🔥🔥

Analyzing the Hierarchical Reasoning Model by Guan Wang We verified scores on hidden tasks, ran ablations, and found that performance comes from an unexpected source ARC-AGI Semi Private Scores: * ARC-AGI-1: 32% * ARC-AGI-2: 2% Our 4 findings:

Not just Anthropic or Grok. It’ll take several players including myself 😎🔥

We are not just living in interesting times, we are shaping the destiny of intelligence itself. 🫡Elon Musk

I’ve never been more confident in AGI than I am today.

Couldn't resist. Here's a pure PyTorch from-scratch re-implementation of Gemma 3 270M in a Jupyter Notebook (uses about 1.49 GB RAM): github.com/rasbt/LLMs-fro…

Computing power per region over time The USA dominates in AI compute power by a huge margin.

When there’s no big AI news, it usually means someone is about to drop a bomb.

Reliable sources say DeepSeek ran into a serious bottleneck. The weak links? MLA and NSA. CC: Chubby♨️ NIK Lisan al Gaib

FAFO

DeepSeek V4 didn’t make it. Blame MLA and NSA.

Like I said, DeepSeek V4 didn’t make it. The weak links are MLA and NSA.

Insider chatter: DeepSeek’s progress is stuck. MLA doesn’t scale, and NSA is broken.

👀

DeepSeek trains on Nvidia GPUs, but they’re cooked, because MLA and NSA are dead ends.

MoE layers can be really slow. When training our coding models Cursor, they ate up 27–53% of training time. So we completely rebuilt it at the kernel level and transitioned to MXFP8. The result: 3.5x faster MoE layer and 1.5x end-to-end training speedup. We believe our

Nothing new, dead end.

Went to stats department as professor. Classic acceleration to irrelevance.

Cracker Barrel has become a joke. Food quality has cratered over the years and the steak tastes awful. Not touching that place again until they fix it.

Solana🔥