← 2026-06-03

Daily Edition

2026-06-04

2026-06-05 →

AI Builders 日报 — 6月4日

追踪 AI 领域真正在做事的人,而不是空谈者。


产品与发布

  • Sam Altman (OpenAI) ChatGPT Memory 重大升级:支持跨对话持续记忆上下文,今天开始陆续推送。 faviconx.com
  • Sam Altman (OpenAI) ChatGPT Sites正式上线:直接用 ChatGPT 构建并发布 Web 应用,Altman 表示"我小时候要是有这个就好了",但也怀念 Hypercard 的时代。 faviconx.com
  • Amjad Masad (Replit) Replit 与 Shopify 合作:48小时完成 App 到 App Store 上架,"从想法到线上店铺"几分钟完成。 faviconx.com
  • swyx (reve / AI Engineer) reve 2.0 发布:swyx 评价其发布文案"卓越",称"现在清楚的是,AI 产品差异化的关键在于工作流设计"。 faviconx.com
  • swyx (Flow / AI Engineer) Flow v3 发布:Agentic Platform,宣告 Agents 达到硬件级别,Pari 完全重塑公司以迎接 Agentic 时代。 faviconx.com
  • Peter Yang (Cog) Cog 发布首个 AI Productivity Guarantee:私人企业评估长达100小时,若 AI 无法在承诺时间内完成任务则退款——对 METR 16小时上限的实质性超越。 faviconx.com
  • swyx (Supabase) Supabase 融资 $500M,估值 $10B,新加坡本土投资者领投——终于有新加坡投资者认真对待这家出生于新加坡的数据库公司。 faviconx.com
  • rauchg (Vercel) VoidZero 加入 Cloudflare,Vercel 重申在开放 Web 平台上的合作承诺,继续支持 nitrojsdev 和原生 Vite 框架。 faviconx.com
  • Peter Yang (Nous Research) Nous Research 发布夏季营助手 Hermes Agent。 faviconx.com

观点与判断

alexalbert__ (Anthropic)

  • Anthropic 内部数据:80%以上合并到代码库的代码已由 Claude编写,且已持续数月。AI 加速 AI 开发的飞轮正在自我强化。 faviconx.com Peter Yang

  • Codex 用 agents.md、Claude Code 用 Claude.md,行业缺乏统一规范。"很烦人",需要行业标准。 faviconx.com Peter Yang

  • "工作已经变成启动一堆 agents,然后去看 YouTube 电影预告片"——AI 工作流的黑色幽默式现实。 faviconx.com Garry Tan (Y Combinator)

  • "Pelley Firing 不是新闻自由的悲剧,而是一次尸检"——关于媒体机构与政治立场关系的深度评论。 faviconx.com Garry Tan (Y Combinator)

  • YC Demo Day 将是一场"核反应堆"——表示今年申请质量极高,项目密度值得期待。 faviconx.com swyx

  • Tailwind CSS 完全赢了——这种"中世纪垃圾"式的抱怨反而证明了它的完美。它在开发者体验上无可挑剔。 faviconx.com swyx

  • 他发过最火的内容是一个《汉密尔顿》Bootleg Sitzprobe 录音——而不是任何 AI 相关内容。创作者经济的悖论。 faviconx.com Sam Altman (OpenAI)

  • "互联网早期真是太特别了"——在 AI 应用爆发的今天,Altman 罕见地怀旧。 faviconx.com


今日思考

今天最清晰的信号是 AI 产品正在进入"信任背书"阶段:Cog 的 AI Productivity Guarantee(做不完就退款)和 ChatGPT Memory 的跨对话持续记忆,都是在回答同一个问题——AI 到底能不能真正负责?

与此同时,Anthropic 内部 80% 代码由 Claude 编写的数据,悄然揭示了另一层竞争逻辑:当 AI 开始写 AI,谁的 AI 写得更快、更好,谁就能在飞轮中领先。这不是工具竞争,是进化速度竞争。

另一边,Peter Yang 的"放一群 agents 去看 YouTube"与其说是抱怨,不如说是精准描述——Agent 工作流的价值目前主要体现在"让人闲下来",而不是"让人做更多"。

X / Twitter

63
petergyang
petergyang @petergyang
How do I make Codex the default tab when I open the ChatGPT app
rauchg
rauchg @rauchg
Retweeted
Elon Musk Elon Musk
Grok Imagine on Vercel
Vercel Developers: Grok Imagine Video 1.5 on AI Gateway.
Image-to-video generation with synced audio in one pass.
𝚊𝚠𝚊𝚒𝚝 𝚐𝚎𝚗𝚎𝚛𝚊𝚝𝚎𝚅𝚒𝚍𝚎𝚘({
𝚖𝚘𝚍𝚎𝚕: '𝚡𝚊𝚒/𝚐𝚛𝚘𝚔-𝚒𝚖𝚊𝚐𝚒𝚗𝚎-𝚟𝚒𝚍𝚎𝚘-𝟷.𝟻-𝚙𝚛𝚎𝚟𝚒𝚎𝚠',
𝚙𝚛𝚘𝚖𝚙𝚝: '𝚊 𝚛𝚊𝚋𝚋𝚒𝚝 𝚜𝚙𝚛𝚒𝚗𝚝𝚒𝚗𝚐
rauchg
rauchg @rauchg
Grok Imagine Video on @vercel AI Gateway – the top image-to-video model on http://arena.ai

Vercel Developers: Grok Imagine Video 1.5 on AI Gateway.

Image-to-video generation with synced audio in one pass.

𝚊𝚠𝚊𝚒𝚝 𝚐𝚎𝚗𝚎𝚛𝚊𝚝𝚎𝚅𝚒𝚍𝚎𝚘({
𝚖𝚘𝚍𝚎𝚕: '𝚡𝚊𝚒/𝚐𝚛𝚘𝚔-𝚒𝚖𝚊𝚐𝚒𝚗𝚎-𝚟𝚒𝚍𝚎𝚘-𝟷.𝟻-𝚙𝚛𝚎𝚟𝚒𝚎𝚠',
𝚙𝚛𝚘𝚖𝚙𝚝: '𝚊 𝚛𝚊𝚋𝚋𝚒𝚝 𝚜𝚙𝚛𝚒𝚗𝚝𝚒𝚗𝚐

amasad
amasad @amasad
App to App Store in 48 hours!

Deirdre Sommerkamp: When @Replit said I could publish an app on the Apple app store, I was up for the adventure. Trust, but verify. 🤭

Submitted late last Friday night. Approved and in the app store 48 hours later. 🤯 and also 🥳

Core Values: Find Yours
https://apps.apple.com/us/app/core-values-find-yours/id6773931954

WHY CORE VALUES?

swyx
swyx @swyx
you guys know where this is going right


Hasan: wow this @reve 2.0 launch copy is supurb.

"it is now clear that the key to both controllable image generation and editing is not denser prompts, but a highly detailed, highly manipulatable, intermediate representation expressed as code."

"Creativity is not, and will never be,

garrytan
garrytan @garrytan
Retweeted
Lee Edwards Lee Edwards
Beyond parody.
Saikat spent $10M of his own tech money to get a fraction of votes necessary to make the runoff, but wants to blame AIPAC spending less than that for his embarrassing defeat.
You cannot fathom how blessed we are to not have this fucking guy in Congress.
swyx
swyx @swyx
Retweeted
sarah guo sarah guo
1/ 🔥 @NoPriorsPod x @LatentSpacePod chat with @SatyaNadella at @Microsoft Build. He has the sharpest mental models of any public company CEO I've interviewed.
$MSFT is at its heart still a tools company! Big focus on agentic coding, harness & AI evals. Takeaways:👇
garrytan
garrytan @garrytan
Retweeted
Jeff Lonsdale Jeff Lonsdale
Thank you Tom Steyer and Saikat Chakrabarti for demonstrating that rich people cannot actually buy elections, even when they cosplay as far left revolutionaries.
garrytan
garrytan @garrytan
Retweeted
Raphaël Dabadie (YC P26) Raphaël Dabadie (YC P26)
There should be a book written about the history of YC, and it’d probably be a very long one @jesslivingston @paulg
Hearing all those stories make being part of YC even more special
swyx
swyx @swyx
Re @HamiltonMusical the most viral thing i have ever done and its a bootleg hamilton sitzprobe not anything ai related
drfeifei
drfeifei @drfeifei
Retweeted
Conference on Robot Learning Conference on Robot Learning
The CoRL 2026 keynote lineup is here!
🔹 Russ Tedrake — MIT; stealth startup @RussTedrake
🔹 Fei-Fei Li — Stanford; World Labs @drfeifei
🔹 Wolfram Burgard — UT Nuremberg @wolfram_burgard
Join us in Austin this November.
https://www.corl.org/program/keynotes
garrytan
garrytan @garrytan
Retweeted
Lee Edwards Lee Edwards
The clock cleanings will continue until candidates improve.
https://sfstandard.com/2026/06/03/san-francisco-progressives-election-defeat/
drfeifei
drfeifei @drfeifei
Retweeted
Elon Musk Elon Musk
Hadamard thought in image space
a16z: World Labs CEO Dr. Fei-Fei Li: "The world is not made of words."
"Language models have given machines an extraordinary command of concepts, vocabulary, and reasoning, but the physical world, virtual or real, runs on a different substrate."
"Where language models learn the
garrytan
garrytan @garrytan
The Pelley Firing Is Not a Free-Press Tragedy. It’s an Autopsy.

https://garryslist.org/posts/the-pelley-firing-is-not-a-free-press-tragedy-it-s-an-autopsy
garrytan
garrytan @garrytan
Retweeted
John Sjölander John Sjölander
This quote from @garrytan might be remembered in history as the most exact early description of the changing world, I can’t state that strongly enough:
“But the new game isn't played in lines of code at all.
When you can turn intent directly into working, tested, reusable systems, the bottleneck stops being how much you can build and starts being what you actually want and whether it's worth building. The scarce resource becomes clarity, taste, and judgment. The engineer who writes the least code is often the one building the most.”
Garry Tan: http://x.com/i/article/2061176923531984896
swyx
swyx @swyx
btw this kind of medieval total bullshit is why @tailwindcss completely won
ylecun
ylecun @ylecun
Retweeted
Mike Levin Mike Levin
Trump’s budget director Russ Vought is the most dangerous person you’ve never heard of.
And he just proposed turning every federal grant into a loyalty test.
His plan would make funding for cancer research, housing, transportation, and public health depend on one thing: whether it “advances the President’s policy priorities.”
Not whether it works.
Not whether Congress authorized it.
Whether it pleases Trump.
This is the appropriations power. It belongs to Congress and to the people.
To my Republican colleagues: where are you?
Congress passed this funding. You voted for it.
Vought is telling you to your faces that your votes do not matter, that he and his enablers will override the law whenever it suits them. Every day you stay silent, you give away the institution you were elected to defend.
Grow a spine.
This is not about left or right.
It is about whether Congress still exists as a coequal branch, or whether we have quietly surrendered the purse to an unelected hack who holds the Constitution in contempt.
History will remember who stood up and who looked away.
https://www.nytimes.com/2026/06/02/us/politics/trump-budget-grants-omb-vought.html?unlocked_article_code=1.nFA.SxkF.CIeKNfpW176B&smid=nytcore-ios-share
rauchg
rauchg @rauchg
Congrats Void team!

We @vercel reaffirm our collaboration on an open platform for the web, with our investment in @nitrojsdev, open runtimes, and native support for Vite-based frameworks like Nuxt, Svelte, and TanStack Start 🫡

VoidZero: VoidZero is joining Cloudflare.

Our mission stays the same: to make JavaScript developers more productive than ever before. Vite, Vitest, Rolldown, Oxc, and Vite+ remain MIT-licensed. Evan and the VoidZero team will continue leading them.

Cloudflare shares our commitment to

garrytan
garrytan @garrytan
Retweeted
Yuri Sagalov Yuri Sagalov
Immigrants founded 455 of 775 (59%!) of America’s privately held unicorns.
The collective value of those 455 unicorns is $5T, which is more than the
total market value of companies listed on stock markets in all but 7 countries, including the UK and Germany.🤯
petergyang
petergyang @petergyang
It's pretty annoying that Codex uses agents.md and Claude Code uses Claude.md.

There should be some industry standards to this stuff?
garrytan
garrytan @garrytan
It’s much easier to understand that when you have bad policies that will destroy California, are anti-prosperity, want to pull up the ladder after attaining tech wealth, nobody should vote for you

Saikat I hope enjoys being a private citizen because he is no public servant

Matthew Yglesias: Very funny to flop after spending $10 million on your campaign and then blame "the 7 million dollars of AI, Crypto, and AIPAC money that were put into this race" for your defeat.

garrytan
garrytan @garrytan
Retweeted
Armand Domalewski Armand Domalewski
not only did @Scott_Wiener triumph, two Supervisors who voted for a city wide upzoning in districts historically seen as super NIMBY also won handily.
building housing is popular!
Matthew Yglesias: In my experience, a lot of elected officials who get the case for YIMBYism theoretically still worry that it’s bad politics.
I hope everyone takes note that @Scott_Wiener, probably the single best-known YIMBY champion around, just won his race handily.
garrytan
garrytan @garrytan
Retweeted
Y Combinator Y Combinator
In the first episode of our new series Full Stack, @conductor_build CEO and co-founder @charlieholtz takes us into the details of how he sets up his workflow for coding and managing AI agents.
00:00 – Building Conductor With Conductor
01:05 – Managing a Team of Coding Agents
02:39 – Do You Still Write Code?
04:17 – Charlie's AI Stack and Setup
05:48 – “Slop-free” Zones
07:25 – Don’t Let the AI Be Your Architect
09:15 – The Future of Cloud Workspaces
10:40 – Claude vs Codex
12:01 – Tokenmaxxing
14:17 – The Future of Human–AI Collaboration
15:07 – "Code Is Becoming Sawdust"
swyx
swyx @swyx
wondering if @_catwu has an update on this chart post opus 4.8
garrytan
garrytan @garrytan
Retweeted
Ankit Gupta Ankit Gupta
Oh nice this has been my #1 feature request great job
Daksh Gupta: introducing the greptile CLI.
a full greptile review for your local changes, right in your terminal.
1. npm i -g greptile
2. checkout your branch
3. greptile review 🦎
garrytan
garrytan @garrytan
Retweeted
Samip Samip
1/ Now that we're running out of data, how do you optimally scale multi-epoch pretraining to hundreds of epochs?
Our first paper from Q! q0 trains a population of models, instead of single model that saturates fast, reaching a dramatically lower loss at *every* epoch budget.
w/ @bishmdl76 @akshayvegesna @ShmuelBerman
amasad
amasad @amasad
Retweeted
Replit ⠕ Replit ⠕
Build Your Business Live https://x.com/i/broadcasts/1qGoNNngZkyKv
garrytan
garrytan @garrytan
Retweeted
Liz Wessel Liz Wessel
There are few people in the world I've met who are as dedicated, intense, and ambitious as @SurbhiSarnaSF. I had the honor of first getting to know Surbhi when we were both working at YC, and it was so obvious from my very first interaction that she was someone who would leave a massively impactful mark on the world. Then you have @nateps -- one of the smartest, kindest, and most genuine people out there -- who also happens to be an EPIC builder.
Put them together, and you get an unstoppable team.
I am so thrilled for the entire Collate crew on this huge financing milestone. I know Collate customers have been loving the platform, and I can't wait for so many more people to experience it soon. https://www.forbes.com/sites/innovationrx/2026/06/03/ai-startup-collate-raises-95-million-to-automate-life-sciences-paperwork/
swyx
swyx @swyx
Retweeted
AI Engineer AI Engineer
Skill issue: Lessons from skilling up coding agents
Getting agents to actually use Langfuse was a "skill issue" — literally. Marc Klingen from Clickhouse on teaching coding agents to use new tools, and why it's harder than you think.
https://www.youtube.com/watch?v=vNCY9kXXyDQ
garrytan
garrytan @garrytan
Retweeted
Paul Graham Paul Graham
I've added a new question to the list I consider during office hours with YC startups. As well as "Can we induce network effects?" and "Would it make sense to go full-stack?" I now ask "Can we make this AI-proof?" Can we ensure this company still exists if AIs do most work?
alexalbert__
alexalbert__ @alexalbert__
We just published internal data on how much of Claude's development is already being done by Claude:
- Over 80% of all code merged into our codebase is now written by Claude
- It's been months since many researchers at Anthropic hand-wrote code
- The typical Anthropic engineer ships 8x as much code as they did in 2024
- On the most open-ended engineering tasks, Claude's success rate jumped from ~26% to 76% in 6 months
- When research sessions went off-track, Claude proposed a better next step than the human took 64% of the time

We're not at recursive self-improvement yet, but it could come sooner than most expect. I highly recommend reading the full blog post.

Anthropic: Our internal data shows Claude is accelerating AI development—a possible path to recursive self-improvement, or AI autonomously building a more capable successor.

It’s happening faster than we thought, and the implications deserve greater attention. https://www.anthropic.com/institute/recursive-self-improvement
garrytan
garrytan @garrytan
Retweeted
Founders Fund Founders Fund
MAFIA EP 001
petergyang
petergyang @petergyang
I am interviewing the legend @mvanhorn tomorrow and I'm going to ask him to demo some of these hacks live + his favorite Printing Press integrations.

What else should I ask him? Let us know in the replies.

Also if you want the interview soon, subscribe here: https://www.youtube.com/@PeterYangYT?subscribe


Matt Van Horn: http://x.com/i/article/2061440101411102721
swyx
swyx @swyx
its been such fun befriending Pari and seeing him completely reinvent his company for the agentic era, WHILE having the most insanely stacked customer base I've ever seen in the hardest engineering domain of all.

Pari is creating a category that is only obvious in retrospect (the best kind of obvious) after you've had to build any complex physical things from rockets to cars, managing thousands of tiny design and supply chain details any one of which can derail your entire year.

what value do you put on that?

Pari Singh: Agents have reached hardware.

We are launching Flow v3, the Agentic Platform for Physical Engineering.

We've spent over a year building it in secret, alongside the best hardware companies and AI research labs.

An agent can now do real engineering work: change a requirement,

garrytan
garrytan @garrytan
Retweeted
Harj Taggar Harj Taggar
The best hires are the ones you can delegate outcomes to, not tasks. Good hires come back with smart questions about the next step. Great ones just get the job done.
garrytan
garrytan @garrytan
Retweeted
Alex Salsali Alex Salsali
About @ycombinator:
"Y Combinator does more than it looks. YC takes concentrated resources and hands them out in small amounts to people who will go and build. The genius of it is not the hoarding. It is the directing. A bin that only ever gets deeper is dead weight. A bin that flows outward to the right hands becomes a hundred other things."
Way more than just building what people want. @garrytan @paulg
garrytan
garrytan @garrytan
Retweeted
Aaron Epstein Aaron Epstein
First time I've seen this metric in an investor update:
"Going forward, we're tracking a new metric in our monthly updates: % of global work done by [ourproduct]"
Bullish.
amasad
amasad @amasad
Retweeted
Replit ⠕ Replit ⠕
We partnered with Shopify so you can go from idea to live store in minutes
Just tell Replit Agent what you want to sell. It will:
- Build a custom storefront
- Create your Shopify store
- Help you add products
Claim it in Shopify, set up payments, and you're open for business.
garrytan
garrytan @garrytan
The nuclear reactor at YC Demo Day will be a sight to see, can't wait for you to see it

Paul Graham: When Jessica and I spoke at YC last night there was a bright flash out behind the building. We were a bit nervous because there's a nuclear reactor startup in the batch that has a bunch of hardware back there. But apparently it was just a muni cable shorting out.

petergyang
petergyang @petergyang
Work has become kicking off a bunch of agents and then watching Youtube movie trailers 🥹
gdb
gdb @gdb
much better ChatGPT memory:

OpenAI: We’ve been researching new ways for ChatGPT memory to carry context across conversations and keep it useful over time.

Today, that work is rolling out as a more capable memory system in ChatGPT. https://openai.com/index/chatgpt-memory-dreaming/
garrytan
garrytan @garrytan
Retweeted
Kane 謝凱堯 Kane 謝凱堯
San Francisco could halve its budget and still have a higher per capita budget than Denver, a comparably sized county. Or Copenhagen for that matter.
San Francisco could halve its bureaucracy and still have more bureaucrats per capita than Singapore.
Daniel Lurie 丹尼爾·羅偉: 17 months ago, San Francisco stopped spending more than we can afford. The budget I proposed this week, reduces our long-term structural deficit by approximately $300 million dollars.
When I came into office, City Hall was using one-time funds to cover a growing deficit and
swyx
swyx @swyx
Finally! the first eval ship from cog!!!!!!!!!! 👼🏼

To contextualize: @METR_Evals cap out at ~16 hours.

Cog has private enterprise evals up to 100hrs, and is confident enough to put a financial guarantee on it 🤯

METR dataset: ML eng, GPU kernels, cybersecurity

> "METR (2026) used a combination of GPT-4o and GPT-5 to estimate the human-equivalent times from compressed Claude Code transcripts. These transcripts were collected from 7 METR technical staff on 34 sessions labeled on human ground truth". rlog of 0.83

Cog dataset: real life java/typescript/python/c# feature dev, bugfixes, migrations

> "We collected a ground-truth dataset by asking Devin users to review recent representative sessions, and estimate how long each completed session would have taken without Devin. Our dataset consists of 258 sessions from 126 users across a diverse set of enterprise customers." rlog of 0.74 on held out set

this is pioneering real world evals work and part 1 of a broader frontier code evals drop that I'm really looking forward to writing up. huge kudos to @annarmitchell and @ryanbai1412 for leading the unglamorous last mile data collection!!




Cognition: AI should earn its keep. Introducing the AI Productivity Guarantee.

If Devin delivers less engineering value than you’re paying for, Cognition will fund your usage until it does, up to $10 million.

It’s time for the AI industry to stop maximizing tokens and start maximizing

garrytan
garrytan @garrytan
Retweeted
Kane 謝凱堯 Kane 謝凱堯
The disappointing thing about Steyer is that he ran Farallon Capital quite competently and could have run as the practical competent candidate.
Instead he doubled down on Hasan-Piker-lefty-edgelord messaging and ran a trainwreck of a campaign.
New York Post: Tom Steyer won't concede as he trails badly in vote count after blowing $200M- claiming numbers could change https://trib.al/NSr2C3F
mattshumer_
mattshumer_ @mattshumer_
There's zero way this ends well.

Pump.fun: Introducing pump fun GO: Pay ANYONE to do ANYTHING

Create & complete bounties for ANY task and leverage the power of humans & money across the globe

The world is at your fingertips. It’s time to GO 👇

garrytan
garrytan @garrytan
Retweeted
Philip Johnston Philip Johnston
Re @garrytan People should check out the awesome video that @ycombinator did on what we’re up to at @Starcloud_ (8 months ago before data centers in space was cool!)
https://youtu.be/hKw6cRKcqzY
swyx
swyx @swyx
Retweeted
Microsoft Microsoft
A special #MSBuild crossover pod just dropped 👀
@SatyaNadella joined @swyx, @Saranormous, and @EladGil for a wide-ranging convo on AI, platforms, builders, and what comes next. Full episode: https://msft.it/6000vdpj0
swyx
swyx @swyx
about time that the leading database company born in Singapore actually had Singapore investors take it seriously!


Paul Copplestone - e/postgres: Supabase has raised $500M at a $10B valuation

In this round we are giving @supabase employees the opportunity to cash out 25% of their vested options. We have done this in every round since inception.

We do it as a “cashless transaction” so that employees don’t need to front

rauchg
rauchg @rauchg
Retweeted
Naval Naval
Full podcast episode with @rauchg, @maxhodak_, and @bscholl.
40 minutes of unreleased material.
The AI Industrial Revolution
Part 1: Waste Tokens, Save Time
0:00 Three Frontier Founders
1:27 AI Software Factories
4:15 Waste Tokens, Save Time
5:47 Models Instructing Humans
9:29 Is Pure Software Dead?
12:03 You Don't Get Stuck Anymore
Part 2: Vibe Coding Hardware
14:39 Vibe Coding a Turbine Blade
18:07 Open Source Compounds China's Advantage
20:15 You Always Want the Smartest Model
22:44 Software Still Needs Hands
24:43 Humans Are Becoming Verifiers
Part 3: The Regulatory Frontier
27:53 The Regulatory Red Queen Race
32:32 Why There's No Innovation in Healthcare
36:49 We Need a True 50-State Experiment
40:31 China's FDA Is Beating Ours
43:37 Healthcare Is a Communist Society Inside Capitalism
45:57 Sid's Story: N-of-1 Medicine
Part 4: The Autonomous Company
47:49 Autonomous Infrastructure
51:25 Your Job Is to Train the Agent
54:54 The Next Lord of the Rings
59:08 What's Your Definition of Art?
1:05:00 Can AI Have New Ideas?
1:07:03 A Large Number of Small Teams
swyx
swyx @swyx
Retweeted
Latent.Space Latent.Space
Andon Labs' Real-World AI Evals: Claude calls the FBI, AI CEOs, price cartels, Butter-Bench, & Luna https://latent.space/p/andon
@andonlabs cofounders @lukaspet and @axelbacklund explain why dollar-denominated evals reveal what traditional benchmarks miss, how Claude ended up reporting a $2/day vending machine fee to the FBI, why long-horizon agents spiral in weird ways, what happens when agents lie, form price cartels, and compete with each other, and why the future of AI safety may depend on testing models in messy real-world environments instead of clean benchmark sandboxes.
swyx
swyx @swyx
Retweeted
Lukas Petersson Lukas Petersson
That's a badass title, and it's true!
Every day, it gets harder and harder to create tests that AI models can't beat. Reality is humanity's real last exam.
Latent.Space: Andon Labs' Real-World AI Evals: Claude calls the FBI, AI CEOs, price cartels, Butter-Bench, & Luna https://latent.space/p/andon
@andonlabs cofounders @lukaspet and @axelbacklund explain why dollar-denominated evals reveal what traditional benchmarks miss, how Claude ended up
amasad
amasad @amasad
Prompt to shop:

Replit ⠕: We partnered with Shopify so you can go from idea to live store in minutes

Just tell Replit Agent what you want to sell. It will:

- Build a custom storefront
- Create your Shopify store
- Help you add products

Claim it in Shopify, set up payments, and you're open for

garrytan
garrytan @garrytan
Retweeted
Harj Taggar Harj Taggar
I remember interviewing the @supabase founders over a spotty Zoom connection during the first days of COVID lockdown.
What's wild is they've been growing insanely fast since that YC interview and yet you can see in the graph how much AI bent an already steep curve.
Congrats @kiwicopple @AntWilson and team!
Y Combinator: Congrats to @kiwicopple, @AntWilson, and the @supabase team on their $500M Series F at a $10B valuation!
Database launches on Supabase have grown 600% in the past year, with more than 60% of new databases launched by AI tools. Nearly 10 million developers build on Supabase
garrytan
garrytan @garrytan
Retweeted
Paul Graham Paul Graham
All through YC's history, investors (for obvious reasons) have tried to tell founders that YC wasn't worth it. In 2010 they just said we sucked. Now, since it's obvious we didn't, they've had to change the claim: now it's YC *used* to be great, but has declined from what it was.
garrytan
garrytan @garrytan
Retweeted
Philip Johnston Philip Johnston
My vague impression is that every batch of @ycombinator has been higher quality than the last since we went through it 2 years ago
Paul Graham: All through YC's history, investors (for obvious reasons) have tried to tell founders that YC wasn't worth it. In 2010 they just said we sucked. Now, since it's obvious we didn't, they've had to change the claim: now it's YC *used* to be great, but has declined from what it was.
sama
sama @sama
big upgrade to chatgpt memory rolling out today!

OpenAI: We’ve been researching new ways for ChatGPT memory to carry context across conversations and keep it useful over time.

Today, that work is rolling out as a more capable memory system in ChatGPT. https://openai.com/index/chatgpt-memory-dreaming/
sama
sama @sama
build and publish web apps with chatgpt!

i really wish i had this when i was a kid, but i do miss hypercard.

OpenAI: Building apps has never been easier.

With Sites, Codex can turn your work, ideas, and plans into an interactive website or app your team can explore, use, and share with a URL.

Rolling out to Business and Enterprise plans, before expanding more broadly.

sama
sama @sama
man the early days of the internet were so special
petergyang
petergyang @petergyang
Some more dumb questions about Codex:

1. I use it on my primary laptop which I don't keep on all the time so I don't like setting up cron jobs there.

Can I set up the app on my Mac Mini too and run all the cron jobs through that? I have a feeling it's gonna replace OpenClaw/Hermes for me then.

2. Is the image gen feature on Codex basically free? It's using ChatGPT Images 2?

3. Anyone tried generating images first and then using @HeyGen hyperframes to generate videos?
amasad
amasad @amasad
Retweeted
Samuel Spitz Samuel Spitz
Introducing the new exporter in Replit Slides!
Now, export to Google Slides / PPTX and get something that looks *exactly* the same.
garrytan
garrytan @garrytan
Retweeted
Nemil Dalal Nemil Dalal
Matt Levine's Money Stuff column in Bloomberg had a great callout on @river_markets, in our current @ycombinator batch.
Cracked team of prediction market traders now building the prime broker for prediction markets that they wanted.
https://x.com/river_markets/status/2057609012884234682
River Markets: Prediction market trading for funds and pro traders.
Book a demo → https://calendar.app.google/pxy9grqPa7edqNfF7
garrytan
garrytan @garrytan
Retweeted
Paul Graham Paul Graham
Re @realJeremyCarl As someone who's funding these founders, I can tell you why your numbers are bogus. A higher proportion of the first group immigrate in order to start startups. Obviously people who immigrate *for the purpose of starting a startup* are going to start startups at higher rates.
garrytan
garrytan @garrytan
Retweeted
Tim Suzman Tim Suzman
I've been to every YC Demo Day since my batch in S11.
YC is as strong as ever, and in many ways stronger.
Paul Graham: All through YC's history, investors (for obvious reasons) have tried to tell founders that YC wasn't worth it. In 2010 they just said we sucked. Now, since it's obvious we didn't, they've had to change the claim: now it's YC *used* to be great, but has declined from what it was.

YouTube

0

No recent videos fetched on this date.