I think it's a good idea to experiment with and discover the limitations of small, untuned models before exposing yourself to the modern very powerful ones. It gives you a better sense of their nature as token predictors and not real sentience.

In the same way, seeing an incompetent stage magician fumble before you see a very talented one perform flawlessly will help you understand that it's all sleight of hand. If you jump right to a professional performance, you might think it's real magic.

Agreed, and you could get a completely unrelated LLM to generate a similar apology without any of the real context, it would make up reasoning just as effectively.

Seems valid to me. I won't read articles with model-generated header images, because it's a good indicator the rest of the text will be slop as well.

For a restaurant, a slop logo gives the impression that the owner doesn't care about the details and has no taste.

Beyond that, the use of generative models is a big moral issue for a growing number of people.

Yes, this isn’t a logical error at all. If you don’t have taste in one area — actually, it’s even worse, you’re not even aware of your own lack of taste — why would I trust your taste in another area?

The best Mexican places I've been to in CA had decor reminiscent of a big-city bus station, despite how good the food was.

Why would lack of taste in graphic design be even remotely related to lack of taste in food preparation?

It's like arguing you wouldn't trust a lawyer with a medical negligence case if they can't suture a wound.

Or you wouldn't trust a graphic designer with a restaurant logo if they can't make good scrambled eggs.

This is assuming that the owner is also the chef, and exclusively concerned with cooking. Being a restaurateur is a multi-disciplinary job. The owner's job is literally to have good taste in all areas of the restaurant business: food, interior design, hospitality, branding and marketing, etc. No one is saying that they have to be a graphic designer. The obvious answer here was to have the good sense to hire a local college art student for like $300 to make an endearing and meaningful logo.

Do people even expect anything from “sports” grill?

Probably more likely to run into a Shoggoth than Cthulhu in the Antarctic, or maybe an Elder Thing if any are still around. Rl'yeh is a bit further north.

Maybe I'm missing something but I don't understand where that 3.9 g expected weight comes from.

Sounds like my Dad, who used to have an uncanny ability to get stuck in elevators. Even got stuck in one with his claustrophobia therapist.

Regarding the interactions shown in the screenshots:

LLMs are pattern-matching machines. They keep the pattern going. Once "the agent disobeys the human's instructions" has made its way into the context, that is the pattern that it's going to keep matching. No amount of telling it to stop will make it stop.

The only possible solution is excising it from context and replacing it with examples of it doing the right thing. Given that these models have massive context windows now and much of the output is hidden from the user, that's becoming less viable.

It makes the behavior more obvious from simply looking at the file, for one thing, and it means you can just lump it into your next `git add -A` without needing to handle it specially.

I've never been a fan of dual grid, and personally prefer the rpg maker approach of using 5 sample tiles and then chopping up and recombining them to make the 47 tiles needed for what I believe is called "blob tiling".

Teams messages expire in 30 days at my job, we use email for anything that needs a paper trail

Yup, email is usually the preferred communication tool of record. In a previous job, our messages on Teams were wiped after 8 days so anything that needed to be recorded had to be in an email or some form of document.

And some platforms like Slack, WhatsApp (and previously Skype) make trying to find archived information such a slog as to not be worth it.

I can search email in two seconds.