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