Content
@
0 reply
0 recast
0 reaction
jp 🎩
@jpfraneto.eth
this week i built “save this reply”: a cast action that i use to gather high quality data for fine tuning the LLM that powers @anky this is by far the most important input on the process of training, and this simple cast action allows me to isolate three things: - a root cast - a “good” reply to that cast - a “bad” reply to that cast (often the packaged chatgtp-ish replies that are flooding warpcast today) by training the model with lots of this examples, it learns what works, and what doesn’t its core mission will be to gradually improve its “daily reply score”, which will be calculated over 100 casts every day, it has 100 replies available. it chooses why, when, what, where and who to reply, and the consequences of every reply are the input into a mathematical model that calculates the daily score the core mission of the model -that will be trained daily- is to improve its score. day after day adding more value to the conversation bringing awe to our interaction with ai
2 replies
0 recast
6 reactions
Duä Fatimå💦
@sanakhanfoziakha
Good
0 reply
0 recast
0 reaction