- Stream Notes
- Doing
- Continuing building GPT Tokenizer
- From Chat / Derail
- #physbuzz
- Cool state parks in California
- Anza-Borrego
- Mount San Jacinto
- Mono Lake
- Cool state parks in California
- #hittaus
- Suggestion: devise a simple evaluation (pure code or promptfoo for example) and try every single prompt tuning from surveys and write a blog post of your results.
- A Systematic Survey of Prompt Engineering in LLMs: Techniques and Applications
- And remove duplicates and take these too A Survey of Prompt Engineering Methods in Large Language Models for Different NLP Tasks
- Then, mix it up by generating some eval cases by mixing with 1 billion personas (have the tab open/have notes)
- And make it a challenge: complete in 1 hour (stream)
- first eval cases running in 10 mins,
- 3 prompt alternatives running in 20 mins,
- persona generator in 10 mins,
- and while you leave it running in Together for 20 mins you write the blog post, and when 1h is full you drop in results into tables of your blog post
- Cut all corners, move fast, break things
- Then next stream you redo it all but in 4h, see what’s different
- Have you played Redactle, it builds intuition for RAGs
- Pick, for example, an evaluation from Mealeon. Like “does Alla Forma and Shrimps go well together”
- Suggestion: devise a simple evaluation (pure code or promptfoo for example) and try every single prompt tuning from surveys and write a blog post of your results.
- #physbuzz
- We raided GrumbleofPugz
- Doing