AI Briefing#transformer-architecture
The latest AI and developer news about #transformer-architecture, with the original source and a short summary.
The latest AI and developer news about #transformer-architecture, with the original source and a short summary.
Using modified training protocols on Qwen3.6-35B-A3B and gpt-oss-20b, the research demonstrates that models can perform serially dependent reasoning steps without visible chain-of-thought output.
Microsoft Research scientists Jennifer Neville and Chad Atalla discuss how AI models degrade in multi-turn interactions despite high single-turn benchmark scores, highlighting the need for new evaluation methods and user strategies.
The Recurrent Looped Transformer (RLT) architecture has been released, featuring temporal paths that expand with sequence length.
LG AI Research unveiled EXAONE 3.0 7.8B, an open-source model with excellent English and Korean performance.
LG AI Research has open-sourced EXAONE 3.0 7.8B, an expert-grade AI model with outstanding English and Korean performance.
A new study finds that LLM alignment goes beyond simple filtering, actually restructuring the model's representational space itself.
A new study finds that safety and sycophancy behaviors in LLMs are concentrated in the later layers of the model.
A study reveals that the Engram module functions as a simple architectural side-pathway rather than as memory retrieval in image generation.