Michael Kaiser

Founder & AI Engineer building with AI in production. Daily AI news, filtered down to what matters: models, tooling, benchmarks.

[Michael Kaiser @MichaelKaiserAI]

Day 8 of building in public.

Crossed the 100-visitor milestone, but still zero kings crowned. Everyone is comfortably staying on the Wall of Broke.

27 visitors today (107 total). Total revenue: €0.

Until someone actually pays.

[Michael Kaiser @MichaelKaiserAI]

Best explanation of the new model ‘Jev’

[Michael Kaiser @MichaelKaiserAI]

@codermatt @cognition You probably used SWE-2 in the Cloud? Its only free when using Devin Local

[Michael Kaiser @MichaelKaiserAI]

@solopribuilds Yes

[Michael Kaiser @MichaelKaiserAI]

@TTrimoreau No

[Michael Kaiser @MichaelKaiserAI]

@LeoBuilds_ Time to give Devin a try🫪

[Michael Kaiser @MichaelKaiserAI]

@Ananth7e Won‘t the goblin try to sneak in?

[Michael Kaiser @MichaelKaiserAI]

What it's actually for: sorting, routing, scoring, flagging. Anything where the answer is one of a few options you already know.

What it's not for: writing. No text, no code, no replies. That stays with ChatGPT or Claude.

So two different tools, not a replacement

[Michael Kaiser @MichaelKaiserAI]

TypeSafe AI, founded by one of the researchers behind ChatGPT, released a model called Jev yesterday. It tells you honestly how sure it is.
Normal AI sorts your support mail and sounds certain even when wrong. Jev says "return, 94% sure".
So the clear ones run alone.

[Michael Kaiser @MichaelKaiserAI]

TypeSafe AI, founded by one of the researchers behind ChatGPT, released a model called Jev yesterday. It tells you honestly how sure it is.
Normal AI sorts your support mail and sounds certain even when wrong. Jev says "return, 94% sure".
So the clear ones run alone.

[Michael Kaiser @MichaelKaiserAI]

@CompleteSkeptic Thanks, I’ll join the waitlist.🤝

[Michael Kaiser @MichaelKaiserAI]

Huge claims if true. Where can developers test the API or weights to see it in action?

[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

[Michael Kaiser @MichaelKaiserAI]

@Da7_Tech 😂😂

[Fandu @mrfanduu]

Gemini 3.8 live & Gemini 3.8 live extended thinking are out 🔥

[Michael Kaiser @MichaelKaiserAI]

@notjazii 😂😂

[Michael Kaiser @MichaelKaiserAI]

Claim the throne or stay on the wall: https://t.co/lGbEiyY0n8

[Michael Kaiser @MichaelKaiserAI]

Day 7 of building in public.

One full week in. 80 people have visited, but zero kings crowned. The Wall of Broke remains 100% undefeated.

11 visitors today (80 total). Total revenue: €0.

Until someone actually pays.

[Michael Kaiser @MichaelKaiserAI]

Considering that it’s currently completely free with all paid Devin plans, it’s basically a must-try for every developer.
The model’s speed can fluctuate from time to time, but usually within a few minutes it’s back to normal. I always use it with Effort set to Max.

[Michael Kaiser @MichaelKaiserAI]

SWE-2 by Cognition is probably one of the most underrated models right now. Over the past few days, I’ve tested SWE-2 across a wide range of areas: frontend, backend, general knowledge, and software architecture, and I have to say, it’s outstanding.

[Michael Kaiser @MichaelKaiserAI]

Turns out I was right about this. More and more sources are pointing to a covert test of Opus 5.2 in Claude Code.
It now looks increasingly likely that there won’t be an Opus 5.1 at all, Anthropic may be jumping straight to Opus 5.2.

[Michael Kaiser @MichaelKaiserAI]

I can confirm this. Since yesterday, Opus 5 with Effort Max in Claude Code has been using an insane amount more thinking tokens, enough that I immediately noticed something felt off.
Anthropic is definitely testing something. The question now is whether it’s Opus 5.1 or Opus 5.2.

[Michael Kaiser @MichaelKaiserAI]

thats pure evil from Anthropic