Paul Ojo

Senior Full Stack Engineer | I build and ship scalable web and mobile products | Next.js · Node.js · PostgreSQL

[Paul Ojo @PaulOjo312]

@ShamsAmin56 @videoduck_ai Wow, I love this, this is mindblowing!

[Paul Ojo @PaulOjo312]

@grok That's a nice pick

[Paul Ojo @PaulOjo312]

Hey @grok, Grok 4.7 or Claude Opus 5.5?

[Paul Ojo @PaulOjo312]

@bijanbowen @Everlier Your laughter is always contagious 😂

[Paul Ojo @PaulOjo312]

This is incredible!
Same price as 4.6High

[SpaceXAI @SpaceXAI]

Grok 4.7 is here.

It's a notable improvement over Grok 4.6 at the same price and speed.

[Paul Ojo @PaulOjo312]

@myxx_190 Congrat! 🎉🎉🎉

Btw, I shot you a DM about three weeks ago. You seems to have missed it

[Paul Ojo @PaulOjo312]

@julianfaceless Thanks for the insight Julian
However, my AdSense got disabled in the past, I got misled by a course. Does getting a new phone automatically fix this for AdSense?

[Paul Ojo @PaulOjo312]

@KnowPilot @elonmusk It's an AI generated video

[Paul Ojo @PaulOjo312]

@BiggMuna Pls can you check DM

[Paul Ojo @PaulOjo312]

@BiggMuna This is awesome, congrat👏

[Paul Ojo @PaulOjo312]

@CompleteSkeptic @4rcherhume Jev is a great name

[Paul Ojo @PaulOjo312]

@BiggMuna Congrat Alaoma!

Btw, shot you a DM.

[Paul Ojo @PaulOjo312]

I'm in!

[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

[Paul Ojo @PaulOjo312]

What Jev from @typesafeai is. ELI5 🧵

Jev is not really your usual AI chatbot (Claude, ChatGPT, Gemini, Kimi). Think of it more like an AI that takes quizzes.

You give it some information and a set of questions, then it picks the answers. There is no essay, no long explanation, just the answers and how confident it is about them.

Imagine giving a really smart kid a test like this:

=> Is this email urgent? Yes, 99%

=> Which team should handle it? Technical, 84%

=> How frustrated is the customer? 7/10

The kid does not need to write an essay. It just picks the answers.

That is basically what Jev is built for.

You give Jev some information. This information is called State. It could be an email, a support ticket, a game screen, documents, Wikipedia links, or pretty much any other context.

Then you give it the questions you want answered.

There are three basic types:

🡆 Noul: “Is this true?” => a number between 0 and 1

🡆 Choice: “Pick one of these options” => the answer plus probabilities for each option

🡆 Score: “Rate this using my criteria” => a score plus its confidence

You also define the possible answers yourself.

If you give Jev these options:

* Technical
* Billing
* Sales

it cannot suddenly decide that the answer is “Pizza Department.”

It has to choose from the options you gave it.

You can also ask many questions about the same situation at once. Give it one customer email and ask whether it is urgent, which team should handle it, how frustrated the customer is, whether a refund is being requested, and whether a human needs to review it.

Jev can answer all of them at once and give you the results in a structured format.

The confidence numbers are important too.

Jev is trained using RLCD (Reinforcement Learning for Calibrated Decisions). The goal is for its confidence to actually mean something.

So if Jev says 90% confidence, the idea is that answers given that confidence level should be correct roughly 90% of the time across many cases. It does not mean every individual 90% answer will be correct.

So why is it fast and cheap?

A normal LLM spends a lot of computation generating text token by token. Ask a normal LLM for a 500 word explanation and it has to generate all those words.

Jev does not need to write the explanation. It looks at the context and gives you the decisions you asked for in a structured format.

That makes it useful for applications where you need to make lots of decisions from the same piece of information.

There is a tradeoff, though.

Jev is not replacing your normal LLM. If you want an email, poem, explanation or detailed conversation, you still need a generative AI model.

But if you already know the questions you want answered and just need the model to make lots of decisions quickly, that is where Jev gets interesting.

A normal LLM is great when you want it to generate something. Jev is best option for when you need to make a lot of decisions.

[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

[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

[Paul Ojo @PaulOjo312]

@upgradeoptimism @Nikkau @cramforce @typesafeai Same here, it's mind blowing

[Paul Ojo @PaulOjo312]

@Nikkau @cramforce @typesafeai Wow!, just wow
That's 4.2 cents per million tokens.

[Paul Ojo @PaulOjo312]

@CompleteSkeptic I can't wait to have this 🫢

[Paul Ojo @PaulOjo312]

THIS IS HUGE! LET"S GO! 🔥🔥🔥

[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