Hacker News new | past | comments | ask | show | jobs | submit | datascientist's comments login


Alignment is hard: "If the LLM has finite probability of exhibiting negative behavior, there exists a prompt for which the LLM will exhibit negative behavior with probability 1." Source: Fundamental Limitations of Alignment in LLMs https://arxiv.org/abs/2304.11082


Creators of the data quality tool for computer vision, fastdup, continue to improve on their free release https://github.com/visual-layer/fastdup

Here's a short video of some recent results for LAION 400M https://www.youtube.com/watch?v=dlRCm29Upu4


Recent perspectives from the creators of Prefect, Dagster, Flyte, and Orchest => https://gradientflow.com/summer-of-orchestation/


Seems to be missing Temporal/Cadence, which I'm very excited about, but I've never heard of Flyte or Orchest.


If you need to scale out or speedup pandas, there's Modin https://modin.readthedocs.io/en/latest/ (which uses Ray from)


Gunnar Carlsson will be teaching a related tutorial ("Using topological data analysis to understand, build, and improve neural networks") on April 16th in New York City https://conferences.oreilly.com/artificial-intelligence/ai-n...


RISE Lab's Ray platform (now includes RLlib) is another option https://www.oreilly.com/ideas/introducing-rllib-a-composable...


Chainer's Define-by-run apporach is also described here https://www.oreilly.com/learning/complex-neural-networks-mad...


A few months ago Mikio Braun wrote a great post on this topic (the need for data scientists with strong software development skills) https://www.oreilly.com/ideas/what-is-hardcore-data-science-...


O'Reilly Radar


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: