Levin Ronskii

Click to edit

[Levin Ronskii @xronsk]

I notice in myself a preemptive feeling of ‘if the project doesn’t launch, whatever’ before the start.

Like during development I look like Schwarzenegger at the gym, but during the presentation I’m a shy girl going ‘you probably won’t like it, but…’

This is bad.

[Levin Ronskii @xronsk]

my tweet went viral (dozens of views, 1 like) —

another reminder to believe in yourself and never give up 🙏

[Levin Ronskii @xronsk]

Beauty is in the eye of the beholder. Left: astigmatism; right: healthy eyes. So it’s not that I’m fat—it’s that your eyes are bad

(le chaton fat)

[Levin Ronskii @xronsk]

@openai please add a checkbox to NOT include general user memory / RAG retrieval of past conversations into the current chat

this would make the chat experience 100% more convenient, because ChatGPT’s judgment is sometimes polluted by context I never asked for

[Levin Ronskii @xronsk]

@marclou Hey Marc. Can organic growth of $66 ARR over six months be considered validation for a lightweight project (a Chrome extension)? Is there any point in continuing it with such a small ARR?

[Levin Ronskii @xronsk]

@airkatakana Beauty is in the eye of the beholder. Left: astigmatism; right: healthy eyes. So maybe it’s not that I’m fat—it’s that your eyes are bad

[Levin Ronskii @xronsk]

Beauty is in the eye of the beholder. Left: astigmatism; right: healthy eyes. So it’s not that I’m fat—it’s that your eyes are bad

(le chaton fat)

[Air Katakana @airkatakana]

i met a very attractive korean girl for dinner tonight

last time i saw her was 6 months ago

when we sat down she immediately said "your got more muscles" and "your skin looks much better" etc

only been looksmaxxing for a month

you can do it too anon

[Levin Ronskii @xronsk]

@airkatakana Beauty is in the eye of the beholder. Left: astigmatism; right: healthy eyes. So it’s not that I’m fat—it’s that your eyes are bad

(le chaton fat)

[Levin Ronskii @xronsk]

@airkatakana Beauty is in the eye of the beholder.

Left: astigmatism; right: healthy eyes. So it’s not that I’m fat—it’s that your eyes are bad

[Levin Ronskii @xronsk]

@KateriSeraphina i thought this was a post about Le Chaton Fat

[Levin Ronskii @xronsk]

my source sent me some photos

they’re saying the employees are in a panic — they’ll have to rebuild their entire workflow because le chaton fat

[Peter Gostev (SF: 22-26 June) @petergostev]

Le Chaton Fat: We have requested an urgent authorisation from the French government to extend the axis to display the result on the Agent Arena

[Levin Ronskii @xronsk]

this is a good time to think about your life and reconsider everything you had planned

[Henok @henokcrypto]

After Fable 5 nothing in Ai feels the same anymore

[Levin Ronskii @xronsk]

can you call yourself an ultrasound specialist if the ultrasound machine does all the work

[Sarah @araseb_]

Can you call yourself a founder if your entire product was built by Claude?

[Levin Ronskii @xronsk]

maybe haiku handles its tasks just fine at 100%

or maybe an upgraded cheap haiku would make sonnet obsolete

[Haider. @haider1]

anthropic ignoring "haiku" models this hard is beyond ridiculous

haiku is a subagent powerhouse and the most important model for cost efficiency in subagents

it's criminal that anthropic hasn't updated haiku since 4.5, while openai has gpt-5.4 nano, which is cheaper and better

[Levin Ronskii @xronsk]

and maybe when ASI is achieved, it’s only a matter of time before big AI gains more power than governments

[Ethan Mollick @emollick]

If AGI is achievable & labs can be banned from using a model internally ONLY if they release the model publicly, the Big Three labs may decide it is better to capture all the value from AGI themselves by expansion & acquisition. Sharing AI access with other firms triggers risk.

[Levin Ronskii @xronsk]

was this a surprise to anyone?

anthropic has said 100 times that tools matter just as much as raw intelligence.

did you really expect a tech company full of TOP engineers NOT to build software dev tooling for their own AI?

it’s just so dumb that anyone finds this “wow” worthy

[How To Prompt @HowToPrompt__]

Researchers show that Claude Code is 98% not AI.

Anthropic never gave us the architecture for Claude Code. There were no docs. Just a tool that every developer is currently obsessing over.

Until it leaked recently.

A research team pulled the source code, analyzed all 500,000 lines, and found something ridiculous.

Only 1.6% of the codebase actually interacts with the AI model.

The core of Claude Code is literally just a simple while-loop. It asks the model what to do, runs a tool, and repeats.

So what is the other 98.4%?

It is hardcore, traditional software engineering.

The researchers found a massive, complex infrastructure designed entirely to babysit the AI and keep it from hallucinating or destroying your computer:

- A 7-mode permission system acting as a security bouncer.
- A 5-layer context compaction pipeline so the AI doesn't forget its goal.
- A subagent delegation mechanism with strict worktree isolation.
- Four different extensibility hooks to manage external tools safely.

Every startup right now is trying to build a better AI model to get better results.

Anthropic did the exact opposite.

They took an existing model and built a fortress of deterministic software around it.

They realized that the AI doesn't need to be smarter. It needs to be managed.

[Levin Ronskii @xronsk]

good posture is the cheapest way to boost your brain performance by 5-10% through improved blood flow. as you know from LLMs, a 5% difference in intelligence can feel like 500%

[Levin Ronskii @xronsk]

lol, you're not using Haiku right. You just need to give it a set of smoke and e2e tests, the ability to call gpt xhigh, write a detailed prompt, and list the errors it made. Basically, you can do all of this yourself and just give him a script that will insert.

[Da7em @Da7_Tech]

The real trick with Opus 4.8 is not to ask it to “do the task.”

That is exactly why some people mistakenly think Codex 5.5 is smarter.

For simple execution — “do this,” “build that,” “fix this” — Codex is often faster, cleaner, and more direct.

But Opus has another gear entirely.

Don’t just give it a task. Give it a quality system.

Example:

“Complete task X, but do not deliver the final result until you verify it through three different review methods, and until an Opus-based reviewer confirms it meets these project standards: A, B, and C.”

This is where Opus separates itself.

The same methodology does not work nearly as well with Codex 5.5.

Codex is strong as an executor, but its agent coordination is weak. When you try to make it work like a multi-agent review system, the final output still feels like it came from one mind, one lane, one narrow execution path.

Opus is different.

Inside Claude, when prompted this way, it can actually feel like a council of experts: builder, critic, strategist, reviewer, and quality gatekeeper all pushing the work from different angles.

That is why the result often has better judgment, less hallucination, more depth, and a level of polish Codex 5.5 rarely reaches through raw execution alone.

One warning: don’t use this casually unless you’re on the $200 Max plan.

This method is powerful, but it burns usage fast.

[Levin Ronskii @xronsk]

And it’s research-only, right? No agentic coding, no creative writing, no conversation, no SWE-bench?

[OpenRouter @OpenRouter]

Introducing the Fusion API, the smartest compound model in the market.

Fusion achieves Fable-level intelligence at half the price.

How it works 👇

[Levin Ronskii @xronsk]

Even if the “US-only” scheme works, it’s hard to imagine 100+ gray-market APIs reselling Fable at a 1% markup won’t pop up within a day.

[Levin Ronskii @xronsk]

That's wrong. She should teach you just how important money is in our world

[Matthew Berman @MatthewBerman]

This entire situation should wake you up to how important open source models are

Get your own model running locally, go into debt if you have to