Skylar Payne
@skylar_b_payne
Empowering early-stage engineering teams to confidently launch AI users love, permanently ditching those 3 AM production fire alarms.
How do you identify a good use case for AI? This is a common question I get from prospects. Let me share how I think about it 🧵
what did you get done today? what worked? what didn't? what are you celebrating?
WOW. 40 days of Effective AI Engineering and y'all have stuck with me. I'm not done yet. But I would love to pause to ask: what else would you like to learn about? What problems are you facing in building AI? Please let me know in the comments!
3 observations about ColBERT / late interaction (use in RAG, etc) 1. Not commutative: sim(a,b) and sim(b,a) are not the same thing 2. Vector order does not matter: MaxSim does not consider order 3. Vectors can be at any level of granularity: tokens, chunks, etc
feeling a bit inspired after chatting with @ChrisLovejoy_ ... returning to my behavioral healthcare roots will be using @GoogleDeepMind Gemma models to run on device (h/t @giyu_codes ) more to come...
i really want to start referring to my audience as "plastics" as an homage to Mean Girls, but idk if that would go over well
As a former LinkedIn employee, I 100% endorse this. Leverage GDPR/CCPA compliance to your advantage. I didn't put a shitton of work into GDPR compliance at LinkedIn for nothing. Make my work matter. Also follow @_colemurray he's the most underrated person around
Few people know this, but you can just export your LinkedIn network and run algos on top of it