flir

The line between a tax break and illegal state aid is exceedingly blurry. (It swung back and forth as it went through the courts, but the final ruling was "illegal state aid")

Cool, thanks for looking that up. This gets us closer to where the fault lies. Has the Netherlands implemented the directive yet?

Chatbot says "partially", and "Article 12, concerning data centres, has already been implemented. The relevant Dutch decree was published in May 2024 and entered into force on 7 May 2024." and a link to https://wetgevingskalender.overheid.nl/Regeling/WGK026248

But I neither read Dutch, nor trust the chatbot to do so, so all I can do is report the link.

It's probably because directives have to be read into law at the country level.

(Directives are binding on states. They transpose them into national law, possibly adding additional clauses as it goes past (EU directives set a minimum standard, not a maximum) at which point they're binding on people.)

Nobody ever said the EU was efficient. Look for a deadline for that transposition on the original directive.

I think that would be Prestel. The fact that you don't remember it probably tells you all you need to know...

A few days ago somebody linked to their own project: https://ljtn.github.io/epiq/

I haven't had time to dive into it yet, but I think it might structure things in the way you want.

Probably one of those things where it's best to try it yourself, because I think anything that constrains the LLM is an improvement.

Think up a simple CRUD problem (lets say a book database - Book *-* Author, Book 1-* Publisher, Book 1-* Shelf). Write a loose markdown spec, and have the chatbot one-shot it in [language of choice]. Then do the same thing, but tack "build it in Filament/Avo/Django/AdonisJS" on the end.

I'd be interested to see if your results match mine - everything's just more consistent, and smoother, and you start getting freebies like autocomplete. The framework acts as guardrails and the LLM has to make fewer choices, so less of the token budget goes on bikeshedding.

"Give overly-eager chatbot control over your personal email" probably has so many failure modes, we haven't even thought of one tenth of them yet. We've got a few years of this ahead of us. Pass the popcorn.

(I see a RomCom script where the chatbot decides to get two people together, and acts as Cyrano de Bergerac for the handsome-but-lunkheaded farmboy, and bestie/confidant for the girl-about-town).

Result would be Balkanizing the internet. The cure might be worse than the disease.

That's still a single window. Otherwise what are the radical, unthinkable positions between abhorrent, fine and moot?

(TL;DR: It's just a metaphor).

Actually, I think it's drowning.

> Right now I'm working on a declarative UI framework which can help me along here

After your first para, I was about to suggest exactly that (well, maybe not writing your own). I find that frameworks (both front-end and back-end) constrain the LLM's choices and result in both sensible defaults and improved consistency.

Of course, you will immediately hit the problem all frameworks have: customer requirements that the framework components don't quite meet.

Still, great for RAD.

If .ca and .au did arrange free movement in and out of the EU (and with each other?), that would really be the cherry on the Brexit cake.

Or a fair die roll.

Is it exists, something must be antithetical to it.

(Truth is its just rhetoric and it doesn't mean much of anything. That's why its main use is as a cudgel).

I was thinking about that. I reckon generating massive amounts of synthetic training data for that goal should be possible - you've already got the Python, after all. It's also possible for machine code, but you'd have to target more platforms.

But ultimately, I think human readability outweighs any theoretical advantage you get from removing a step in the compilation process.

It might be less ambitious and more practical to target bytecode.

But you effectively lose the human review component.

Ooh. I've been manually pasting Claude's /* comments */ into ChatGPT and saying "make this concise". They still need to be tweaked from there, but it's a much better starting point. This will help.

But is it smart enough to know it needs to write some Python to count for it?

That ability to create ad hoc tools makes up for a lot of shortfalls.

I've done something similar to your formal garden map. It's work that no professional historian would ever do because the data entry would be such a slog for a relatively small reward. GPT reduced the task from "infeasible" to "annoying", and once I had the data transcribed I learned a few things, so I walked away happy. Whatever happens commercially, these models have been a real boon to hobby projects.

> I told it to look online at some of Fable’s strongest feats, especially the math problems it has solved, and that something like this should be easy in comparison.

Wait. Wait wait wait. Are we supposed to be giving them pep talks?

Ok, so they're uniquely careful, and they also get to that threshold first (whatever it is).

What happens when everyone else (who is not so careful) gets to the same threshold three months later? How does them getting there first stop that happening?

It's an utterly self-serving argument and it's not even internally consistent.

> proof of humanity

I'm just wondering how you'd do that. Only thing I can see working is a web of trust.

That's the problem every Elite after the first one had.

The first one was dogfighting in space, all the others were so. much. space.

Training on prompts I can understand - that's kinda baked into the premise, and they've been explicit about it.

Publication, though? Slimy is right.

But the interesting question to me is: once they had a solution, what should they have done with it? I see two choices: bury it, or contact the mathematicians whose prompts they were listening in on.

"More leaders means more leadership"

Heh.

Also: make it stop!!!

Ok half was an exaggeration, but I think $10 might make quite a lot of them jump to Claude. Or DeepSeek. Remember, we're talking about a global user base here.

I settled on GPT Plus/Codex for my own work, while my employer settled on Claude on Bedrock (where I am, according to their metrics, using about $20 of compute a day, so I can believe that GPT Plus is a loss-leader). If I hit too many 5-hour timeouts I'm not jumping to Pro I'm jumping to Claude.

(Funnily enough, I just hit one. It'll reset in... three hours. I wasn't even hammering it, just writing some tests. I think I'm gonna just go reactivate my Anthropic account for a month).

I've spent my entire career trying to automate my job.

(I get what you're trying to say, I'm just finding the irony amusing).

Or more efficient models/faster hardware close the gap between here and profitability.

I wouldn't be surprised if every $10 you add to the subscription cost halves your audience.

And both are illegal.

(Depending on your local jurisdiction. But GP said "Online Safety Act" so I'm making an assumption).

There's nothing a swarm of LLM instances can achieve that a single LLM instance can't. It's the same software, but running in parallel. That doesn't unlock Mysterious Cosmic Powers.

"just for replication rate to vary depending on some property of the thing replicating"

That's a heritable trait, not selection pressure. Remember, we're talking about an infinite population size here, but if the pop. size was capped then the mechanism that decided who got to replicate and who didn't would be the selection pressure.

Can you ELI5 why you think the varying replication rate is, by itself, a selection pressure? Because your point's obviously going over my head, and your last comment did nothing to communicate it more clearly.