Eugene Shvarts

building the future of ai @typesafeai

[Eugene Shvarts @mathfax]

@negativestats @ctjlewis O he olo mark

[Eugene Shvarts @mathfax]

@altryne Yaaaa! Honestly I’m just trying them all as I run + make demos

[Eugene Shvarts @mathfax]

We're pushing updates to cookbooks, best practices, and use cases as fast as we can, but the community is 1000x-ing us here.

The Factorio-ing is real too; some of the highest leverage additions are automating your automation, like teaching your agents how to use Jev directly in your applications. One example in the wild: https://t.co/hTcW4QMU59

[Eugene Shvarts @mathfax]

@sheherenow_ You thought we studied model dumbness but it’s actually model dumbness

[Eugene Shvarts @mathfax]

@dorkitude @sheherenow_ Amazing use-case, and a great argument for S1 calls peppering every interface between agent tool calls

[Kyle Wild @dorkitude]

tl;dr Jev makes web search more context-efficient. I'm open sourcing a thing that does this for you: https://t.co/JcHYc06r57

So, I ran out of Codex tokens last week, then ran out on my backup account. Same for my Claude tokens.

So I did a little context usage audit, and one of the biggest culprits was web research -- I "read" a lot of academic papers. Like, all of them.

I figured TypeSafe's new Jev model could help me make something more efficient than burning expensive Astra and Fable tokens parsing through web results in the raw. Plus I needed a search CLI for my Pi/Kimi workflows.

So I made this, and it's pretty dope!

[Eugene Shvarts @mathfax]

@luizribeiro @typesafeai Okay sick but can you structure the criteria?

[Eugene Shvarts @mathfax]

maybe we should've gone with this instead of Doom

[Max Rozen @RozenMD]

this is art

[Pete Soderling @petesoder]

3 months ago, an anonymous account commented "Y'all are sleeping on the most important talk of this conference". The speaker was @CompleteSkeptic and his talk was on RLHF's deal with the devil, and the argument behind Jev.

Diogo argued that training AI to produce answers people prefer comes with a trade-off. You get polished, coherent responses. You lose the unusual alternatives through mode collapse.

The problem is that the unusual alternative can be the correct one.

Three months later, @typesafeai's Jev, a model designed for decisions, is now generally available to everyone.

If Jev has caught your attention over the weekend (the internet is on fire with it after all!), Diogo’s @AICouncilConf talk is worth watching. It lays out the argument behind what his team is building and why they chose this direction.

Diogo's talk: https://t.co/627kG7eMOh

[MotherDuck @motherduck]

Text classification in MotherDuck just got ~50x faster at ~1% of the cost.

prompt_jev() is a SQL function powered by Jev, TypeSafe's new system one model. 100k rows: 40s, $0.50, frontier-LLM accuracy. The LLM took 32 min and $37.

Read on:

https://t.co/XIE2cw6rUS

[Eugene Shvarts @mathfax]

Our users are getting better at teaching how to use Jev than we are!

[Sydney Runkle @sydneyrunkle]

here's a quick breakdown of @typesafeai's new jev model: what it is, and how to plug it into your agents!

[Eugene Shvarts @mathfax]

@sydneyrunkle @typesafeai What an awesome tutorial! 🙌

[Eugene Shvarts @mathfax]

@deepfates damn feels like a hack though. Of course the distribution of internet text would think a thing thinking to itself is more likely to have qualia

[Eugene Shvarts @mathfax]

@GoSailGlobal Yes we've got tons! Semantic search was my baby https://t.co/CNQpMDDJiC

[Paolo Rosson @redp314]

can @typesafeai Jev count the r's in strawberry?

no.
47% says 3, 47% says 2. a coin flip, same as every LLM.
On 70% on 168 test words, it undercounts doubled letters like everyone

then I gave it the letters instead of the word: ["s","t","r","a","w","b","e","r","r","y"]
168/168. same model, same question, 260ms
@CompleteSkeptic is this expected?

[Eugene Shvarts @mathfax]

One of the most fun parts of this company is @CompleteSkeptic getting to go off on any topic he feels like either nerding out about or shitposting on. Always a terrific listen.

[Latent.Space @latentspacepod]

Jev and the System One Model: RLCD, intelligence/$, reliable AI, & the end of chat-first AI https://t.co/H2bZXCENyW

@typesafeai CEO @CompleteSkeptic explains why AI can solve extraordinarily hard problems yet still fail to automate basic work, why Jev is built for reliable decisions inside software instead of chat, why TypeSafe rejects public benchmarks and refusals at the API layer, why data and the right task matter more than brute-force compute, how System One Models could reshape coding agents and software, and why even with $1 billion he wouldn’t pre-train a model from scratch.

[Eugene Shvarts @mathfax]

@UnCorped @browserbase okay this made me laugh on my commute. Rare Jev pun encountered.

[Eugene Shvarts @mathfax]

I was honestly surprised by the number of absolutely sick computer use demos that hit the TL right after we launched; we know that we're going to dominate in that space, but user creativity always shines brighter than you can predict.

@browserbase was so on this that they hosted a Jev Day hackathon less than a week after our launch, and @kylejeong has a thorough guide on what Jev's for and how to get Jev working for you.

Wait till y'all see what the image model can do 😎

[Kyle Jeong @kylejeong]

https://t.co/MAj6LuRCZ3

[Eugene Shvarts @mathfax]

@Siddharth87 @browserbase Editorial with semantic search goes hard. I already use Jev, e.g., for targeted edits in anti-AI writing patterns across any auto-generated content

[Eugene Shvarts @mathfax]

@huh_go @browserbase You can train it on whatever you want! World's [y]our oyster

[Eugene Shvarts @mathfax]

@tenobrus Reliable decisions was the clearest example of where LLMs totally miss the mark while being super wasteful w their intelligence

[Eugene Shvarts @mathfax]

@tenobrus “Engineers want control again” misses the point; obviously agents can and will write system one calls into code themselves. What’s been missing is appropriate primitives to apply intelligence exactly where and how it’s needed to dramatically improve overall intelligence per $