| Thanks for your comment. We have also tried it ourselves in relbench (Stanford's predictive modelling on relational databases), and surpassed tabular foundational models: https://github.com/Leeroo-AI/kapso/tree/main/benchmarks/relb... Our hypothesis is that similar to RL principle, you can experiment in the environment (predictive modelling on relational databases), learn from it, and compound your domain harness with these bank of lessons, instead of re-training foundational table model, which puts these learnings in model weights. This enables faster iteration, and transparent learning over time. As a proof for this domain, Kapso leads in classification, regression and recommendation tasks vs. best tabular foundational model (Kumo AI/Nvidia): https://star-project.stanford.edu/relbench/leaderboard/#clas... |
| |
| ▲ | granthamctaylor a day ago | parent | next [-] | | Thank you for sharing! I can appreciate surpassing foundation models in RelBench, but I’ve found that peers in our space strongly prefer RelArena because it reruns submitted implementations under shared evaluation protocols. IMHO Prior Labs does a great job of providing a controlled and transparent environment. In RelBench, hypothetically, a solution could mine for days, weeks, or months for specialized, handcrafted features. I’d would be extremely interested in how the cumulative cost of that exhaustive search scales to new, enterprise-scale databases. As per the leaderboards you reference: "The RelBench leaderboard uses self-reported prediction files, so we rely on the authors to submit in good faith ... For a more controlled and transparent evaluation on RelBench tasks, please see RelArena". I would love to see Kapso’s performance in a controlled environment like RelArena, to understand how this approach compares with Relflow and other relational learning methods in an apples-to-apples comparison! | | |
| ▲ | alirezamshi a day ago | parent [-] | | Yeah, I agree with that. We have started Kapso for predictive modelling on relational databases before RelArena came out, so Relbench was the only popular public benchmark. We surely try Kapso on it too. To test it further in live environment (additional to enterprise environment of our customers), we participated with Kapso in AI olympiad this year, and surpassed all human participants, and became top 3 AI systems: https://ioai-official.org/ai-models-track/ioai%c2%b2-2026-re.... In IOAI2, six expert-designed tasks, a single-GPU, and 2 hours budget were given to autonomous agents for deliver solutions without any human intervention. | | |
| ▲ | granthamctaylor a day ago | parent | next [-] | | Woah, now that is extremely impressive. I think Kapso could be used as an _incredible_ resource to AB test internal
architectural changes to complex models with many moving parts ... | |
| ▲ | alirezamshi a day ago | parent | prev [-] | | > Woah, now that is extremely impressive.
I think Kapso could be used as an _incredible_ resource to AB test internal architectural changes to complex models with many moving parts ... I think if a task is iterative and needs experimentation to find the optimal solution, then Kapso can be used there (hopefully helps)! |
|
| |
| ▲ | alirezamshi a day ago | parent | prev [-] | | I am very interested to hear others' perspectives on this topic: Should we integrate experimental learnings by evolving the agent harness, or do RL-based post-training, and leave it in LLM's weights? Or Both? | | |
| ▲ | rraithel27 4 hours ago | parent [-] | | I think evolving the harness makes the most sense. Models get swapped as new ones come along, and anything learned into weights would be lost. Lessons in the harness can carry over. I work in a regulated industry and traceability is extremely important. This includes explaining why the system did something and being able to undo what it learned. A lesson with its evidence attached can be traced and deleted, a weight change not so much. Can you trace a result back to the lessons that produced it? | | |
| ▲ | alirezamshi 3 hours ago | parent [-] | | Great point, it makes even more sense in regulated industry. They can better trust an AI system that transparently writes and executes the learnings rather than being in model weights with no transparency. Regarding your question, Yeah, in our bank of lessons, each lesson card has credence score, boundary (where it works), and full evidence of experiment traces. As you run kapso, these credence score and boundary get better with more evidence from experiments. By any chance, did you encounter any use-cases in these industries, that we can try Kapso? |
|
|
|
