diegoperini

May I steal this quote as if it's mine? :bow:

I have the opposite experience where Gemini (even the flash models) has the only useful model for my reverse engineering related use case. My hunch is Google utilizes its free access to entire Google search indices to train itself from niche non-English speaking community websites, much frequently and in a "relevant" manner, which in the end gives these models the most up to date info for this particular kind of work. Every other model is just either 10 years outdated with their answers or simply hallucinates like waaaay crazy.

If I am not mistaken, it's even shown in the marketing materials to build suspense.

> Water blobs vs. fireballs. Pretty sure there's a nice videogame idea hiding in there somewhere.

I like the way you think. :)

That analogy may not be suitable for this case because value proposition between the aesthetics vs the function is different for visual art projects compared to software. There is also the maintainability factor where most aged software (especially the closed source ones in private sector) change maintainers every few years. Old maintainers most often lose access to the source code and become unreachable after leaving their job.

"Protection against malicious use" isn't as cool as "model welfare". I'm renaming my authentication function to "examineCrest()".

I recommend replacing "hi chat" in the test spell with "Mortal plane, I greet thee".

> I'm not quite sure why you'd want to run Erlang on it, but the hardware exists.

Erlang is invented before IoT was a thing to facilitate distributed computing for telecommunication in a highly reliable manner. It makes perfect sense to adapt it for driving fleets of cheap IoT devices.

> And here are my clearly unimpressed “friends” >:(

These friends don't get it!

Low resource footprint, written in Go, embed-able in any Go project as a library, compiles to mobile with little to no modification, supports config change without restart, has plugin API.

These were the reasons why we used it in my previous job.

> it has a truckload of data

It would be "artificial" only if LLMs performed badly despite having an equal amount of data containing examples of eastern customs in its training set. Even that's arguable since we don't (didn't) have the benchmarks for this particular case before.

It's no different than GPT answering a prompt with "That's a wonderful idea!", except it's in a different language than English. It's a good thing if LLMs can do this in every language and for any culture with no compromise.

> yet the end result was complete shit

Could you elaborate why? It looks like a useful protocol.

> Linux separates things such that I was looking at C files in drivers/platform/x86 and header files in include/linux/platform_data/x86. And the ACPI code lives other places as well. It’s all very orderly, but at times it felt like navigating a grocery store that arranges products in alphabetical order. Logical, but not exactly cozy.

So beautifully put!

You get to choose your own, more efficient "PI" for your model. Still, it's a valid question.