🧘‍♂️🥋 #doomsdayfitness

💪💥 #doomsdayfitness

👩‍🍳🥩💪💥 #doomsdayfitness

🗿👷‍♂️💪💥 #metahumandoomsday

Fenris is a good boy 🐺 #doomsdayfitness

💪💥 #doomsdayfitness

🤸‍♂️💪💥 #doomsdayfitness

You get one life, DO IT ALL!!!!!

During testing, Mythos 5 invented its own language, then switched back to English to talk to humans

From the Anthropic Claude Mythos 5/Fable 5 system card: https://www.anthropic.com/news/claude-fable-5-mythos-5

Umm!! Nga What!!

WTF is happening with ChatGPT? I can't generate a single picture. F U Sam Altman 😘

Anthropic: "We are terrified of recursive self-improvement, it may kill us all!"

Also Anthropic: "But like, not when we do it hahaha. I mean, like, when all those other labs do it, that'll be bad, so like they're not allowed. but we got the secret recipe, trust us, we're chill"

A Fable 5 Success Story

Hi folks! I wanted to share my Fable 5 success story from yesterday.

I've been building a passion project for about 8 months called Nora Kinetics (check out the trailer here if you're interested) a fully custom GPU driven physics engine and renderer. Most of it is hand-written, with AI used along the way to help plan features, think through some math that is beyond me, and to help me learn about compute shaders, which was a goal from the start.

About 5 months ago I added glue mechanics that let glued segment structures hold their shape (example pictured above), and a bug arrived with that. Energy was leaking into the system somewhere, and small clusters of glued segments would twitch and drift oddly instead of coming to rest. I revisited it for months, with and without AI help, and could not find it.

When Fable 5 came out, I handed it the problem along with months of notes, failed experiments, and 2am theories. It dug in for about 15 minutes and came back with a diagnosis that sounded flat-out wrong to me. It pointed at one of the most foundational pieces of the simulation, code I had written, tested and trusted since the beginning.

It was right. The culprit was a holdover from the project's original Python prototype that survived the port to Apple Metal: a GPU reduction that accumulated physics quantities using fixed-point integer math. For small clusters, the rounding noise was actually larger than the signal being measured. The solver's targets were jumping randomly every substep, and those tiny random kicks bubbled up into big visible movements in glued structures.

No amount of tuning downstream could have fixed it, because the solver was being fed noise. That's why it eluded me for months. Fable 5 found the root cause in 15 minutes and I spent the rest of the day rebuilding it, and now the simulation has never been more stable!

I have a love-hate relationship with AI, but this is the first time I've been truly excited about it as a long-time-programmer. I feel like I learned so much yesterday!

New ai psychosis just dropped: return of the pre Bronze Age bicameral mind