Sakana's Open Baseline Challenges Discovery Loop

This makes Discovery Loop easier to judge than most frontier AI research products, because Sakana has turned autonomous research from a private demo into a public benchmark. Sakana has open sourced The AI Scientist code, and the Nature paper says the template based version and Automated Reviewer are available on GitHub. That means performance can be compared on concrete tasks like idea generation, code edits, experiment execution, and paper writing, instead of on anecdotes alone.

• Sakana is not just publishing a paper. It has also published code repositories for both The AI Scientist and AI Scientist v2, plus experimental writeups. That gives competitors and researchers a runnable starting point, which is rare in this category and raises the bar for any closed system claiming better automation.
  2 nature 4 github 5 arxiv
• The comparison is especially direct because both systems start in ML research, where experiments are already digital. A user can ask the system to improve benchmark accuracy or reduce latency, then watch it generate variants, run jobs, score outputs, and pick the next branch. That is the same core loop Discovery Loop is building for its first market.
  3 sacra
• Google DeepMind is a different kind of reference point. Co-Scientist is built as a collaborative hypothesis engine around Gemini, search, databases, and tools like AlphaFold. It shows that large labs are also productizing scientific agents, but Sakana provides the cleaner public baseline because its code is openly inspectable and reproducible.
  3 sacra 7 deepmind 8 blog

Going forward, the winners in automated research will be measured less by visionary claims and more by reproducible loops, stronger benchmark gains, and lower cost per useful experiment. Public baselines like Sakana's will push Discovery Loop to prove that its system can search wider, learn faster, and deliver better results than open alternatives.