Ramakrishna

It's happening!

[Chubby♨️ @kimmonismus]

Claude Fable 5 has produced a hand-checkable counterexample to the Jacobian conjecture, an open problem dating to 1939.

The conjecture says that a polynomial map with a constant non-zero Jacobian determinant must have a polynomial inverse.

An 87-year-old problem was casually solved by Fable 5 on a sunday evening. Insane times to live in.

[Kimi.ai @Kimi_Moonshot]

🚀 Hello, Kimi K2 Thinking!
The Open-Source Thinking Agent Model is here.

🔹 SOTA on HLE (44.9%) and BrowseComp (60.2%)
🔹 Executes up to 200 – 300 sequential tool calls without human interference
🔹 Excels in reasoning, agentic search, and coding
🔹 256K context window

Built as a thinking agent, K2 Thinking marks our latest efforts in test-time scaling — scaling both thinking tokens and tool-calling turns.

K2 Thinking is now live on https://t.co/YutVbwktG0 in chat mode, with full agentic mode coming soon. It is also accessible via API.

🔌 API is live: https://t.co/EOZkbOwCN4
🔗 Tech blog: https://t.co/n7xxaszqzF
🔗 Weights & code: https://t.co/4ukcXB0iP6

[Ramakrishna @heremytatersat]

@divya_venn Maybe an advantage in early career but breaking glass ceiling is harder when not taken seriously. Potential liability to tie professional success to relationships.

[Ramakrishna @heremytatersat]

@iScienceLuvr @prajdabre1

[Jordan Mechner @jmechner]

In 1985, I asked my brother David (age 15) to be the rotoscope model for my new game, Prince of Persia.

38 years later, I've made him a (present day) cartoon character in my new graphic novel memoir REPLAY.
https://t.co/bsLWplOdYd

Thanks, bro!

@01FirstSecond @stripepress @princeofpersia

[Taelin @VictorTaelin]

Seriously - this is great. I can't overstate how good it is. I spent a LONG time to get a half-decent run with Opus back then. Other models could barely draw a frame. GPT-4o just... plays the game. This is Pokémon Red. In a terminal. To details. It remembers everything I do, it gets the maps right, it emulates the battles accurately. It is not good. It is greatness. Humanity is headed towards a beautiful place. I'd not be underwhelmed if this was called GPT-5. Good job @OpenAI. You're a good company.

(Err I hope Nintendo doesn't sue you.)

[Ramakrishna @heremytatersat]

@epicnephrin_e @Anindit27658436 You didn't have to prove that it was real. 😭

[Simon Willison @simonw]

Wow there's a lot of clever technical details in how Cloudflare's new Python support works - running Python in WebAssembly in server-side Pyodide with a whole bunch of performance optimizations

[Dreaming Tulpa 🥓👑 @dreamingtulpa]

if ("OpenAI" == "ClosedAI" && "StabilityAI" == "UnstableAI") {
return "Irony++";
}

[Aviral Kumar @aviral_kumar2]

Super simple code change to get value-based deep RL scale *much* better w/ big models across the board on Atari games, robotic manipulation w/ transformers, LLM + text games, & even Chess!

Just use classification loss (i.e., cross entropy), not MSE!!

https://t.co/0IHSgN4pBj🧵⬇️

[bitbutter @mormo_music]

had to open an invisible door for them

[cts🌸 @gf_256]

Everyone knows Tupper's "self-referential" formula, one that draws its own formula.

That formula is from a 2001 SIGGRAPH paper.

The lore: Tupper is doing king shit, demoscene but in a graphing calculator that he invented

Thread of more sick formulas from the paper:

[chess24 @chess24com]

Mickey Adams says winning the #LondonChessClassic at the age of 52 may be his best ever result, "because other tournaments that I won were when I was in my prime as a player, and it’s completely different now". Final report:
https://t.co/mzv5Wt5cu7

[Trung Phan @TrungTPhan]

Taylor Swift’s cardio training routine in the 6 months prior to the Eras Tour was very legit

[Ben Recht @beenwrekt]

I live-blogged my course this semester. It was the most rewarding experiment I’ve tried in my teaching. Would do again. 10/10.

[Horace He @cHHillee]

I reverse-engineered AlphaCode2's submission history and manually performed the Codeforces evals.

I'm ... again concerned that data leakage is affecting the results.

For the DP problem highlighted in the AlphaCode2 release, look at AC2's solution vs. the tutorial.

(1/5)

[Brian Roemmele @BrianRoemmele]

For example: Because the Gemini Ultra model was trained deeply on YouTube data it can extrapolate a series of static images from a scene in a video (The Matrix) and write a textual narrative from it.

I tested on ChatGPT-4 Turbo and no it could not reason this output.

[BBC News (World) @BBCWorld]

The Indian siblings taking the chess world by storm https://t.co/C5Cz3RapEc