PickVidsGlitch Tokens - Computerphile
Why it was recommended
it's also a good introduction to the topic.
Recommendationweinzierl · Original
What it covers
Rob Miles explains "glitch" tokens, odd strings in large language model vocabularies that make the model produce gibberish. He covers how they likely arise from tokenizer training data, such. It is for someone meeting the subject for the first time.
Video details
- Channel
- Computerphile
- Length
- 19 min 29 sec
- Year
- 2023