While you’re here, poke around this blog and site, and you’ll see why I talk about these things, what I think, what I do to help. You might also be curious about LawQi.
But enough about me. So let’s get into it.
Wait! Now you’re telling me I have to think about model architecture?
Sort of, yeah, but not always. Today’s lesson is a head’s up that the features and limitations of a given model will vary, and that we can’t just say assume that newer ones behave like older ones, but with more bells and whistles, fewer errors and lower cost.
We’ll save the discussion of the specific differences between GPT-4o, Gemini 2.0, Claude 3.5 Sonnet, DeepSeek R1 and all their brethren and cousins for another day. For today, just an overview that available models broadly fall into one of 3 major categories and that it’s worth thinking about which sort you are using and its suitability to the purpose of your session.
Click a type below to see the details.