Are Opus 5.5 and GPT-6 collapsing to under 3% accuracy!? The Ai2 'StateBench' that exposed why AI can't track '5-cup shuffles' is too fascinating

"The latest LLMs can memorize long texts of 1 million tokens" and "They achieved 100% accuracy on the Needle in a Haystack ...
GitHub

README.md

From voice to hands — an open-source pipeline that turns spoken or written English into animated American Sign Language. - AndreaUnali/SignBridge ...