Video yükleniyor...
Video Yüklenemedi
I got a rubik's cube to solve itself with TypeSafe AI 's Jev and it solves it like a person does, 94 moves, not the 22 move optimal solution. Jev isn't an LLM, it just answers one question in ~250ms with a probability. so I put the beginner method... show more
208,053 görüntüleme • 1 ay önce •via X (Twitter)
33 Yorum
@typesafeai Worth noting the 94 vs 22 gap comes from the scaffold, not the model: the beginner method deliberately constrains the solution space. The real design win is that deterministic code validates each probabilistic pick.
@typesafeai Yes totally!
@typesafeai
@typesafeai A bit paradoxical 🤣
@typesafeai Actual AI engineers coming out of the woodwork. Love it!
@typesafeai Thank you! Much appreciated
@typesafeai Now we load Jev into a robot solver and sell it for $50 a pop!
@typesafeai Hahaha gotta do it now 😎
@typesafeai This does not seem any better than using an LLM though. An LLM could have written you the exact code to solve it, and it would have cost you nothing.
@typesafeai This is a different approach altogether! Of course it coukd be done with an llm but less efficiently!
@typesafeai Incredible. I wonder if someone could make a mixed system that would use Jev for 90% of things but delegate to an LLM when it needed to think longer about a decision, kinda like how humans work.
@typesafeai That’s exactly the idea! And how to use it!
@typesafeai Gods number is 20, so every Rubik’s cube no matter the position can be solved in 20 moves or less (meaning the 22 move solution can’t be optimal) :)
@typesafeai Can’t argue with this!
@typesafeai How did you guys get access?
@typesafeai Just signed up to the waitlist
@typesafeai So there is no lookahead / test time compute sort of thing happening here, its a direct general purpose classifier?
@typesafeai that just sounds like coding a rubik's solver with extra [pointless] steps :)
@typesafeai The interesting part is that it uses the same approach a person would now what a computer would do
@typesafeai You gave jev an algorithm and it executed it. That‘s orders of magnitudes slower that compiling the algorithm to machine code and executing it directly.
@typesafeai Superb.
@typesafeai Thank you! It’s a great new tool!
@typesafeai Indeed! 👍
@typesafeai When is this available via api!?!??!
@typesafeai You can sign up to the waitlist. I got access in a day or so
@typesafeai Ty
@typesafeai Can you play a game of chess with Jev?
@typesafeai Definitely!
@typesafeai Really cool stuff. There are many algos already present. Does jev beat them? Recently built this with jev as well
@typesafeai Reminds me of my childhood. So fun.
@typesafeai One of the best childhood activities!!
@typesafeai That's not impressive Im calculating n reductions in the delete path in under 4 minutes - Jev is terrible I looked at it who ever engineered it they're clueless wouldn't even listen to them - this is like sharing an md file and acting like you won.
@typesafeai GLiNER can do the same agentic decision making tasks like Jev while being local, free and open weights. it's so lightweight it'll run on a laptop.
![this harness stack is f*cking gold TypeSafe AI + LangChain finally dropped the layer that makes agents affordable at company scale: jev running inside the langchain agent loop as middleware the problem every big team hits: each fork in the loop is another full LLM call. at 10 agents that's a bill. at 10,000 it's the budget > routing before: every task goes to the frontier model now: jev picks the least costly model that can finish it. lookups go fast, architecture goes powerful > guardrails before: the "is this command dangerous?" check sat in the closed parts of claude code, codex, cursor now: AutoModeMiddleware(tools=["bash"]). one line, any agent, blocked before the tool runs > decisions before: one LLM call per question now: every question about the same state in one request, answered in parallel > audit before: "the model decided" now: the answer, the probability and the confidence stay in agent state > adoption no rewrite. pip install langchain-typesafe and drop it into the agent you already run TypeSafe reports up to 200x faster inference and 400x lower cost than comparable LLMs on classification. the LLM still reasons and writes. jev takes the thousands of small calls in between already live: browserbase agents for fractions of a cent, a live trading agent, email triage at scale full loop mapped below: 7 agents, 4 jev gates per step count the forks in your agent loop tonigh](https://image.24vids.com/tw-2104876056893104461/amplify_video_thumb/2104866254146203648/img/J8LnvuFvwbGYaJ7N.jpg)
