you forgot Frank Jr. SAY HIS NAME, DAVE

Depending on what side of LinkedIn you're on, you're going to run across JEV, hyped as the new shit of LLMs. It is not. It's the marketing hype of discriminative models, induced by some folks discovering that frontier LLMs really shouldn't be used for classification tasks (for a bunch of reasons)

for a moment I thought you said the quiet part out loud...

LLMs really really suck at being random. Much like hallucinations, that isn't a bug. It's how autoregressive models work, it's by design. Testing an LLM with a coin flip tells you nothing about its actual "intelligence" or reasoning capability; it just reveals human training data bias. I like 7.