A team from MIT built a model that scores 61.9% on ARC-AGI-PUB using an 8B LLM plus Test-Time-Training (TTT). Previous record was 42%. (via Localllama)
Global AI News Aggregator
By
–
A team from MIT built a model that scores 61.9% on ARC-AGI-PUB using an 8B LLM plus Test-Time-Training (TTT). Previous record was 42%. (via Localllama)
Leave a Reply