AI Model Collapse Is Happening: Treat Data as Code Now
If you’ve asked an LLM for a simple command lately and watched it flail through three wrong answers, you’ve already…
-
-
RBF Attention Reveals Dot‑Product’s Hidden Norm Bias
Swapping dot‑product attention for RBF attention sounds like an architectural revolution. In Raphael Pisoni’s experiment, it turned out to be…
-
Claude vs ChatGPT: Why Claude Feels More Honest and Accurate
A 100‑question “bullshit benchmark” sounds like a joke until you see the chart. In BullshitBench v2, Anthropic’s Claude models sit…
-
Rebuttal Experiments Are Breaking Peer Review Right Now
A lot of people in AI quietly agree on one thing about rebuttal experiments: they make their papers better. More…
-
Voxtral TTS: Mistral’s Open Model, Hype vs Hardware
Imagine you ship a voice agent that talks to customers all day, and then your TTS provider changes their pricing,…
-
Local LLM Coding: $500 GPU Beats Claude: Not the Story
A frozen 14B Qwen model, quantized and running on a single RTX 5060 Ti, scores 74.6% pass@1 on LiveCodeBench after…
-
Karpathy Autoresearch: 700 Experiments Rewire AI Research
If you tried to copy Karpathy autoresearch this weekend, the first thing you’d hit isn’t the 630 lines of Python….
-
Persona Drift: Why LLMs Go Insane Under Repetition
A model gets pinged every few seconds for the time. Nothing else. After enough rounds, it starts acting “fed up,”…
-
The NZ Town Subsidising Silicon Valley’s AI Boom
A farmer outside Invercargill stands at a fence line and tries to picture it: the paddock across the road, not…
