News for hackers. Stuff that matters.

A comment on GPT-4o's Memory Breakthrough (Needle in a Needlestack).

Re:GPT-4o's Memory Breakthrough (Needle in a Needlestack) by cararemixed on Tuesday May 14, 2024 @05:33PM (#40360403)
What's the chance that these limericks are now in the training set? As others mention, it'd be interesting to come up with a way to synthesize something sufficiently interesting so it always evades training fit.
[ Parent | on HN ]
Re:GPT-4o's Memory Breakthrough (Needle in a Needlestack) by sftombu on Tuesday May 14, 2024 @06:43PM (#40361000)
Previous answer to this question:

https://news.ycombinator.com/item?id=40361419

[ Parent | on HN ]
Re:GPT-4o's Memory Breakthrough (Needle in a Needlestack) by causal on Tuesday May 14, 2024 @11:30PM (#40362795)
Your test is a good one but the point still stands that a novel dataset is the next step to being sure.
[ Parent | on HN ]
Re:GPT-4o's Memory Breakthrough (Needle in a Needlestack) by dontupvoteme on Wednesday May 15, 2024 @05:02AM (#40364644)
One could also programmatically (e.g. with nltk or spacy, replace nouns, named entities, etc) modify the dataset, even up to the point that every test run is unique.

You could also throw in vector similarity if you wanted to keep words as more synonyms or antonyms.

[ Parent | on HN ]