MindBase: what it takes to stop an LLM from writing to its own wiki
Every demo of an LLM maintained wiki shows the writing half: sources go in, tidy pages come out. MindBase https://github.com/frankchu91/mindbase llm wiki , an
AI research
Papers, releases and incidents in AI and the hardware it runs on, checked against the primary sources.
The Sabotage Yardstick: Open Models Tie Sonnet on Vigilance, and Serving Weighs More Than Bits
A benchmark you have probably never heard of ran almost seventy complete coding campaigns on almost sixty models between April and October 2026, cost its
An 8GB Laptop Runs a 35B MoE Model at 33 Tokens per Second: the llama.cpp Flag and Its Cliff
One llama.cpp flag turns a 4-token-per-second laptop into a 33-token-per-second one - and a second machine shows where the cliff is.
ETH is false: two AI labs, two Lean proofs, one shared date
Two labs, two Lean-verified proofs: the Exponential Time Hypothesis fell on 26 October. What falls with it, what stands, and where the crossover sits.
Novelty Scores That Measure the Prompt
Ai2 published a benchmark today and the headline number is uncomfortable for anyone who works with judges, including me. Given the same pairs of research