Why Perfect Prompts Aren't Everything
Nothing is loaded from YouTube until you press play, and then without tracking cookies.
About this video
Once you have a good starting prompt, the model fine-tunes itself to fit the problem. At every step, it adapts to produce the most useful output. The key is how you handle gradients and prefixes, which can be just as crucial as the initial prompt. Explore why prompt quality alone isn't the end game! #AI #MachineLearning #PromptEngineering #Tech #DataScience #NEARai #NEARprotocol #NEARuserownedAI
Transcript
167 wordsCaptions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.
- 0:00
oh I mean if if you wanted to find I guess yeah to increase the chance of Ever Getting to a successful trajectory right yeah but then you can fune the things themselves so at that point assuming your promp was at least somewhat good at the beginning to get to any success the goodness of your prom doesn't matter anymore because the prom just becomes the perfect prompt for this particular problem right yeah uh and not just the at the input level right at
- 0:25
every yeah like level what whatever the key value is that whenever you need to Output the next token would be the most useful yeah which is you know contains instructions and whatever else it can yeah and if the model if you don't actually send gradients right then you just like drop one prefix at another I guess probably as expensive as plugging in low on top of it but yeah so that's that's something interesting relatively easy to try