Paul Barba
@happily-human.bsky.social
Machine Learning, AI and Natural Language Processing Expert. Soon to be published author. Curious human.
Random AI / NLP Paper Thoughts Wednesday. They don't let me name things for a reason.
arxiv.org/pdf/2411.14708
For certain cases, DeepMind shows that embedding a list of numbers via LLM and plopping an MLP on top outperforms proper regression with XGBoost working on the, y'know, actual numbers.
arxiv.org/pdf/2411.14708
For certain cases, DeepMind shows that embedding a list of numbers via LLM and plopping an MLP on top outperforms proper regression with XGBoost working on the, y'know, actual numbers.
arxiv.org
December 4, 2024 at 3:37 PM
Random AI / NLP Paper Thoughts Wednesday. They don't let me name things for a reason.
arxiv.org/pdf/2411.14708
For certain cases, DeepMind shows that embedding a list of numbers via LLM and plopping an MLP on top outperforms proper regression with XGBoost working on the, y'know, actual numbers.
arxiv.org/pdf/2411.14708
For certain cases, DeepMind shows that embedding a list of numbers via LLM and plopping an MLP on top outperforms proper regression with XGBoost working on the, y'know, actual numbers.
I have to assume David Hume would be raising a Series A to buy more GPU's to scale LLM's into cognition, and Immanuel Kant would be scoffing at the field and demanding we give proper attention to neurosymbolic architectures, right?
November 29, 2024 at 4:58 PM
I have to assume David Hume would be raising a Series A to buy more GPU's to scale LLM's into cognition, and Immanuel Kant would be scoffing at the field and demanding we give proper attention to neurosymbolic architectures, right?
Just spitballing, but having worked through DevOps, MLOps, now LLMOps, should we just OpsOps and be done with it? Here's a tool to build tooling. Have fun. Done.
November 27, 2024 at 6:12 PM
Just spitballing, but having worked through DevOps, MLOps, now LLMOps, should we just OpsOps and be done with it? Here's a tool to build tooling. Have fun. Done.
I'm unreasonably happy that searching #NLP here gives me natural language processing results and not neurolinguistic programming results. Score 1 for Bluesky.
November 26, 2024 at 3:10 PM
I'm unreasonably happy that searching #NLP here gives me natural language processing results and not neurolinguistic programming results. Score 1 for Bluesky.
What's everyone's thoughts on the value of traditional #NLP (sentiment, NER, supervised learning, etc.) vs. just asking an #LLM for insights? I think the repeatable, structured results have value in tracking insights over time, but might just be holding on to what once was...
November 25, 2024 at 10:31 PM
Hello Bluesky Universe. To kickstart the process of moving to a new social network - if I should know you, reach out! I work in AI/NLP, I write, I am a curious human being. Or, tell me who I should follow. Thanks!
November 25, 2024 at 7:04 PM
Hello Bluesky Universe. To kickstart the process of moving to a new social network - if I should know you, reach out! I work in AI/NLP, I write, I am a curious human being. Or, tell me who I should follow. Thanks!