AI Insiders - Prompt Hacks
Added 2023-12-05 16:54:43 +0000 UTCWhether you're new to LLMs or consider yourself a Prompt Engineer, here is a 30 min tour of my most-used Prompt Hacks, plus expert guest insight from Google DeepMind's Tim Rocktäschel. Covering everything from prompt chaining to self consistency, rephrase and respond to EmotionPrompt.
Slides: https://docs.google.com/presentation/d/1IlOXhv4JUX7SOmPaMKQGe5yagNw3Zr0w83voYMImVv8/edit?usp=sharing
Comments
Can't wait for a finetuned version of GPT that automatically identifies the most suitable prompting strategies for a given prompt and uses these to prompt itself instead of the original prompt. At least this would save quite a lot of time.
Phillip Lakaschus
2024-01-04 08:49:02 +0000 UTCAmazing stuff! Keen to try all these out!
Graham Herdman
2023-12-14 04:22:58 +0000 UTCThanks so much Bojan, more to come!
Philip
2023-12-09 19:54:23 +0000 UTCLove this practical content, huge value for the patreon
Bojan Savic
2023-12-09 19:28:03 +0000 UTCGreat theory, could definitely be part of the explanation, along with the recent fad of 'tipping' a model
Philip
2023-12-09 18:26:37 +0000 UTCAs for the Emotional prompting , I have a theory. I think , as You said , those emotions were in training data . But I think , that its not about models Attention , but more about how the text followed after those Emotions used in text. Usually , if sentence " This is important for my career" is written somewhere , its followed by very good performance of person ( e.g. student , employee) as a continuation of that text - maybe some book , some TV show script , ... . So its just higher quality text for training , because hard Emotion , triggered better performance of the human , all recorded in the text . So the explanation should be in Pattern that it follows , not the attention that model pays to emotions.
Rastislav Dujava
2023-12-07 20:24:38 +0000 UTCThanks so much Martin!
Philip
2023-12-07 11:52:11 +0000 UTCall this is amazing thanks Philip! : )
Martin Percy
2023-12-07 09:35:15 +0000 UTC
