R to @hardmaru: To give some more context on what we are building with Daiwa Securities: During our…
Ranking
No observed public metrics; popularity remains neutral/archived.
Merged summary
TL;DR - Sakana AI (via co-founder David Ha) details a partnership with Daiwa Securities moving from technical verification into full deployment, applying its AI Scientist and AB-MCTS agent frameworks to automate market-data gathering and analysis for Daiwa's wealth management division. It's a concrete case of research-grade agentic search methods being productionized in regulated financial services.
- The stack combines Sakana's AI Scientist (autonomous research/analysis pipeline) with AB-MCTS (adaptive branching Monte Carlo tree search for inference-time scaling) rather than a single-model deployment.
- Verification phase focused narrowly on rigorous, at-scale gathering and analysis of complex market information — not trading or advice generation.
- Analysis quality reportedly improves continuously by incorporating direct end-user (consultant) feedback, i.e. a human-in-the-loop refinement loop.
- Stated goal is human-AI collaboration: offloading data-processing "heavy lifting" so financial consultants can focus on personalized client advice.
- Note: this is a company announcement thread; no benchmarks, metrics, or evaluation results are provided.
Sources (1)
R to @hardmaru: To give some more context on what we are building with Daiwa Securities: During our…
TL;DR - Sakana AI (via co-founder David Ha) details a partnership with Daiwa Securities moving from technical verification into full deployment, applying its AI Scientist and AB-MCTS agent frameworks to automate market-data gathering and analysis for Daiwa's wealth management division. It's a concrete case of research-grade agentic search methods being productionized in regulated financial services.
- The stack combines Sakana's AI Scientist (autonomous research/analysis pipeline) with AB-MCTS (adaptive branching Monte Carlo tree search for inference-time scaling) rather than a single-model deployment.
- Verification phase focused narrowly on rigorous, at-scale gathering and analysis of complex market information — not trading or advice generation.
- Analysis quality reportedly improves continuously by incorporating direct end-user (consultant) feedback, i.e. a human-in-the-loop refinement loop.
- Stated goal is human-AI collaboration: offloading data-processing "heavy lifting" so financial consultants can focus on personalized client advice.
- Note: this is a company announcement thread; no benchmarks, metrics, or evaluation results are provided.