nhlism.bsky.social - Profile | ThreadSky | a Reddit-style client for Bluesky

nhlism.bsky.social

Senior ML at 🍎 Had fun at Alexa Prize 2017. Interested in NLP and bounding boxes around things. Love less attention in text and more attention in vision. OMSCS 2025(?) Recurve Archery and weird projects 🇨🇿

24 posts 75 followers 379 following

Posts 12 Comments 12

In a world of more and more LLM generated data, I turned to the only thing I know for sure is human generated -> my own journal. I'm doing data analysis on 5 years of my own journaling data.

submitted 54 days ago • 4 comments

after eating west coast oysters for 5 days straight the last restaurant tells me they got recalled

submitted 108 days ago • 0 comments

#neurips2024 logistics

submitted 112 days ago • 0 comments

#neurips2024 food

submitted 112 days ago • 0 comments

no no no no no

submitted 113 days ago • 0 comments

at Symbolic AI and Foundational models panel because I used RDF in 2017 and desperately want it to be relevant again so I can be cool

submitted 113 days ago • 2 comments

Diffusion(?) based generation of a samoyed by #meta

submitted 113 days ago • 0 comments

Salesforce has agents doing tasks on slack which is both interesting tool-using-LLM but also has just a tiny bit dystopian-dead-internet vibes

submitted 113 days ago • 0 comments

Got baited by Anime posters at #neurips

submitted 113 days ago • 0 comments

To try: have a autoregressive model predict number of tokens needed before generating. One problem is once you commit to generating N tokens you don’t know the gradients for 1..{N}…Max_N, which might be better choices. We could run all Max_N lengths and take min gradient which seems expensive

In a world of more and more LLM generated data, I turned to the only thing I know for sure is human generated -> my own journal. I'm doing data analysis on 5 years of my own journaling data.

after eating west coast oysters for 5 days straight the last restaurant tells me they got recalled

#neurips2024 logistics

#neurips2024 food

no no no no no

at Symbolic AI and Foundational models panel because I used RDF in 2017 and desperately want it to be relevant again so I can be cool

Diffusion(?) based generation of a samoyed by #meta

Salesforce has agents doing tasks on slack which is both interesting tool-using-LLM but also has just a tiny bit dystopian-dead-internet vibes

Got baited by Anime posters at #neurips

To try: have a autoregressive model predict number of tokens needed before generating. One problem is once you commit to generating N tokens you don’t know the gradients for 1..{N}…Max_N, which might be better choices. We could run all Max_N lengths and take min gradient which seems expensive

Biggest source of anxiety at #neurips so far

Black Friday is the best time to purge all subscribed mailing lists