From the source

Can LLMs invent better ways to train LLMs?

Research paper

Sakana AI releases a research report on using LLMs to discover preference optimization algorithms, including the state-of-the-art DiscoPOP algorithm, and open-sources the codebase and model checkpoints.

From the source
we’re excited to release our report, Discovering Preference Optimization Algorithms with and for Large Language Models .
sakana.ai

Other stories

Recursive Self-Improvement through Collective Intelligence
A method for models to improve via multi-agent teams
Sakana AI· 11 OctResearch paper
SAIL: Scaling In-Context Imitation Learning
Method for reliable robot trajectory generation via test-time scaling
Sakana AI· 27 SepResearch paper
The Next Frontier: Welcoming AI Pioneer Jürgen Schmidhuber to Sakana AI

Announced: 12 Jun 2024 15:00 UTC
Detected: 28 Aug 2026 21:48 UTC (+807d after)

Visit source ↗sakana.ai

Sakana AI

· 23 Sep

Introducing Sakana AI's Frontier Intelligence Group (FIG)

Sakana AI launches Frontier Intelligence Group to explore alternative AI paradigms beyond scaling.

Sakana AI· 17 Sep

Inside Sakana AI's Product Team

Sakana AI's Product Team converts research technology into products for global users.

Sakana AI· 14 Sep

Training 1000-layer networks without backpropagation

A local-learning method trains 1000-layer networks without backpropagation.

Sakana AI· 13 SepResearch paper

Can LLMs invent better ways to train LLMs?

From the source

Recursive Self-Improvement through Collective Intelligence

SAIL: Scaling In-Context Imitation Learning

The Next Frontier: Welcoming AI Pioneer Jürgen Schmidhuber to Sakana AI

Introducing Sakana AI's Frontier Intelligence Group (FIG)

Inside Sakana AI's Product Team

Training 1000-layer networks without backpropagation