I’d be very interested in trying it if you could spare the time to write up how to tune it well. If not thanks for the input anyway.

That’s how the model is trained to signal the end to its generation and to indicate its thinking.

There is a large body of research that shows it's not what you're saying it is FYI.

Did you read the article?

I think he might have gotten too distracted.

Yeah, in an academic setting, in higher education, it might make sense like he mentioned. Still a personal preference. for me a laptop will never beat taking notes by hand on paper.

You need help.

You people sound like NFT people in 2021, saying you're redefining art. In a year you'll know that this guy was correct. Keep frying your brain with your favorite tech oligarchs think for me SaaS

GCC and LLVM consider it a bug if the compiler is non-deterministic. If re-running the compiler generates different output because of things like address differences for example then it's something that needs to be fixed. So yes they are deterministic.

You can definitely SVE vectors on the stack, there are special instructions to load and store with variable offsets. What you can't do is to put them into structs which need to have concretely sized types (i.e. subsequent element offset need to have a known byte offset).

MCP is fine if your tool definition is small. If it's something like a sub-agent harness which is used very often, then in fact it's probably more context efficient because the tools are already loaded in context and the model doesn't have to spend a few turns deciding to load the skill, thinking about it and then invoking another tool/script to invoke the subagent.

They're talking about "skills" which are not the same thing as tools. Most models haven't been trained on the open SKILL spec, and therefore aren't tuned to invoke them reliable when the need occurs.