They also have an increasingly disturbing tendency to end a response with a question. Seems like an over engineered reward in RL to keep the conversation going.
Anthropic publishes system prompts and at least for the case of Claude 3.5 Sonnet 2024-11-22 asking questions is explicit.
"Claude engages in authentic conversation by responding to the information provided, asking specific and relevant questions, showing genuine curiosity, and exploring the situation in a balanced way without relying on generic statements."
This book actually makes some sense to me as a software engineer. We’ve all learned this at school, but they were just scattered pieces of knowledge. This book actually offers a way of systematic organization of useful knowledge in production. Content is actually not for learning, but for quick check and review. The organization might not be perfect, but really is a way of reflecting on our understanding in this field.
Everything apart from design or software architecture as such. (Unless you include that one weird class on UML and Sybase) Math, electronics, algorithms, encryption, physics, legal side of things like gdpr, networking, compilers, UX, embedded systems, chip design, different CPU architectures, reading research papers, databases (about acid, not specific implementations), ...
But the part of "how to manage the process of creating this" and related ideas you were basically supposed to figure out on your own. Or actively ask the teachers. It feels very sink-or-swim now in retrospect. I've never been explicitly taught about testing for example, but if you didn't write them for your projects, I don't think you'd complete them.
A really interesting review. The idea of relative error makes sense in most cases, but when we need to do subtraction and difference matters, maybe absolute error is actually better.
If you are doing only additive operations, then, yes, absolute error might actually be the best choice. But as soon as multiplications start to show up, they are enough trouble that they tend to dominate the whole error propagation show. Since many real calculations have multiplication in them, you end up having to optimize the whole thing for multiplicative operations, and so we end up just using relative errors everywhere.
You can, of course, do a very specialized optimization for one particular algorithm, but that tends to not be a very good use of time. Usually. (Counterexample: Kahan summation!)
Subtraction is the very case where relative error might matter most, and error relative to magnitude of the original unsubtracted numbers causes the most surprises.
Absolute error has useful applications, without any doubt, and regardless of arithmetic operation, but it probably doesn’t make sense to say it’s “better” without a specific problem in front of us, and without specific goals and priorities. Error tolerance is always up to the user.
