whydeso

AI researcher · agentic builder · future architect
I find the edge so you don't have to

[whydeso @whydeso]

https://t.co/A28UjMYlZO

[whydeso @whydeso]

@monokern sent you a DM bro

[whydeso @whydeso]

SEND THIS PROMPT TO YOUR AI AGENT

it finds every place in your project where you're paying LLM prices for a yes-or-no, tells you what it costs you per month, and moves the most expensive one to Jev.

most agents burn half their budget on decisions that never needed a single generated word.

Run it tonight. Copy this

[whydeso @whydeso]

https://t.co/A28UjMYlZO

[whydeso @whydeso]

@qwinsi0x stale book kills even a great model, you got the point

[whydeso @whydeso]

Every AI trading bot you've seen is answering questions about a market that no longer exists.

An LLM takes 3–30 seconds to think about an order book.

The book moves 40 times in that window.

By the time the model answers, the market it was answering about is gone. that's why nobody could run a real AI trader 24/7.

Jev answers inside a single 300ms block. 6 questions at once: regime, toxic flow, direction, liquidity, quote quality, inventory pressure.

Running it every block, all day, all month: $10–25.

The same loop on a frontier LLM: $50–150 per hour.

for the first time, retail can run the brain of a market maker for less than a Netflix subscription.

[Roan @RohOnChain]

Jev is the FASTEST AI model ever built for trading

It makes calibrated buy/sell decisions in under 100 ms

That is one real decision on every single block, 24/7

In this article I've shown EXACTLY how to build HFT trading system with Jev (from scratch) https://t.co/4hmZDzN00Z

[whydeso @whydeso]

@0xSpivach that’s the point, the loop finally costs less than the latency

[whydeso @whydeso]

@qwinsi0x love that the tool killed a dead strategy instead of selling another curve

[whydeso @whydeso]

I pointed 8 bots and Jev at 3,412 posts across X, LinkedIn and YouTube.

13 seconds later I had a full 7-day content plan. It cost $0.37.

the same reading, done by a human: 6 hours 40 minutes.

here's the part nobody building agents has clocked yet — most of what your AI does isn't writing. it's deciding. which post is worth reading. which model to use. which action is safe to run.

and you're paying frontier-model prices for what is basically a switch statement.

Jev does only the deciding. 200x faster. 400x cheaper. output tokens free.

people already ran 500 emails through it for 3.5 cents, and compacted a Claude session from 1,000,000 tokens to 86,000 in 1 second.

the full build — setup, code, and 5 ways to deploy it — is in the article.

send this to whoever's still running their agents the expensive way.

[whydeso @whydeso]

https://t.co/A28UjMYlZO

[whydeso @whydeso]

STOP PAYING A WRITING MODEL TO TICK BOXES.

You built an agent to do research. Now it calls an LLM to decide whether to open a page, keep a source, pick another worker, or stop. You're paying for all of that before you get the report.

That's the bill I wanted to pull apart in this article.

Jev caught my attention because it gives those small judgments somewhere else to go. The useful question is which calls you can move out of your agent without making the whole thing worse.

So I wrote the full build: account setup, Python code, a decision router, and the handoff back to the model doing the writing. The video sketches the workflow. The article gets into the wiring.

There are examples for browser tasks, inbox triage and choosing which model gets a request. You can start with the part of your agent that keeps making the same expensive judgment.

I also included where this falls apart. Jev can return a perfectly valid answer that's wrong. Giving it permission to act on that answer is your decision.

That deserves as much space as the demos.

The guide covers review thresholds, spending limits and checking that a task actually finished. And when an ordinary rule already solves the problem, the code should keep that job.

Read it with your own agent open. Find the calls you're paying for that never needed a written answer.

You can find all the information in the article.

[whydeso @whydeso]

https://t.co/A28UjMYlZO

[whydeso @whydeso]

@pylok 1b or regex for the routing, save the big model for the actual judgment

[whydeso @whydeso]

https://t.co/A28UjMYlZO

[whydeso @whydeso]

@0xMovez This has been around for a couple of days now, and I only found out about it now, thanks for saving it

[whydeso @whydeso]

the guy who co-invented ChatGPT just dropped a model built for one thing: making decisions in real time

not chatting. deciding

40-400x cheaper, output tokens free, fast enough to run an agent every 300ms block

people are already wiring it into on-chain trading bots that read the price feed and execute trades themselves, block by block, 24/7, for basically nothing

we spent 3 years making AI that talks. Jev makes AI that acts

this is the shift nobody's pricing in yet

[Diogo Almeida @CompleteSkeptic]

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?

I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev

• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions

AFAICT the shortest path to AI-based economic revolution

[whydeso @whydeso]

@RohOnChain jev only judging and the math staying in deterministic code is the right split

[whydeso @whydeso]

I asked GPT-6 Astra to build the entire White House in 3D.

every room. every corridor. the Oval Office, the East Room, all of it — fully explorable, rotate it, fly inside, walk the halls.

1 prompt. zero code. no Blender, no 3D team.

you're watching a $50k architectural model built in minutes.

[whydeso @whydeso]

@AIMind_Ai To be honest, I liked the gemini version more

[whydeso @whydeso]

@thomasgoai That's what I wanted

[whydeso @whydeso]

I gave GPT-6 Astra my Twitter avatar and asked it to build me a 3D game.

what it made shouldn't be possible from one sentence.

a fully playable 3D world, my character, real physics, running at 60fps in my browser — zero code written.

we went from "AI writes text" to "AI builds you a game starring you" faster than anyone noticed.

drop your @ and watch what it makes of you

[whydeso @whydeso]

https://t.co/2yWwDogfSi

[whydeso @whydeso]

@RoundtableSpace beating a 4B from the laptop is the part local people have been waiting on

[whydeso @whydeso]

@gippp69 memory that survives the restart is the whole missing piece

[whydeso @whydeso]

@andreysuperior filing cabinet with no reader was the problem, this loop actually closes it