What's the trick to training on this dataset? Lots of folks having trouble!
@jeremyphoward
-
Pioneer Recognition Gap in Tech Innovation History
By
–
I don't mean to imply it's just a wiki thing. It does seem to have been an issue throughout history — in reading biographies and histories of tech innovation, it does seem often the folks that first build things can fade into the background, vs the folks that come later.
-
Kaggle Acquisition: Exec PR Over Historical Accuracy
By
–
In this case, the underlying issue appears to be that Kaggle got lots of attention when it was acquired, and the execs at that time did PR stuff. So writing about the company after the acquisition is focused on the POV of those execs, rather than historical accuracy.
-
Wikipedia Revisionism Risk Through Sequential Editorial Changes
By
–
In thinking about it more, it seems like a systematic issue for wikipedia. If later editors change articles based on new references, replacing articles based on contemporaneous references, it opens the door for just this kind of revisionism.
-
Security vulnerability discovered affecting large language models
By
–
Ouch it infects LLMs too https://
x.com/linylinx/statu
s/1708154091782676619?s=46
… -
Logprobs utility in language model applications
By
–
It's super helpful when used with logprobs though!
-
What emerging AI technology hasn’t peaked yet?
By
–
OK so what's the new thing that hasn't peaked yet?
-
Banning Educational Content: Impact on Learning and Knowledge Access
By
–
But if we ban the right books and web pages how will they learn how to do it?
-
Disabling Hugging Face Logging Warnings in Libraries
By
–
Yeah unfortunately Huggingface are extremely verbose with dumping out loads of warnings across their projects. First thing I do when I use an HF lib is to disable their logging.