Probability scores let developers set their own risk thresholds

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 20:02

   So in that example of the personal assistant, the fundamental basic concept is that Jeff is going to take the input, which could be data and or instructions, and will translate this not into text, which is what your language model does, but into a set of probabilities. So they will say, this is urgent, yes or no, escalate, yes or no, and if you escalate, send this to this department or that department, right?

2. 20:37

   Yes, and also, there's like interesting nuance here, where you don't just ask, Jeff, should I escalate? Because that's a very high-level decision. It requires experience and trust with working with someone to know if you should escalate or not. um the idea is that you can keep on decomposing your questions into like the lowest level you we have the ability to break it down like oh i want to filter out all inbound sales from my inbox and i want to maybe filter out like recruits emailing me right um because like we have like a like a specific email list for that like that might make sense but there might be a world like oh i've been i filtered out the wrong email let me add new behavior an example is if someone else is on an email i probably shouldn't be on that email but if they say diogo can you look at this

3. 21:33

   maybe i should all of a sudden see that and this is like nuanced smart behavior that requires like a smart person to think about that behavior so either way a system is getting programmed either by natural language or code but if you do it by code it's cheaper faster and guaranteed to run um and the idea is you can keep on like drilling down into the behavior you want as you make it more and more and more great for your use case rather than with llmc there's only so much control you have and the developer gets a score back for whatever little um sort of you need they're focusing on is that email uh angry or not angry 0.85 and if it's above 0.9 then trigger this action if it's below don't trigger that action yeah yes and also actually your your note on the probabilities is also really really great there's a reason why we do that and we learned

4. 22:32

   this the hard way um like when we like shipped the open ai models they don't have like fine grain controls. And what ends up happening is, I'll use the simplest analogy, which is refusals, right? Like, how does it make sense that you have a response? A refusal is when the model refuses to answer. Yes, I'm sorry, I can't do that. And it's kind of crazy that for some responses, you just refresh it. And sometimes it refuses, sometimes it doesn't. Like, that's insane behavior. And also, if you're building on top of this, that your threshold for refusals might be very different, right? Like, if you are Disney, you probably have an extremely different threshold for refusals than if you are character AI, right?

5. 23:12

   Like character AI likes to do like role play and stuff. Disney is very protective of their characters and IP. It doesn't make sense to just have a discrete decision. The most powerful way to empower developers is to give them back as much information as you can so they can then choose for their application what is the threshold. Like, oh, I accidentally misclassified an email from an extremely important investor.

6. 23:36

   Let me tune that threshold lower because I'm okay with false positives. I really can't stand false and negatives.