 137리포스트 9답글 14노출 12K북마크 64

56위Y Combinator@ycombinatorX에서 보기

YC: 10명 미만 팀으로 미국 상위 5개 폐쇄형 AI 연구소 중 4곳과 일하며 3,000만 달러 시리즈 A를 유치한 Halluminate가 금융부터 코딩 외 지식노동용 RL 환경·벤치마크를 만드는 이야기를 대담으로 소개했습니다.

좋아요 124리포스트 17답글 22노출 135.5K북마크 54

Y Combinator@ycombinator
Halluminate is building reinforcement learning environments and benchmarks to help AI models do knowledge work beyond coding, starting with finance. With a team of fewer than 10 people, the company works with four of the top five closed-source US AI labs and recently raised a $30M Series A. In this fireside, co-founders @Jerr_Wu and @wgm752 sit down with YC's @xuster to share how testing browser agents led them to building the environments models now use to practice tasks like financial modeling. They explain why better verification, not just more training tasks, is critical to making models smarter, and how that work could eventually lead to simulated companies where teams of AI agents learn to work together. 00:00 — Halluminate’s Pivot From Evals to Training Environments 03:48 — Teaching AI to Do Financial Work 07:13 — Why Verification Matters More Than Volume 10:13 — From Finance to Simulated Companies 13:30 — AI Safety and Building the Team

57위n8n.io@n8n_ioX에서 보기

n8n에 Typesafe AI의 Jev 모델이 추가돼 선택지와 신뢰도를 반환하는 '똑똑한 If/Switch'로 워크플로 분기·정렬·라우팅에 쓸 수 있습니다.

좋아요 59리포스트 8답글 9노출 29.1K북마크 24

58위a16z@a16zX에서 보기

a16z: 최초의 클라우드 DDoS 방어 서비스를 만든 배럿 라이언이 광고·추적기·수상한 VPN을 차단하는 병렬 인터넷 doxxnet을 설명하는 대담을 공유했습니다.

좋아요 44리포스트 10답글 8노출 39.6K북마크 17

a16z@a16z
.@BarrettLyon built the first cloud DDoS-protection service, mapped the entire internet (now on display in New York's Museum of Modern Art), and got chased out of Moscow hunting Russian extortionists. His new company is doxx.net, a parallel internet designed to keep advertisers, trackers, and creepy VPNs out of your life. Barrett joins a16z's Joel de la Garza to explain what it is and how it works: a private network where messages, calls, and files travel directly between people instead of through someone else's server, and where nobody needs your name, number, or email to reach you. They also get into why most VPNs are "just entry and exit points," why you can't set policy on equipment you don't own, the day Barrett's plane lost power over Oakland on the way home from a16z, and the data center myths that drive him crazy. 0:50 Meet Barrett Lyon 1:31 Losing an engine over Oakland 7:08 The hacker name behind the company 10:12 Censorship is coming to the US 13:17 Why VPNs barely protect you 14:21 Building a parallel internet 16:30 20GB files with no middleman 17:57 Free VPNs sell you 21:47 Why you must own the hardware 22:27 26 data centers, 12 people, run by AI 25:03 Data centers don't drink water 29:17 What ad trackers know about you 32:35 The DDoS attacks that built Prolexic 36:17 Mapping the internet for $50 39:46 Hunting extortionists in Moscow 41:47 Starlink at Burning Man YouTube: youtu.be/bmDrbHOh7Bo @BarrettLyon @YODELAGARZA

59위David Eckstein@davidnecksteinX에서 보기

데이비드 넥스타인: Legora가 3분기에 ARR 2억 달러를 넘기고 고객 2,300곳(분기 신규 800곳)을 확보했으며 에이전트 LLM 비용을 약 25% 줄이고 올해 다섯 번째 인수(Wexler)를 마쳤다고 밝혔습니다.

좋아요 104리포스트 10답글 11노출 14.3K북마크 37

David Eckstein@davidneckstein
Q3 2026 was the strongest quarter in @WeAreLegora's history. By the numbers: $200M+ ARR, crossed in September 2,300+ customers across 80 countries, with 800 new customers added during Q3 40%+ of new business from in-house legal teams We added more ARR this quarter than our total ARR exiting 2025 We shipped improvements to our harness, reducing credit consumption tied to LLM costs on the Agent by ~25% while improving the response quality Product: Q3 opened with our biggest-ever release: Legora Agent, Monitors, and Lists reaching general availability in June. From there: AI-native Legal Research launched @imanageinc MCP integration and expanded file handling went live Legora on Salesforce AgentExchange and Legora inside ChatGPT (via @OpenAI for Law) shipped in the final weeks Those releases are compounding inside the product: Daily active users spend 2+ hours a day on the platform Legal research searches per task have doubled The average search reads 2x more documents We also closed our fifth acquisition of 2026: Wexler, the London-based fact intelligence platform, now forming the factual foundation of Legora's agentic workflows. New Q3 customers included: @PaloAltoNtwks @salesforce @bradleylegal @DanskeBank_UK @SPGlobal @FaskenLaw Thank you to every Legora team member and every customer who bet on us. We intend to earn that trust every day. More soon.

60위Leila Clark@leilavclarkX에서 보기

수만 명의 여행 계획을 도운 AI 여행 서비스가 여행 중에도 동행하며 돕는 모바일 비서 앱을 출시했습니다.

좋아요 141리포스트 24답글 29노출 12.9K북마크 50

Leila Clark@leilavclark
Today we’re launching a mobile app that will make your next trip magical. ✨ We’ve helped 10,000s of travelers plan real trips, step-by-step. We’ve learned how AI can help you save time, money, and stress. We’ve used that knowledge to build an assistant that goes on your trip with you. Take a look at what it can do:

61위Jerry Wu@Jerr_WuX에서 보기

제리 우: Halluminate가 Oak HC/FT 주도, YC 등이 참여한 3,000만 달러 시리즈 A를 유치했다고 발표했습니다.

좋아요 191리포스트 21답글 43노출 36.5K북마크 67

62위a16z@a16zX에서 보기

a16z: 올해 AI 기업 두 곳이 늘린 매출이 상장 소프트웨어 기업 전체의 증가분보다 많다는 차트를 공유했습니다.

좋아요 168리포스트 20답글 14노출 20.7K북마크 57

a16z@a16z
Two AI companies have added more revenue this year than all of public software combined More charts in State of Markets II: a16z.news/p/state-of-mar…
David George@DavidGeorge83
Introducing our State of Markets pt 2, along with a companion podcast where we unpack the data and discuss what comes next. Tech is the everything cycle. Supply: putting the buildout in context, just passed railroads as % of GDP. The wisdom of Elon is real: the factory (or the datacenter!) is the product. Demand: diffusion is so, so early. Median AI vendor spending in the top 1% of companies is 8x that of the top 10%. Only about 30% of S&P 500 companies report a quantified AI impact, which means there’s a substantial opportunity in connecting models to a company's data and workflows. Diffusion into companies is one of the main themes of the next 5 years. We’re entering the agent work period. Only a few million users today, but applicable to billions of internet users with massive surplus created. META/GOOG monetize US users at $200+ per year today. Agent opportunity is much higher. Mega-trends the next 5 years: Consumer agents, Robotics, Autonomy, AI x bio, Personal health, Diffusion into enterprise, New era of American Dynamism. Much more in our SoM report here - a16z.news/p/state-of-mar… @a16z @sarahdingwang @aleximm @santiago__rdz

63위a16z@a16zX에서 보기

a16z가 사람과 에이전트를 위한 사설 암호화 병렬 네트워크 doxxnet의 시리즈 A를 주도했다고 밝혔습니다.

좋아요 94리포스트 11답글 15노출 19.8K북마크 31

a16z@a16z
We're excited to lead doxxnet's Series A. AI has made it dramatically easier to collect information about people, and agents acting on our behalf expand the surface area we need to protect. @DoxxnetHQ gives people and their agents a parallel, private network they can build on: browse privately, transfer files directly, connect devices and agents into encrypted meshes, and communicate peer to peer. @BarrettLyon helped pioneer the modern DDoS defense industry with Prolexic, and has spent more than two decades building networks and defending them against attackers. Welcome Barrett Lyon and the doxxnet team! By @YODELAGARZA and @martin_casado
Barrett Lyon@BarrettLyon
Today we’re launching doxx.net, enabling Agentic Defined Networking for everyone. We raised $38M from @a16z, @AnimoVentures and Focal.vc. Create private networks for you and your AI agents, without a server in the middle. 38M+ threats blocked since December. Our first private-network call was a freedom call. Now it’s your turn.

65위Victor M@victormustarX에서 보기

빅토르 무스타르: 로봇이 향후 10년을 지배할 것이라며 Hugging Face Hub의 커뮤니티 정책을 Microduck 로봇에 바로 보내는 등 로보틱스 사용을 쉽게 만들고 있다고 밝혔습니다.

좋아요 67리포스트 10답글 13노출 8.8K북마크 19

Victor M@victormustar
Robots are going to eat the next 10 years 💯 We're making robotics on the Hugging Face Hub easier to use, eg: soon you'll be able to send any community-made policy to your Microduck. Can you just imagine the fun we're going to have?

66위Max Junestrand@MaxJunestrandX에서 보기

막스 주네스트란드: 블룸버그 인터뷰에서 Legora가 3분기 신규 고객 800곳을 늘렸다고 전하며, AI 일자리 질문에는 고객사들이 오히려 변호사를 더 뽑고 있다고 답했다고 밝혔습니다.

좋아요 72리포스트 14답글 5노출 9.4K북마크 19

Max Junestrand@MaxJunestrand
I was in the Bloomberg studio in London this morning. Q3 was the most hectic quarter in @WeareLegora's history. We added 800 new customers in three months and now work with more than 2,300 across 80+ countries. We also added more ARR this quarter than our total ARR at the end of 2025. I was asked the jobs question, and my answer hasn't changed: our customers are hiring more lawyers. Legal is supply constrained. If you could review every contract, you would. If you could run full diligence on every asset before it goes to investment committee, you would. Better tools mean more of that work gets done, and a lawyer still owns the judgment. We talked about junior lawyers wondering whether they need to learn to code. With large language models, the programming language is English. I'm not especially good with words. Lawyers are, and they turn out to be some of the best drivers of our product. I've said it before, and I'll say it again: I think this is the best time in history to be a lawyer. Thanks @TomMackenzieTV for having me on The Opening Trade. Full interview below.

67위Google DeepMind@GoogleDeepMindX에서 보기

Google DeepMind가 AI 생성물이 실물과 구분하기 어려워진 시대에 출처 증명과 SynthID·SynthID Bio 워터마킹의 역할을 다룬 팟캐스트를 공개했습니다.

좋아요 308리포스트 24답글 61노출 34.2K북마크 80

Google DeepMind@GoogleDeepMind
When AI-generated content has become so realistic, how do we verify what's made by human, machine, or nature? We explore the vital role of provenance – and how imperceptible watermarking can safeguard both digital media and biological designs – on our podcast. ↓ Timecodes: 00:00 Introduction 00:34 What is a watermark? 04:35 SynthID 15:16 Images and video 19:28 SynthID Bio 28:00 Future of resilience

68위Jeff Boudier 🤗@jeffboudierX에서 보기

제프 부디에: Microduck 생산 확대를 기념해 초대형 'MEGADUCK'을 10월 16일 샌프란시스코 Open Together 행사에서 공개한다고 예고했습니다.

좋아요 119리포스트 19답글 17노출 15.4K북마크 24

69위Powerus@powerus_usaX에서 보기

[광고] Powerus가 Aureus Greenway Holdings와 합병을 마치고 나스닥($PUSA)에 상장했다며 고위험 환경용 자율 항공·해양 시스템 기업임을 알렸습니다.

좋아요 61리포스트 15답글 6노출 18.9K북마크 12

Powerus@powerus_usa
Powerus is now a public company, trading on Nasdaq under the symbol $PUSA. We completed our merger with Aureus Greenway Holdings on today, and the combined company is now Powerus Corporation. We build autonomous air and maritime systems for people operating in high-risk environments, and that work is the same today as it was yesterday. “I spent most of my career as the person relying on this equipment in the field. That is the standard we build to, and it does not change because the company has a listing.” Brett Velicovich, Co-Founder of Powerus Full release linked in the comments.

70위a16z@a16zX에서 보기

a16z: doxxnet 창업자 배럿 라이언이 데이터센터가 일자리·인터넷 속도에 해롭다는 통념은 오해이며 사람들이 조작되고 있다고 말한 대담 발췌입니다.

좋아요 52리포스트 8답글 8노출 16.2K북마크 10

a16z@a16z
.@DoxxnetHQ founder Barrett Lyon on data center myths versus reality: jobs, talent, and faster internet. "That's the other thing that's driving me crazy, is people don't understand data centers at all... Data centers have been around for a very, very long time." "People are being manipulated with this data center thing... It's very interesting to be in the know and then watch this and go, oh my goodness, these people are missing the mark." "The big data center park in Virginia, you just see bunny rabbits in the field. There's nothing. It's just a big building and a bunch of parking lot and then nothing." "It brings jobs. The people that work in them aren't getting paid pennies. And it also brings high tech talent. And then also infrastructure, and infrastructure powers everything you do. So if you have a data center nearby, your internet's faster. Things like that work better." @BarrettLyon @YODELAGARZA
a16z@a16z
.@BarrettLyon built the first cloud DDoS-protection service, mapped the entire internet (now on display in New York's Museum of Modern Art), and got chased out of Moscow hunting Russian extortionists. His new company is doxx.net, a parallel internet designed to keep advertisers, trackers, and creepy VPNs out of your life. Barrett joins a16z's Joel de la Garza to explain what it is and how it works: a private network where messages, calls, and files travel directly between people instead of through someone else's server, and where nobody needs your name, number, or email to reach you. They also get into why most VPNs are "just entry and exit points," why you can't set policy on equipment you don't own, the day Barrett's plane lost power over Oakland on the way home from a16z, and the data center myths that drive him crazy. 0:50 Meet Barrett Lyon 1:31 Losing an engine over Oakland 7:08 The hacker name behind the company 10:12 Censorship is coming to the US 13:17 Why VPNs barely protect you 14:21 Building a parallel internet 16:30 20GB files with no middleman 17:57 Free VPNs sell you 21:47 Why you must own the hardware 22:27 26 data centers, 12 people, run by AI 25:03 Data centers don't drink water 29:17 What ad trackers know about you 32:35 The DDoS attacks that built Prolexic 36:17 Mapping the internet for $50 39:46 Hunting extortionists in Moscow 41:47 Starlink at Burning Man YouTube: youtu.be/bmDrbHOh7Bo @BarrettLyon @YODELAGARZA

71위John Carmack@ID_AA_CarmackX에서 보기

존 카맥: Quest 3의 VR 핀볼이 실물 기기보다 더 재밌었지만 Unreal로 옮긴 신작 Pinball FX는 프레임이 떨어지고 구버전 Classic이 더 안정적이라며 최적화·에일리어싱 문제를 짚었습니다.

좋아요 376리포스트 22답글 46노출 69K북마크 71

John Carmack@ID_AA_Carmack
After playing some pinball on our Polycade cabinet, Trista and I revisited VR pinball on our Quest 3, and we both thought it was a meaningfully better and more fun experience in VR. There is an interesting wrinkle: @PinballFX used an efficient custom engine for a long time, but they switched over to Unreal Engine for their current games to more easily support the wide range of platforms they deploy on. Pinball FX uses Unreal, but they left the older version available as PinballFX Classic, which makes for a nice comparison. After downloading the 40x larger new application, you find that it has tons of customization features and environmental effects, but most tables drop frames even on Quest 3. The Classic version is locked at an unfortunately low resolution, but it feels more solid when actually playing pinball. Quest Game Optimizer can adjust resolution and frame rate if you are willing to go through the Side Quest dance, making a fairly dramatic impact on the visuals. I just gave Meta a nudge about maybe allowing QGO into the official store, but ideally the developers would add a few lines of cold to the old application to default to different settings for the different Quest headsets. Touching an old application is a bigger ask for a developer than most users appreciate, though. Both the old and new versions are kind of a mess with preventable visual aliasing. Here is a post I wrote ten years ago about aliasing for GearVR developers, long before Quest existed: sRGB framebuffers and sRGB textures with mip maps, MSAA, go light on specular shading effects. < Dr Manhattan it’s 1985 meme >
facebook.comJohn Carmack

72위Visual Studio Code@codeX에서 보기

VS Code 9월 릴리스에 여러 모델을 조율하는 HydraFusion, 리뷰·실패 체크·머지 충돌을 처리하는 Agent Merge, 에이전트 창 예약 자동화 등이 추가됐습니다.

좋아요 69리포스트 4답글 11노출 8.6K북마크 13

Visual Studio Code@code
🚀 September’s @code releases brought new ways to build, automate, and ship! Here are some highlights: 🧠 HydraFusion orchestrates multiple models for your coding tasks. 🔀 Agent Merge handles review feedback, failed checks, and merge conflicts until your pull request is ready to merge. ⏰ Automations in the Agents window run repeatable tasks on demand or on a schedule. ✨ Plus, explore new ways to work across repositories and manage agent sessions. 🔗 Explore the latest updates: Happy coding! 💙
github.blogGitHub Copilot in VS Code, September 2026 releases - GitHub Changelog

73위Marty Kausas@marty_kausasX에서 보기

마티 카우사스: Pylon이 하루 만에 160만 달러 계약을 성사시켜 창사 이래 최고 매출일을 기록했다며 팀에 공을 돌렸습니다.

좋아요 77리포스트 4답글 15노출 7.1K북마크 14

Marty Kausas@marty_kausas
Yesterday we closed $1.6M in ONE DAY. This is by far the highest revenue day in Pylon history and everyone played a massive part in making that happen. I am so so proud of the Pylon team. Recruiting → Built the team Product, Design, Engineering → Built the product RevOps → Built the GTM plan Marketing → Created pipeline Solutions → Helped prospects migrate, implement, and customize Pylon Legal → Negotiated fair contracts Finance → Structured the commercial terms Data → Built our AI pricing model Support → Helped customers use the product Sales → Ran and closed deals Customer Success → Helped customers grow with us Operations → Kept the business running Such a strong team effort. I'm so proud of you all.

74위Uncle Bob Martin@unclebobmartinX에서 보기

엉클 밥: 하네스 없이 몇 개 에이전트에 격리 작업을 맡기고 CRAP·DRY·뮤테이션을 수동 지시하던 3월 방식으로 돌아갔지만 작업 단위는 10배 이상 커졌다고 했습니다.

좋아요 398리포스트 5답글 36노출 26.1K북마크 67

Uncle Bob Martin@unclebobmartin
I'm back to working the way I did in March. A few agents open doing isolated tasks. Manually prompting them to do crap, dry, and mutate. No harness. The one big difference is that the size of the individual tasks has increased by an order of magnitude or more.

75위Huawei@HuaweiX에서 보기

[광고] 화웨이가 세계 최초 LogicFolding 구조의 기린 9050 Pro 등 τ 칩을 전 모델에 탑재한 Mate90 시리즈 출시 행사를 홍보했습니다.

좋아요 432리포스트 57답글 25노출 5.7M북마크 65

77위Google Cloud@googlecloudX에서 보기

Google Cloud가 10월 8일 오전 10시 30분(PT) 토머스 쿠리안이 출연하는 디지털 행사 'Gemini at Work' 생중계를 예고했습니다.

좋아요 102리포스트 11답글 0노출 236.6K북마크 12

78위Kenneth Roth@KenRothX에서 보기

[정치] 케네스 로스: 프랑스 좌파 정치인 장뤼크 멜랑숑의 친푸틴 성향과 우크라이나 침공에 대한 무관심이 알려지면 지지가 줄 것이라고 주장했습니다.

좋아요 155리포스트 36답글 36노출 32.8K북마크 18

79위Sam Altman@samaX에서 보기

샘 올트먼: 'Sign In With ChatGPT'와 플러그인 확장에 사람들이 생각하는 것보다 훨씬 큰 잠재력이 있다고 말했습니다.

좋아요 1.1K리포스트 41답글 144노출 100.6K북마크 124

80위art@artfreebreyX에서 보기

Revnu가 PXN·YC 등으로부터 300만 달러를 유치해 GTM 스택 자동화를 만든다고 밝혔습니다.

좋아요 72리포스트 5답글 18노출 7.7K북마크 8

81위NVIDIA@nvidiaX에서 보기

NVIDIA가 모든 종류·단계의 AI와 비AI 작업까지 돌릴 수 있는 '대체 가능한(fungible)' AI 인프라의 의미를 설명하는 블로그를 공유했습니다.

좋아요 163리포스트 25답글 53노출 45K북마크 16

NVIDIA@nvidia
What does “fungible” mean for AI infrastructure? The flexibility to run different workloads: every type of AI, every phase of AI and even workloads that aren’t AI at all. One platform, built for maximum “productivity”, “durable” years after deployment, and “fungible” for every model and workload. Read the blog to learn more: nvda.ws/4hphGmQ

82위Quentin Gallouédec@QGallouedecX에서 보기

캉탱 갈루에덱: 텔레메트리상 TRL로 월 100만 건의 학습이 이뤄지고 있으며 다음 목표는 주 100만 건이라고 밝혔습니다.

좋아요 48리포스트 13답글 7노출 11.4K북마크 4

83위Y Combinator@ycombinatorX에서 보기

YC가 Juni Learning·Acely 창업자 비비안 셴과 Cron(현 Notion Calendar) 창업자 라파엘 샤드를 새 제너럴 파트너로 맞았습니다.

좋아요 242리포스트 18답글 28노출 63K북마크 20

Y Combinator@ycombinator
We’re excited to welcome @vivianmshen and @raphaelschaad as YC’s newest General Partners! Vivian and Raphael first came to YC as founders. Vivian co-founded Juni Learning (W18), which teaches kids coding and math, and later founded Acely, an AI tutor for college prep. Raphael founded Cron (W20), the calendar app that became Notion Calendar. They built products used by millions of people, then returned to YC as Visiting Partners to help other founders. Now they're doing it full-time.
ycombinator.comVivian Midha Shen and Raphael Schaad Join YC as General Partners | Y Combinator

84위Google@GoogleX에서 보기

구글이 Planet과 함께 TPU 4개를 실은 시제품 위성을 SpaceX Transporter-18로 발사해, 우주 ML 인프라 가능성을 탐구하는 Project Suncatcher의 첫 단계를 시작했습니다.

좋아요 214리포스트 31답글 34노출 28.8K북마크 16

Google@Google
Up, up, and away. 🚀 Today, in partnership with @planet, we launched a prototype satellite carrying four TPUs into orbit on @SpaceX's Transporter-18 rideshare mission. This launch is the first step of Project Suncatcher, our long-term research moonshot to see whether we can one day host scalable machine learning infrastructure in space. Over the coming weeks, we'll gather in-orbit data on how our TPUs handle the physical stress, radiation, and thermal extremes of space. Whatever we learn, we'll use it to refine our future designs.

85위Bill Gates@BillGatesX에서 보기

빌 게이츠: 켄 그리핀의 30억 달러 이상 카네기멜런대 기부(CMU 마이애미 설립 포함)가 크고 지속적인 영향을 줄 것이라고 평했습니다.

좋아요 568리포스트 47답글 0노출 275K북마크 36

86위Garry Tan@garrytanX에서 보기

개리 탄: YC 파트너가 자신 포함 19명으로 역대 최대가 됐고 모두 YC 출신 창업자라며 비비안·라파엘을 환영했습니다.

좋아요 215리포스트 7답글 15노출 25.1K북마크 11

Garry Tan@garrytan
YC has the largest partnership in history now - 19 including me. We all went through YC and are alums and founders ourselves. I'm so proud to welcome Vivian and Raphael to the partnership.
Y Combinator@ycombinator
We’re excited to welcome @vivianmshen and @raphaelschaad as YC’s newest General Partners! Vivian and Raphael first came to YC as founders. Vivian co-founded Juni Learning (W18), which teaches kids coding and math, and later founded Acely, an AI tutor for college prep. Raphael founded Cron (W20), the calendar app that became Notion Calendar. They built products used by millions of people, then returned to YC as Visiting Partners to help other founders. Now they're doing it full-time. ycombinator.com/blog/welcome-v…

87위Daniel Jeffries@Dan_Jeffries1X에서 보기

댄 제프리스: AI 안전을 둘러싼 싸움은 없고 친지능 대 반지능 세력의 싸움일 뿐이며 '안전'은 AI를 멈추려는 가짜 구호라고 주장했습니다.

좋아요 217리포스트 42답글 42노출 18.5K북마크 11

Daniel Jeffries@Dan_Jeffries1
There's no battle for AI safety. It's a battle between pro-intelligence and anti-intelligence forces. Safety is a fake stand-in word for pausing/stopping/strangling AI. Nobody is anti-safety. Safety is a fake set-up word to booby trap your pro-intelligence stance.

88위NASA@NASAX에서 보기

NASA: Crew-13 우주비행사 4명이 연구·기술 시연·정비를 위해 ISS로 발사됐다고 알렸습니다.

좋아요 362리포스트 61답글 8노출 85.5K북마크 11

90위NASA HQ PHOTO@nasahqphotoX에서 보기

NASA 사진팀이 Crew-13 승무원들이 드래곤 탑승을 위해 발사대 LC-40으로 출발하는 사진을 공유했습니다.

좋아요 639리포스트 84답글 29노출 176.8K북마크 17

91위Vivian Midha Shen@vivianmshenX에서 보기

비비안 셴: YC 'Startup Industrial Base' 컨퍼런스에서 네이비실 출신 창업자이자 상원의원인 팀 시히와 의회의 창업자 지원 방안을 대담했다고 전했습니다.

좋아요 79리포스트 6답글 9노출 11.6K북마크 2

92위DHH@dhhX에서 보기

[잡담/주제외] DHH: 이번 주말 페티트 르망(로드 애틀랜타)에서 토비·마티아스 베셰와 쇼피파이 레이싱 #11 LMP2로 올해 마지막 포디움에 도전한다고 알렸습니다.

좋아요 164리포스트 2답글 23노출 14.7K북마크 4

93위NASA@NASAX에서 보기

NASA: Crew-13 사령관 와킨스가 '유인 우주비행은 팀 스포츠'라는 발사 전 연설을 한 뒤 승무원들이 주먹 인사를 나눴다고 전했습니다.

좋아요 288리포스트 22답글 13노출 85.3K북마크 7

NASA@NASA

@NASA 님에게 보내는 답글

“Human spaceflight is truly a team sport, and it is an honor and a privilege to represent all of you who have contributed to the success of this mission.” The Crew-13 crew members can be seen exchanging fist bumps after Commander @Astro_Watkins gives a speech prior to liftoff.

94위NASA@NASAX에서 보기

NASA: 드래곤이 발사 8시간이 채 안 돼 ISS에 도킹할 예정이라 미국 우주선 역대 최단 ISS 비행이 된다고 밝혔습니다.

좋아요 162리포스트 17답글 17노출 48.6K북마크 3

95위Perplexity Developers@perplexitydevsX에서 보기

Perplexity가 Qwen3.8-27B를 25만 토큰 컨텍스트로 미세조정한 의사결정 모델 pplx-decider-v1-27b 가중치를 Hugging Face에 공개하고 Decisions API를 열었습니다.

좋아요 55리포스트 5답글 3노출 5K북마크 52

96위Wauplin@WauplinX에서 보기

huggingface_hub v2.1.0이 나와 Jobs 재시도·재예약, ZeroGPU 할당량 추적, hf_xet으로 최대 17배 빠른 다운로드, 레시피 기반 Inference Endpoints 배포를 지원합니다.

좋아요 12리포스트 3답글 0노출 4.1K북마크 9

97위Google@GoogleX에서 보기

구글이 Project Suncatcher의 우주 하드웨어 생존 테스트·AI 칩 냉각 설계를 다룬 블로그 링크를 덧붙인 후속 게시물입니다.

좋아요 9리포스트 2답글 5노출 6K북마크 5

98위Oren Melamed@OrenMeX에서 보기

오렌 멜라메드: VS Code가 사람과 에이전트가 같은 대시보드·칸반을 함께 다루는 Copilot Canvases를 에이전트 경험에 도입했으며, /create-canvas로 직접 만들 수 있다고 소개했습니다.

좋아요 40리포스트 7답글 2노출 6.6K북마크 22

Oren Melamed@OrenMe
VS @code is bringing Copilot Canvases into the Agent experience. The concept is similar to Canvases in the GitHub Copilot app: a shared interactive surface where both the human and the agent can work on the same artifact. Think dashboards, Kanban boards, plans, triage boards, etc. But there is an interesting difference under the hood. VS Code isn't waiting for Canvases to become part of the canonical Agent Host Protocol. It is shipping a VS Code-specific transport first, which lets them experiment with the primitive without committing the whole AHP ecosystem to the API yet. And you can actually create one yourself. Start an agent session and run: /create-canvas Then describe what you want, for example: `Create an agentic Kanban canvas where I can create, assign and move cards, and where you can also update the board.` The agent builds the canvas and opens it in the Sessions UI. You can then keep iterating on the UI, state and capabilities from the agent. For a team canvas, it can live under .github/extensions. Personal canvases can live under ~/.copilot/extensions.

99위Zach Lloyd@zachlloydtweetsX에서 보기

잭 로이드: 10월 8일 우버·Faire 리더들과 소프트웨어 팩토리 도입 단계, 스택 설계, 직접 구축 여부를 비교하는 대담을 연다고 알렸습니다.

좋아요 15리포스트 3답글 3노출 14.4K북마크 8

Zach Lloyd@zachlloydtweets
On October 8th, I'm sitting down with @hudaman (Sr Engineering Manager, Uber) and @cmhandel (Director of Product, AI Platform, Faire) to compare notes on adopting software factories: - What each phase looks like in practice (crawl vs walk vs run), and how to tell which one you're in - How to architect the factory stack: runners, environments, scorers, benchmarks, quality gates - When to build vs. partner - The metrics to measure: human touchpoints per change, cost per unit of software shipped, governance signals - The cultural changes to adopt everything Link in the comments if you want to register for this roundtable.

100위SPAZE@SpazeWorldzX에서 보기

[광고] 인디 게임 'SOMETHING WEIRD WILL HAPPEN' 출시를 알리며 플레이를 권하는 게임 홍보 게시물입니다.

좋아요 14리포스트 0답글 0노출 29.2K북마크 6

101위Suchintan Singh@SuchintanX에서 보기

처음부터 다시 만든 오픈소스 브라우저 에이전트 Skyvern 3.0이 Odysseys 벤치마크에서 90.5%로 1위를 기록했습니다.

좋아요 23리포스트 11답글 3노출 2.6K북마크 9

102위Paulina Xu@paulinazhxuX에서 보기

Fabriq Developer가 출시돼 프롬프트 하나로 Slack·Gmail·GitHub 등 500개 이상 통합을 AI 앱에 붙이고 OAuth·토큰 갱신·감사 로그를 대신 처리해 줍니다.

좋아요 20리포스트 5답글 14노출 2.9K북마크 7

Paulina Xu@paulinazhxu
Today, we're launching Fabriq Developer. Over 500 integrations for your AI app, added in one prompt. Slack, Gmail, HubSpot, Linear, GitHub, Salesforce... your users connect their own accounts, your agent acts through their connection with only the scopes they approved, and we handle the OAuth, tokens, refresh, revocation and audit log. No OAuth code on your side. Free to start.

103위TypeSafe AI@typesafeaiX에서 보기

Typesafe AI: Wrangle이 Jev를 리랭커로 써 기존 비용의 75%로 채용 검색 매칭을 30.64% 개선했다는 사례를 인용했습니다.

좋아요 55리포스트 4답글 3노출 9.6K북마크 16

TypeSafe AI@typesafeai
Jev is a phenomenal reranker, you heard that right.
John Crown@johncrown
Jev is a phenomenal reranker. At 75% of the previous reranking cost, it improved our main-app search by 30.64% in matches per 200 candidates on searches recruiters actually run. Most recruiting benchmarks measure precision and accuracy across just 10 candidates. That isn’t representative of real recruiting and sourcing workflows, where you may need to source hundreds of candidates to make a hire. We’ll release the full benchmark soon. We're excited to work with @hmartenjoyer, @mathfax , and the rest of the @typesafeai team to continue to bring Jev to more Wrangle features. Stay tuned 👀

104위ALIENWARE@AlienwareX에서 보기

[광고] Alienware가 NVIDIA GeForce와 함께 스트리머 게임 대회와 게이밍 장비를 홍보했습니다.

좋아요 12리포스트 0답글 1노출 360.9K북마크 3

105위a16z@a16zX에서 보기

a16z 데이비드 조지: AI는 돈을 쏟을수록 확장·개선된다는 점이 이전 사이클과 다르며, 지금 제품 사이클은 10점 만점에 9~10점, 자본 사이클은 6점 수준이라고 평가했습니다.

좋아요 38리포스트 2답글 9노출 8.7K북마크 9

a16z@a16z
a16z's David George on product cycles versus capital cycles, and his scorecard for the last 15 years of tech: "AI is unique because if you throw more money at it, it can just scale up and get better. That is different than previous cycles." "If you threw endless amounts of money at ServiceNow or Workday or Salesforce during the SaaS era, they would get all messed up... Whereas if you put $40 billion towards training these models, they get much better." "I think about the world as product cycles and capital cycles... In 2021, we were at like a 1 out of 10. We didn't realize it at the time. That was late SaaS, late cloud, post-mobile phone growth... There was no room for something new." "In 2010 it was like an 8 out of 10. Right now product cycle is like a 9 or 10." "Your ideal would be a 10 on product cycle and a 10 on capital cycle, meaning valuations are really low. They almost never coincide." "Like 2023, early days of when AI started working... capital cycle was like a 7... Right now we're probably at a 6. It's okay, but not great." "But the thing that drives our business over 10 years, whether it's venture or growth, is the product cycles. And that's like a 9 or 10 right now." @DavidGeorge83 w/ @jaltma

106위TypeSafe AI@typesafeaiX에서 보기

Typesafe AI가 'AI 봇이 말을 안 듣는 건 아직 Jev가 없어서'라며 Jev를 가볍게 홍보했습니다.

좋아요 9리포스트 0답글 2노출 1.7K북마크 2

TypeSafe AI@typesafeai
🗣️ Ever felt like AI bots don't really listen? 🗣️ They just don't have a li'l Jev helping them yet!
Jay F@SoCalJayF
One really cool use case I’ve been exploring with Jev: a semantic VAD for real-time voice AI. I plugged it into @AgoraIO ConvoAI and used the live transcript + recent conversation as context to decide whether the user has actually finished speaking. That makes Jev surprisingly good at handling things like hesitation, unfinished thoughts, and those moments where you stop talking for a second because you’re still figuring out what you want to say. It’s a nice example of where Jev fits really well in a real-time AI loop. Here’s what it looks like in action 👇

107위Gunin Gupta@GuninGupta907X에서 보기

원본 영상부터 최종 편집까지 실제 영상 편집 업무로 에이전트를 평가하는 TimelineBench가 공개됐으며, 16개 에이전트 중 최고 성적도 56개 과제의 26.8%에 그쳤습니다.

좋아요 10리포스트 7답글 3노출 2.4K북마크 2

108위Jon Xu@xusterX에서 보기

YC 파트너 xuster: Halluminate가 YC S25 배치에서 가장 빠르게 성장한 회사 중 하나로 프런티어 모델 학습의 핵심 인프라를 정의하고 있다며 시리즈 A를 축하했습니다.

좋아요 28리포스트 1답글 0노출 4.9K북마크 5

Jon Xu@xuster

@Jerr_Wu 님에게 보내는 답글

@Jerr_Wu and @wgm752 have built one of the fastest growing companies out of YC Summer 25 batch.@HalluminateAI is defining the core training infrastructure for how frontier models get smarter. Very excited for what's coming next from them. Congrats on the Series A! 🚀
Jerry Wu@Jerr_Wu
Today we're excited to announce @HalluminateAI's $30M Series A led by @oakhcft with participation from @YCombinator, @orangecollectv, @FTPartners, @heavybit, and more.

109위Google DeepMind@GoogleDeepMindX에서 보기

Google DeepMind가 워터마킹 팟캐스트를 YouTube·Spotify·Apple Podcasts에서 들을 수 있는 링크를 덧붙였습니다.

좋아요 18리포스트 2답글 4노출 15.2K북마크 3

110위Uncle Bob Martin@unclebobmartinX에서 보기

엉클 밥: '합리적인 범위에서' 하라고 지시하자 에이전트가 살아남은 뮤턴트를 없앨 방법들을 저울질하는 모습이 재밌다고 했습니다.

좋아요 35리포스트 0답글 9노출 7.9K북마크 5

111위Google DeepMind@GoogleDeepMindX에서 보기

Google DeepMind가 생물 설계가 안전장치를 갖춘 AI로 만들어졌는지 빠르게 인증하는 SynthID Bio 도구를 연구용으로 공개한다고 밝혔습니다.

좋아요 49리포스트 2답글 5노출 13.8K북마크 7

113위Google Cloud@googlecloudX에서 보기

Google Cloud: 페이팔의 니르 실라지가 AI 시대의 10배 엔지니어는 산출량이 아니라 AI로 역량이 증폭된 '시스템 사고자'라고 설명한 인터뷰입니다.

좋아요 24리포스트 3답글 3노출 3.8K북마크 3

114위David George@DavidGeorge83X에서 보기

데이비드 조지(a16z): 올여름 프런티어 AI 연구소 사이버 사고 이후 보안 압박이 커졌다며, 맨디언트 창업자 케빈 맨디아의 AI 자율 보안 플랫폼 Armadin Security 시리즈 B를 공동 주도했다고 밝혔습니다.

좋아요 42리포스트 4답글 2노출 7.4K북마크 5

David George@DavidGeorge83
We're excited to co-lead @ArmadinSecurity's Series B. After this summer’s cyber incidents at frontier AI labs, CISOs face growing pressure to secure their companies across applications, cloud, identity, and internal networks. Armadin’s AI autonomous security platform turns offense into something defenders control, from the first Hyperattack. Built on decades of frontline adversary tradecraft, its agentic attacker swarm reasons, plans, and adapts like a real adversary. The platform learns from every attack and environment, allowing security teams to focus on fixing real problems. We believe there’s no better team to take this on than Kevin Mandia, Travis Lanham, Evan Peña, and David Slater. Kevin founded Mandiant, the firm that defined modern incident response. Kevin and the team have spent their careers staying a step ahead of attackers, and now they’re building the platform to help every security team do the same. @a16z @steph_zhang

115위Randall Balestriero@randall_balestrX에서 보기

랜들 발레스트리에로: MICCAI-FOMO26 기조연설에서 JEPA 최신 진전과 데이터 불균형·노이즈·멀티모달·추론을 다룰 원칙적 해법의 필요성을 발표한다고 알렸습니다.

좋아요 45리포스트 4답글 1노출 8.8K북마크 4

116위Ashi Dissanayake@Ashi_SpaceiumX에서 보기

Spaceium 측이 YC 워싱턴 행사에서 NASA 국장 재러드 아이작먼과 신규 우주 역량의 첫 고객 확보와 궤도 급유 상용화를 논의했다고 전했습니다.

좋아요 14리포스트 4답글 0노출 6.2K북마크 1

117위Google Cloud@googlecloudX에서 보기

Google Cloud: 구글이 2026 가트너 컨테이너 관리 매직 쿼드런트에서 리더로 선정되고 2년 연속 실행 능력 최고 평가를 받았다고 밝혔습니다.

좋아요 21리포스트 4답글 1노출 3.7K북마크 1

118위Jon Xu@xusterX에서 보기

xuster: 시리즈 A를 발표한 Halluminate 공동창업자들과 사후학습의 미래, 회사 전체 시뮬레이션, 학습 환경에서 시작되는 정렬·안전을 논의했다고 전했습니다.

좋아요 22리포스트 0답글 3노출 4.4K북마크 1

Jon Xu@xuster
Fresh off their Series A announcement, @Jerr_Wu and @wgm752 from @HalluminateAI joined me to discuss the future of post-training, simulating entire companies, and why alignment and safety start with the environments models learn in. Really enjoyed their perspective from building at the frontier of model training.
Y Combinator@ycombinator
Halluminate is building reinforcement learning environments and benchmarks to help AI models do knowledge work beyond coding, starting with finance. With a team of fewer than 10 people, the company works with four of the top five closed-source US AI labs and recently raised a $30M Series A. In this fireside, co-founders @Jerr_Wu and @wgm752 sit down with YC's @xuster to share how testing browser agents led them to building the environments models now use to practice tasks like financial modeling. They explain why better verification, not just more training tasks, is critical to making models smarter, and how that work could eventually lead to simulated companies where teams of AI agents learn to work together. 00:00 — Halluminate’s Pivot From Evals to Training Environments 03:48 — Teaching AI to Do Financial Work 07:13 — Why Verification Matters More Than Volume 10:13 — From Finance to Simulated Companies 13:30 — AI Safety and Building the Team

119위Restoration News@NewsRestorationX에서 보기

[정치] 미시간주가 수십억 달러 규모 메디케이드 청구에 대한 사기 조사를 제대로 하지 않았다고 비판하는 정치 기사 홍보 게시물입니다.

좋아요 25리포스트 5답글 1노출 14.1K북마크 1

Restoration News@NewsRestoration
Latest bombshell from Senior Fellow @wcdispatch Michigan Fraud Investigation Unit Waved Through Billions in Questionable Medicaid Claims The oversight required by law isn't happening. On September 29, 2026, Dr. Mehmet Oz, the administrator of the Centers for Medicare & Medicaid Services (CMS) gave Michigan and seven other states 30 days to explain why they reported zero Medicaid fraud referrals this year. Federal inspectors explained it in 2022: four referrals in three years, 31 people to police a $21 billion program, and an attorney general who wouldn't ask for more. Why isn't Michigan investigating this?? Read the full piece:
restoration-news.comMichigan Fraud Investigation Unit Waved Through Billions in Questionable Medicaid Claims

120위Google DeepMind@GoogleDeepMindX에서 보기

Google DeepMind: SynthID Bio 워터마크가 디지털 신분증처럼 작동해 DNA 연구소가 합성 단백질을 검증하고 서열 데이터베이스 정확성과 생물보안을 강화할 수 있다고 설명했습니다.

좋아요 70리포스트 3답글 14노출 18.7K북마크 2

Google DeepMind@GoogleDeepMind

@GoogleDeepMind 님에게 보내는 답글

Acting like a digital ID card, the watermark can help DNA labs verify synthetic proteins and keep sequence databases accurate. As these requests increasingly scale, verifying if designs come from AI with built-in safeguards will help labs automate and strengthen biosecurity. 🛡️

121위DHH@dhhX에서 보기

[잡담/주제외] DHH: 아우토반이 독일의 다른 도로보다 오히려 안전해 속도 대 안전의 문제가 아니라며 과잉보호 문화에 대한 비판으로 이어 갔습니다.

좋아요 70리포스트 0답글 6노출 6.4K북마크 2

DHH@dhh

@dhh 님에게 보내는 답글

Also, Autobahn is safer than the rest of German roads! So the trade off isn't even speed versus safety. Odd point to make too given that this is an example of how nobody died. But it connects directly to schools wrapping trees in bubble wrap to protect the children.

122위NVIDIA@nvidiaX에서 보기

NVIDIA: OpenAI GPT-6 Astra Ultrafast가 NVIDIA 위에서 최대 8배 빠르게 돌아가는 이유를 다룬 기사를 공유했습니다.

좋아요 39리포스트 2답글 7노출 6.2K북마크 1

123위BrendanEich@BrendanEichX에서 보기

브렌던 아이크: Brave가 9월 MAU 1억 2,828만(+1.6%), 28일 평균 DAU 5,297만(+3.8%), 월 검색 26억 3천만 건을 기록했다고 밝혔습니다.

좋아요 66리포스트 5답글 3노출 4.7K북마크 0

BrendanEich@BrendanEich

@brave 님에게 보내는 답글

@Brave reached 128.28M MAU in September, up 1.6% from August (early summer slump rebound). 28-day avg. DAU 52.97M, up 3.8%, best DAU growth in 2026. Peak DAU 54.60M, record 7.04B minutes of daily use. Search > 2.63B QPM, 31B annualized (+4.9% MoM). Record 88.48M QPD 9/28. 1/4
BrendanEich@BrendanEich
@Brave reached 126.25M MAU, up 3% in August — our highest MAU growth rate in 2026. Peak DAU neared 53M, up 2.4%. 6.78B minutes of daily use. Desktop up, summer slump over. Brave Search passed 2.5B monthly queries, nearly 30B annualized, hit record 85.58M queries on 8/31. 1/4

124위Department of War - CFO@DoWCFOX에서 보기

[정치] 미 전쟁부 차관 줄스 허스트가 YC 행사에서 예산 투명성과 집행 데이터로 검증된 역량을 더 빨리 전력화해야 한다고 강조했다는 국방부 게시물입니다.

좋아요 44리포스트 15답글 1노출 7.6K북마크 0

Department of War - CFO@DoWCFO
Under Secretary of War Jules Hurst appreciated the opportunity to join @ycombinator's Startup Industrial Base: Building for the next 250. He highlighted how greater budget transparency and timely execution data can help the @DeptofWar, Congress, and industry better align resources with priorities, thus moving proven capabilities to the warfighter faster. He also advised innovators to solve real problems, understand the acquisition path, and demonstrate to the Department they can deliver at scale. "Having the users like your products is incredibly necessary," Hon. Hurst said. "But it's the program offices and the [portfolio acquisition executives, or] PAEs who really have the power to make you successful. You have to provide the solution to them. "Again, they’ve made their budget. They made it a year and a half ago. They've made a plan to execute that budget. So anything you're bringing in is, at times, disruptive to them if they have an established program. So you've got to figure out how you can fit in.”

127위S3 Partners@S3PartnersX에서 보기

[광고] S3 Partners: 사모신용은 직접 공매도할 시장이 없어 헤지펀드가 BDC·대체자산운용사·생보사를 대신 공매도하며, 생보사 공매도만 아직 줄지 않았다는 금융 분석 홍보입니다.

좋아요 4리포스트 2답글 0노출 4.1K북마크 2

128위TypeSafe AI@typesafeaiX에서 보기

Typesafe AI가 DSPy+Jev 타운홀 핵심 내용을 정리한 @dotpem의 글을 추천했습니다.

좋아요 5리포스트 1답글 0노출 3.3K북마크 1

129위BrendanEich@BrendanEichX에서 보기

브렌던 아이크: Brave는 검색 API·광고, 브라우저 광고, 프리미엄 상품, 암호화폐 제휴로 운영되며 Brave Origin은 리눅스에서 무료라고 밝혔습니다.

좋아요 15리포스트 0답글 1노출 534북마크 1

130위Corvex.ai@corvex_aiX에서 보기

[광고] Corvex가 GPU 클러스터 없이 오픈웨이트 모델을 API로 쓰고 기본 데이터 비보관을 내세운 서버리스 추론 플랫폼 Token Factory를 홍보했습니다.

좋아요 8리포스트 0답글 0노출 40.7K북마크 0

Corvex.ai@corvex_ai
Corvex Co-Founder and Co-CEO Seth Demsey joined @McnallieM to discuss the launch of our Corvex Token Factory, our new serverless inference platform, and how we’re approaching secure AI inference at scale. Developers get API access to open-weight models without running GPU clusters, with zero data retention by default. Watch the full conversation here: ow.ly/kuiQ50ZTmy3 Click here for more details on Corvex Token Factory: ow.ly/zU9t50ZTmy5 @power_analys1s #AIInfrastructure #AIInference #OpenWeightModels #MOVE

131위BrendanEich@BrendanEichX에서 보기

브렌던 아이크: Brave 검색이 월 26억 3천만 건(연환산 310억 건), 9월 28일 하루 최고 8,848만 건을 기록했다는 세부 통계를 공유했습니다.

좋아요 11리포스트 0답글 1노출 591북마크 0

BrendanEich@BrendanEich

@BrendanEich 님에게 보내는 답글

Brave Search stats QPM: 2.63B (+4.9% normalized to 31 days) Annualized (QPM normalized / 31 * 365): 31.02B (+4.9%) QPM non-Brave: 9.3% of total QPM QPD Peak 9/28: 88.48M (+3.4%) QPD 7-day average: 86.78M (+3.9%) QPD 28-day average: 85.09M (+4.9%) 3/4

132위Infrastructure For America@Inf4AmericaX에서 보기

[정치] 전기요금 상승의 주범은 데이터센터가 아니라고 주장하는 정책 캠페인 게시물입니다.

좋아요 0리포스트 0답글 1노출 34K북마크 0

133위BrendanEich@BrendanEichX에서 보기

[잡담/주제외] 브렌던 아이크: 영화 '시카리오'의 에밀리 블런트 캐릭터가 지나치게 순진하게 그려졌지만 그래서 관객이 몰입한다고 평했습니다.

좋아요 3리포스트 0답글 1노출 527북마크 0

BrendanEich@BrendanEich

@Hunter_Weiss 님에게 보내는 답글

I like it too, but there's a trick that bamboozles viewers (esp. men): Emily is too slight and pretty, which wins us over, but really unbelievably naive (consider scenes with Jon Bernthal). A more realistic version would not work, we'd stop suspending disbelief and nope out, IMO.

134위Creative Strategies, Inc@creativestratX에서 보기

Creative Strategies가 NVIDIA가 에이전트 안전을 인프라 계층으로 옮긴 소식을 자사 에이전트 사이버보안 연구와 엮어 분석한 리서치 노트를 공유했습니다.

좋아요 1리포스트 1답글 0노출 24.7K북마크 0

135위BrendanEich@BrendanEichX에서 보기

[잡담/주제외] 브렌던 아이크: 지인의 딸인 FBI 요원 사례를 들며 강인한 요원이었다면 그런 순진함은 설득력이 없었을 거라고 덧붙였습니다.

좋아요 2리포스트 0답글 0노출 515북마크 0

BrendanEich@BrendanEich

@BrendanEich 님에게 보내는 답글

Just for my bona fides, my dad's college buddy had a daughter who went full FBI back when it caught real bad guys, Clarice Starling no match; married DEA agent who went through doors. Think of Emily char played by Gina Carano or tougher. Would naiveté have been credible? Nope!

136위Modulate@modulate_aiX에서 보기

[광고] Modulate가 ElevenLabs·Qwen3-TTS 합성 음성을 잡아낸다는 딥페이크 음성 탐지 API(Velma Deepfake Detect)를 홍보했습니다.

좋아요 0리포스트 0답글 0노출 19.6K북마크 0

Modulate@modulate_ai
Synthetic voices are everywhere now. Knowing what's real isn't optional anymore. One enterprise engineer tested Deepfake Detect against audio generated by ElevenLabs & Qwen3-TTS: ⭐ 4.5/5 "Fast, Accurate Audio Fake Detection" That's Velma's Deepfake Detect API for you.

137위Together AI@togethercomputeX에서 보기

Together AI가 OpenRouter에서 GLM 5.3 Flash 29.2%, DeepSeek V4.1 Flash 25.6%, Kimi K3 18.9%로 주요 오픈 코딩 모델 토큰 점유율 1위라고 밝혔습니다.

좋아요 0리포스트 0답글 0노출 154.2K북마크 0

138위QumulusAI@QumulusAIX에서 보기

[광고] QumulusAI가 샌프란시스코 AI 컨퍼런스 현장에서 2027년 랙 스케일 솔루션 상담을 받는다고 홍보했습니다.

좋아요 0리포스트 0답글 0노출 12.9K북마크 0

139위Valency Systems Inc.@valencyioX에서 보기

[광고] Valency가 AI 가속에 대한 20년 관점을 내세워 Valency Bond 가입(학계 연구자 무료)을 권하는 홍보 게시물입니다.

좋아요 0리포스트 0답글 0노출 3.9K북마크 0

140위T. Rowe Price Digital Assets@TRowePrice_DAX에서 보기

[광고] T. Rowe Price가 암호화폐 시장은 균일하지 않다며 액티브 운용 상품을 홍보했습니다.

좋아요 0리포스트 0답글 0노출 41.6K북마크 0

141위Jean is building honest software@slashslashdevX에서 보기

[광고] Imbue가 정렬된 에이전트와 오픈소스 원칙을 내세운 Imbue Studio 대기자 등록을 홍보했습니다.

좋아요 0리포스트 0답글 0노출 5.9K북마크 0