소쿠릿🦋 @devkcuri

"Vector databases are dead". And yet one of the most interesting infra companies in the world right now... is a vector database. I reviewed 20 expert calls on @turbopuffer + all public data on the company. Co was founded in Ottawa in Oct 2023 by @Sirupsen (ex-Shopify). Here is what we know: - $1m to $100m run-rate in 19 months - profitable the entire way - <$1m of outside capital raised (!) to date - <40 people, no sales team - customers: @cursor_ai, @AnthropicAI, @NotionHQ, @Atlassian, @linear, @harvey, @tryramp... So what does Turbopuffer do? Well, every AI model needs a "find what's relevant" step before it can answer (could be your codebase in Cursor, your workspace in Notion...). Turbopuffer is that search layer, and it is rented as an API. It combines both classic keyword search (exact text hits) and vector search (search by meaning. Basically so that "cancel subscription" also returns "stop billing" topics). Great, but why does it win? Mostly one architectural bet. Everyone else kept data in RAM (~$3,600 per TB p.m.). Turbopuffer keeps it on S3 (~$20 per TB) with an SSD cache on top (~$70 per TB all-in). The one drawback is accepting a slower first query, but it runs 10-100x cheaper at scale. The calls and references are the most one-sided I have ever read on any infra company: Of the 20 calls, 13 were current customers across 9 orgs. Zero churn. Implied NPS ~ +70 (6 of 8 scores were 9s or 10s) - A leading productivity company migrated 10bn+ vectors off Pinecone - A large collaboration software company ran a formal RFP against Pinecone, Qdrant, Weaviate and LanceDB. Turbopuffer won, and now carries ~75% of their semantic search traffic in the EU - A fast-growing AI support platform: identical workload quoted at 5x+ price on Pinecone. They went as far as saying that even if Turbopuffer doubled their pricing tomorrow, they wouldn't blink - A large Canadian telco (2 yrs in prod): "we've never been paged, infra has never gone down"... and went from a cost projection of millions to thousands - A leading AI talent marketplace deprecated Pinecone entirely; on their new workload Turbopuffer came in 50-100x cheaper than Mongo, and they said it is one of the few products developers would cry over if taken out And the best line of all, from one of the experts: "if turbopuffer didn't exist, I don't think Cursor could have existed as a product" Agentic AI is also a tailwind, not the threat people assumed: agents perform 25-50x more queries than human users. And agent loops are inference-bound so they tolerate retrieval latency ==> that's exactly the trade-off Turbopuffer makes. There is one open question here: AWS basically already copied the architecture natively (S3 Vectors, with claims of it being "90% cheaper"). That direct threat was flagged 5 times in my 20 calls reviewed. And that's actually super cool... Turbopuffer's answer is clear: keep cutting prices faster than anyone can chase (Elon's playbook on EVs with Tesla). TP has cut query prices by up to 94% since launch... and stayed profitable doing it. And so ultimately: does compounding cost advantage + total focus beat distribution? It did once before: that is literally Snowflake vs Redshift (and my favourite horse near that race is @ClickHouseDB in the same vein, per my earlier posts) Thoughts @HarryStebbings, @jaminball, @GavinSBaker, @andy_pavlo, @bernhardsson?

today, Vercel's Eve agents can run on agentOS agent frameworks require a heavy sandbox with 1 GB+ of RAM to run but most agents just read files + run simple scripts agentOS provides a lightweight OS for these workloads built on WebAssembly & V8

Rivet @rivet_dev

2 weeks ago

Introducing Vercel Eve × agentOS A secure, lightweight OS without the sandbox 🧬 WebAssembly & V8, not a microVM 🪶 Library, not infra / SaaS 📁 Mount anything as a filesystem (S3, Google Drive, …) ⚡ ~4.8ms cold starts, 254× cheaper to run⚡

17 26 471 65K 463

3.5 프로도 안나왔는데 4는 무슨..

Logan Kilpatrick @OfficialLoganK

2 weeks ago

We have started our most ambitious pre-training run yet, for Gemini 4, and are excited by the progress : )

1K 659 14K 1.9M 913

@Sayak1e 거기에 너무 마음쓰지마세요 저는 맨날 개소리하지 말라고 하거든요

열심히 써서 한도 거의 다쓰면 냅다 리필해주는것 너무 좋아.. 덕분에 여러가지 많이 하고있구

Tibo @thsottiaux

2 weeks ago

10M! New day, new usage reset for paid users of Codex and ChatGPT Work. Lands in the next hour. Enjoy.

3K 1K 21K 2.7M 910

난 마크다운이 싫던데 간단히 쓰라고 만들어둔 물건 이것도쓰고 저기도쓰고 좀아닌것같아

今どきのAIはハードも書けます。 RiscVコアと周辺SoC作らせてFPGAでMotion JPEG再生出来るようなシステムが1週間くらいで出来たよ。 1行もコードを書いてないです。 RTL記述からシミュレーション、実機デバッグまで全部やらせました。

코덱스는 또 5h 제한을 없애버렸네...

Morning. The last 48 hours of Codex and ChatGPT Work have been intense! Three important updates: - Temporarily removing the 5 hour usage limit restriction for all Plus, Business and Pro plans - Rolling out changes that will make GPT 5.6 Sol more efficient across the board and that will be reflected in less usage being used so that it can take you further. Exact impact to be quantified and shared - We hit 6M active users, and are landing a usage reset in the next hour Go do things

이거 공감 많이 가는듯 lethain.com/revised-rules-…

Claude Fable 5 will be available again globally tomorrow. After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding and debugging will fall back to Opus 4.8. We’ll continue to refine these classifiers over the coming weeks to reduce false positives and better distinguish genuine misuse from legitimate requests. We’ve also begun drafting a consensus framework—with Amazon, Microsoft, Google, and other Glasswing partners—for assessing the severity of AI jailbreaks and how AI developers should respond to them. We invite other industry partners and model providers to join us in this effort. Finally, we’re scaling up our collaboration with the US government on model testing and safeguards. This will include pre-release access to models and safeguards for evaluation, information sharing on jailbreaks and misuse, and dedicated resources for joint research. Thank you to our users for your patience, and to our partners across the government, industry, and the research community who worked alongside us to make Fable 5 available again. Read our full blog: anthropic.com/news/redeployi…

추천 타임라인 쓰레기같은것밖에 안나오네 리트윗좀 그만해야겠다

codex에서 gpt-5.6-sol로 라우팅되고 있다는 소문이 사실로 확인되었습니다. 일부 데이터 및 훈련 허용한 사용자에게 A/B 테스트가 이루어지는 모양 chatgpt.com/codex/cloud/se… 위 링크에서 라우팅 여부 확인가능함

The first nail in the coffin of every sandbox-only startup:

Secret level 6: there’s a problem, but it’s not worth solving, so leave it alone. This is why org with a ton of level 1s can win. They don’t get distracted by side quests

Alex Lieberman @businessbarista

a month ago

I stole this idea and now use it with every single employee. It’s the best illustration I’ve seen of teaching someone to be high agency. It says there are 5 levels of work: Level 1: “There is a problem.” Level 2: “There is a problem, and I’ve found some causes.” Level 3:

195 692 7K 1.1M 11K

これはまじですばらしいな。 —— SMPTE、規格カタログ全編を無料公開　1世紀にわたる有料化体制を撤廃-cined.com/jp/smpte-opens…

Building products in the age of AGI

SQLite 数据库的作者 Richard Hipp 解释，为什么他的项目一律不接受外部的 PR。 PR 不是免费的。你实际上是对我提要求：你开发了这个很棒的功能，然后希望我帮你维护它、帮你编写文档、帮你测试，并在接下来的二十五年里一直为你维护它。 Linus 曾说过一句名言：Free 既可以指免费啤酒，也可以指言论自由。但还有另一种 Free：免费的小狗。“瞧，我这儿有只免费的小狗送给你。” 你明白我的意思了吧？ 提交一个 pull request 就相当于有人送你一只小狗。一天下来，你的小屋里就多了一只小狗。你不能把它扔掉——你有道义上的责任照顾它，直到它自然死亡。 我可不要任何免费的小狗。ruanyifeng.com/blog/2026/06/w…

Seeing chatter about Fable 5 being accessible - can say categorically this is false, we are not serving any Fable / Mythos traffic Looking into possibility of UI bug on front-end (e.g. based on historical context), but also very real chance it’s just people shitposting…

sam mcallister @sammcallister

a month ago

@kimmonismus We are currently serving exactly 0 traffic to Fable 5. This could be a UI bug though, will track it down.

48 22 400 429K 49

🚨 BREAKING: Anthropic is "confident that in the coming days [Fable 5] will become available again" - Anthropic's International Managing Director we might be back