AI News 2025-06-26

General

Research Insights

  • Dense SAE Latents Are Features, Not Bugs. They find pairs of opposing features that fire very frequently. Far from being useless, they find these encode meaningful concepts.
  • Sakana AI: Reinforcement Learning Teachers of Test Time Scaling (preprint). Rather than using to RL to improve solution, they focus on improving the ability for the model to teach other models. RL reward is based on how useful generated training examples are (to smaller learning models); rather than being rewarded on correctness of their final answer.

LLM

Agents

Audio

  • ElevenLabs introduces 11ai, a voice conversational assistant; exploits MCP to enable connection to resources (calendar, etc.).
  • ElevenLabs introduces Voice Design v3, an improvement to their text-to-voice system for designing a voice.

Image Synthesis

Video

World Synthesis

Science

  • Google DeepMind releases AlphaGenome (including API capabilities); it takes base-pair sequences as input, and predicts genomic behavior outputs.

Robots

This entry was posted in AI, News and tagged , , , , , , , , . Bookmark the permalink.

Leave a Reply