HighGoldstein

> but I also assume that there is very little training data available for Smalltalk.

Is this actually an issue? I was under the impression that the conclusion was that LLMs will invariably be terrible at writing good, working code (since their training sets are 90% garbage produced by humans), hence why agentic development now is heavy on verification tools and other software that can signal to the LLM whether the code works and has any code smells. Also why AI agents nowadays seem to write and execute scripts for a lot of tasks instead of spitting code directly. Hence, the volume of training data shouldn't be as relevant as it used to be.

"I prefer Claude Code, so it must be the best"

Truly a developer mindset.

Software engineers in a good chunk of Europe/US/Canada/Australia make 6 figure salaries, of course they can easily afford an iPad Mini.

Wasn't the whole point of Fable that it's Mythos but with all the cybersecurity stuff lobotomized out of it?

They also took the correct path for hundreds of other problems, where they didn't have a user that already did.

> Our TranslatePod Pros will automatically translate every language in real-time.

Wasn't this the utopian ideal of the Star Trek Universal Translator? Is it good when Gene Roddenbery writes it as utopian sci-fi, but bad when the AI people make it a reality?

> What benefit are AI generated images to an AI?

Apparently quite a bit as long as you're not indiscriminately feeding random slop back in. An AI-generated artifact deemed "high quality" or somehow desirable by humans is a good training input depending on what you're trying to do. Humans don't train exclusively on external inputs either.

This may look intuitively obvious, but the current shortform video situation probably would've looked like dystopian sci-fi to someone in 2010.

"Do people really want to be seen making using Java part of their personality?"

Has it? The i7-5960X was 8C/16T, Ryzen 9 9950X (and variants) are 16C/32T. Even the Threadripper PRO 9000 WX is 96C/192T and supports a staggering 2TB of RAM.

How so? Memory bandwidth and capacity has been steadily increasing. On desktop for example, 10 years ago an i7-5960X was the absolute top of the line and it had a limit of 64GB, with most motherboards in practice being limited to 32-48GB, and this was an unthinkable amount of RAM at the time for desktop. Now the top end desktop chips support 192-256GB, and a year ago (before the RAM price spike) you could realistically buy that much if you wanted for a reasonable price.

Watch as memory manufacturers learn nothing from the car industry, and cry for tariffs when the Chinese inevitably annihilate them.

I think we are still far from nailing down good LLM benchmarks, because the more general-purpose your software the harder the question of what makes it good becomes. Is Python a good programming language? Is Java? Is C? I think it's a similar class of problem. You can benchmark rudimentary things like execution speed similar to how you can benchmark tokens/second, but these metrics don't tell the whole story.

The average person already does this with obvious AI slop.

> In practice, KYC does little to stop any of that

Source? The fact that it's not infallible doesn't mean that it's ineffective.