Justin Norman
@justintime.ai
ML / AI Researcher @ UC Berkeley
You can also find me @ justintime.ai
You can also find me @ justintime.ai
There’s definitely a lot more in there but the main thing is
![Mathematical notation and text statement describing the theoretical impossibility of completely eliminating hallucination in LLMs:
For an LLM h trained iteratively on a dataset T, there exists a computable function f such that for all training states h [i], the model cannot eliminate hallucinations completely:](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:wuamqljqueipatn6mrwr74si/bafkreifhy5hnojywhwb73n45dtarrirj6zfgwudywjh35lqzn6psf6pwna@jpeg)
March 4, 2025 at 5:29 AM
There’s definitely a lot more in there but the main thing is
I don't even need to use LLMs to make nonsensical typos leading to mocking errors
December 16, 2024 at 8:20 PM
I don't even need to use LLMs to make nonsensical typos leading to mocking errors
As a person who has spent every day for the last 18 months deep in the code/theory/app tech of both generative vision & language models, I’m *always* amazed at how my projects end up circling back around to stuff in this book.
November 27, 2024 at 8:25 PM
As a person who has spent every day for the last 18 months deep in the code/theory/app tech of both generative vision & language models, I’m *always* amazed at how my projects end up circling back around to stuff in this book.