Tomas Klohna

neurotypicality doesn’t sit right with me | computering @ aisle | prev pushing paper @ google

[Jürgen Schmidhuber @SchmidhuberAI]

2026 update: Canada adopted the old motto "AI For All" or "AI∀" of our AI company NNAISENSE, founded in 2014 😀 See Prime Minister @MarkJCarney's post: https://t.co/gjhxrNSsEH

[Mitchell Hashimoto @mitchellh]

Fable is a good model. As with all new models, it is simultaneously excellent and entirely unremarkable (relative to other models). It is slow and expensive, and the "loops are all you need" discourse they are pushing is obvious in the context of someone using Fable-class models

What I've found so far is that for broad scope design (code architecture) tasks, Fable is unremarkable. Or, not better enough to justify its cost and speed.

But in highly targeted goal-oriented loops, it is another beast entirely. It is very slow but produces very good results.

I let it churn on optimizing a SwiftUI-layout resolver in Go I wrote and it was able to bring it down to an order of magnitude I could not reach myself (micro => nanosecond scale). But it took 2 hours and $40 to do it and I had to claw back some changes it overfit to Apple Silicon. Still, very worth it.

In comparison, for "implement this feature/change" iterative work, I ran head-to-head Fable vs GPT5.5 vs. GLM-5.1. They all produced equally acceptable final results, but GPT5/GLM did it in a couple minutes and Fable was churning away for 40 minutes. And GLM cost me less than a dollar, GPT5.5 ~$1.50, and Fable cost $9.

You can see that in this context, interactively working with an agent is nonsense. Its too slow. You need to write loops to keep the agent working and you probably want to highly parallelize the work being done. As with all things, I think a balance makes sense...

My sense is that I'd reserve Fable for targeted, surgical analysis and work. Not for daily driving everyday tasks.

I'm going to keep spending a shitload of money (relatively) and maining Fable for the rest of the week to continue to judge, will report if anything changes. I'll continue to head-to-head as well.

[antirez @antirez]

So, what happened to use lately? Especially in Europe. In the US there are those few unicorns but where is all the rest of the AI scene? We need to recover our industrial ethics and stop accepting a narration that see ourselves boiled.

[Minh Nhat Nguyen @menhguin]

glad to know Mythos' safety concerns have been addressed right as Anthropic also secured tens of billions in inference compute 👍

[Tomas Klohna @coffee_puck]

@mitsuhiko it got a loud what the fuck from me

[Tomas Klohna @coffee_puck]

@mitsuhiko half asleep, i looked at it during a shower and thought the display is dying

[Tomas Klohna @coffee_puck]

@tomcupr @VaclavSlajs @matesola @MKecera Largely skill issue tady…

Pouzivam to uz delsi dobu a stacilo zjistit, ze se nemusi pouzivat ty headery ale umi to i OAuth. Normies to pak umi nastavit jak v UI Claude tak i ChatGPT.

Funguje to vyborne, diky za to.

[Sam Altman @sama]

https://t.co/7Am81lOEP3

[Sterling Crispin 🕊️ @sterlingcrispin]

This is the dark reality of data center water usage and they don’t want you to know the truth

[Justine Tunney @jartine]

@NielsHoven What we need is a culture that values doing the work. Instead we have a culture that values administering the work. So the system makes everyone an administrator, while the people who did work feel unvalued and retire.

[Tomas Klohna @coffee_puck]

@halvarflake why? the harnesses and prompts aren’t comparable.
the total with severities is better metric in my eyes since it doesn’t matter how you got there.

[Peter Wildeford🇺🇸🚀 @peterwildeford]

In a debate between Jensen Huang and Jensen Huang, who will win? It's honestly hard to tell...

[Tomas Klohna @coffee_puck]

why would you think the enshittification will skip models, huh?

[Stanislav Fort @stanislavfort]

We pointed the same scanner with just GPT-5.4-NANO in it at the full FreeBSD kernel (>10k files)

Result: <$100 spent + several new kernel bugs + a potential 20+ year-old memory safety issue now under investigation by the security team

Found by the cheapest model available.

[Jeremy Osih @dinnaiii]

introducing https://t.co/92YeKAD5kM

replace "hub" with "inspect" on any github url and chat with any repo.

- pi agent that runs in your browser.
- just-bash with a VFS built on top of the github API
- everything stays local.

turns out pi + just-bash is all you need.
cc @badlogicgames @cramforce

[M @mh012012]

@ClementDelangue For anyone who missed this part deep in Anthropic’s 200 page model card: Their harness prompted Mythos separately for each file. The harness design is similar. And Anthropic to my eyes never tested whether this harness with Opus would find the same bugs.

[Tomas Klohna @coffee_puck]

@badlogicgames @nicopreme @huggingface i like it, but the ampcode thread sharing has good ux worth copying.
idk maybe cf would sponsor the hosting?

[Tomas Klohna @coffee_puck]

@bryan_johnson Did you drop RLT?

[Tomas Klohna @coffee_puck]

@steveruizok @pencildev you will have more luck with @tomkrcha directly

[Tomas Klohna @coffee_puck]

@thruxo @TylerGlaiel who tf do you think owns ltt