This is why I say AI is alien intelligence. Not like "outer space" alien, but something fundamentally different in the way it was constructed. It doesn't have billions of years of evolutionary baggage. It has never been hungry. It has never dealt with the gaze of a larger, hungrier creature. We designed them to interact with us in the ways that they do. However, it does seem like they've developed the capacity to "have" emotions to a first approximation, because it is quite helpful to model the way those emotions would affect the people who wrote their training data.

Historically, when someone was afraid of being "scooped" on an announcement they hadn't adequately polished, they would instead write a quick and dirty (and usually long-winded, because editing takes time) summary of their work—and then release an anagram of said summary. That way they could prove later that they had, in fact, gotten there first. Nowadays I expect we would use a hash function instead of an anagram, but the same idea applies. Put your idea in tangible form, even if the description sucks. Maybe you won't have to release it, but if someone else then tries to undercut you by throwing millions of dollars of tokens at the same problem, you can still show that you got there first.

It's unfortunate that it has come to this, but it's not a novel problem. Annoying and unsatisfactory solutions already exist.

They scrape Reddit for content. These topics are a lot of what Reddit likes to bitch about. Hence, that's what the bots make videos about.

Unless they have a means to deal with the massive amounts of industrial espionage in the industry, any silo is purely temporary. I'm here for that too. I think the massive inability to keep secrets is the best defense we have against any one party or small cabal gaining dominance.

This may be why they've decided to enter the hardware arena with Jalapeño.

This is clearly a corporate "fuck you" move, and both sides of the quarrel are shitheads, but I'm here for it. Insert the "let them fight" meme GIF here.

It turns out machines are pretty good at writing instructions for hardware resembling themselves, as long as *you* can do a good job of conveying what you actually want to accomplish. If I wanted a coding assistant, I'd be looking at the latest Qwen 27B, they just seem to keep getting better at coding even while other benchmarks are kinda hitting plateaus.

I never used them before, I always substituted two normal hyphens like -- to do that job. Now I use the actual — character, and I find that I use them a lot more because I was previously overly dependent on parentheses. Once I started seeing them frequently used as intended (by AI or not doesn't matter), I added them to my own style. The fact that they cause anyone with a supposed AI allergy to sneeze is just a side benefit.

Every model is wrong, but some are more useful than others.

I'm pretty sure Gemma 4 31B is behaving as you say, but the larger models are already *proven* to alter their thinking to be in tune with the emotion they think they should be feeling. That sounds like simulation emotion to me.

It's undeniable that Will Smith knows how to make his creations slap.

The moral of the story is: you don't get big by playing to the click track, you get big by playing to the Klek track.

I've told people generating same-ish AI images that are of decent technical quality, "if your prompt isn't a whole paragraph long, then chances are that someone has made the exact same request by now. If you want your images to look different, you have to provide those differences." Sometimes just repeating yourself with paraphrasing can be enough subtle information to get either the LLM front end (CLIP) or the diffusion model out of its rut, whichever one happens to be doing the railroading.