roykishony

4 ms·

1. 1.▲
   Show HN: Revaiso – AI writing in Google Docs as native suggestions(chromewebstore.google.com)

   5 points by roykishony 4mo ago|0 comments
2. 2.▲
   by roykishony 2y ago

   Try it out here: https://github.com/Technion-Kishony-lab/data-to-paper
3. 3.▲
   Show HN: Data-to-paper AI-scientist platform just out in NEJM AI(ai.nejm.org)

   6 points by roykishony 2y ago|2 comments
4. 4.▲
   by roykishony 2y ago

   Thanks so much for these thorough comments. You suggested some directions for more complex analysis that could be done on this data - I would be so curious to see what you get if you could take the time to try out running data-to-paper as a
5. 5.▲
   by roykishony 2y ago

   and yes we are implementing CoT and OPA - but surely there is ton of room for improvements!
6. 6.▲
   by roykishony 2y ago

   thanks. will certainly look deeper into txtai. our project is now open and you are more than welcome to give a hand if you can! yes you are right - it is built completely from scratch. Does have some similarities to other agent packages, bu
7. 7.▲
   by roykishony 2y ago

   wow - thank you for the meticulous check - these are issues we should certainly fix!
8. 8.▲
   by roykishony 2y ago

   Thanks much for these thoughtful comments and ideas. I can’t but fully agree: pre-registered hypothesis is the only way to fully guard against bad science. This in essence is what the FDA is doing for clinical trials too. And btw lowering t
9. 9.▲
   by roykishony 2y ago

   thanks! indeed currently we only provide the LLM with a short tldr created by Semantic Scholar for each paper. Reading the whole thing and extracting and connecting to specific findings and results will be amazing to do. Especially as it ca
10. 10.▲
    by roykishony 2y ago

    Thanks everyone for engagement and discussion. Following the range of comments, just a few thoughts: 1. Traceability, transparency and verifiability. I think the key question for me is not only whether AI can accelerate science, but rather
11. 11.▲
    by roykishony 2y ago

    yes - LLMs tuned based on data science publications will be great. need a dataset of papers with reliable and well-performed analysis. Notably though it works quite well even with the general purpose LLMs. The key was to break the complex
12. 12.▲
    by roykishony 2y ago

    yes that's sounds like the type of data that will be fun to try out with data-to-paper! The repo is now open - you're welcome to give it a try. and happy to hear suggestions for improvements and development directions. data-to-t
13. 13.▲
    Show HN: "data-to-paper" – autonomous stepwise LLM-driven research(github.com)

    139 points by roykishony 2y ago|49 comments
14. 14.▲
    by roykishony 4y ago

    Excited to launch "Quibbler", an open-source Python package for interactive data analysis. Fun to use. Nothing to learn. Your standard code effortlessly comes to life! With the amazing Maor Kern and Maor Kleinberger. https:/
15. 15.▲
    Show HN -- Quibbler: Your Data – Interactive(github.com)

    5 points by roykishony 4y ago|3 comments