I mean… they didn’t specify it had to be random (or even uniform)? But yeah, it’s a good showcase of how GPT acquired the same biases as people, from people…

Ever since reading about the challenge of deleting an image from your profile, a GUI for that. It should not only be an API call, not should you have to contact your instance admin to do it. It should be completely self-service from your profile page.

Which is still weird.

Alexander Sawchuk, then an assistant professor of electrical engineering at the University of Southern California … along with a graduate student and the SIPI lab manager, was hurriedly searching the lab for a good image to scan for a colleague’s conference paper. … Just then, somebody happened to walk in with a recent issue of Playboy. The engineers tore away the top third of the centerfold so they could wrap it around the drum of their Muirhead wirephoto scanner…

https://en.wikipedia.org/wiki/Lenna

Everything about the story sounds like it was a rush job, a decision made on a whim, after exhausting their existing catalog of test images. And who bring a Playboy mag to their university’s computer lab, and advertises their possession? They don’t even say who it was, probably to protect them from any embarrassing professional consequences. To me, that’s probably the strongest reason to retire it: it’s unprofessional.

Well said.

I would say that the “positive vibes only” trait is part of it, but the far bigger problem was the character limit. Even when it was double from 140 to 280, that still doesn’t not leave room for nuanced opinions. And then, the least nuanced opinions also become the most easily spreadable. Both traits really reward our worst instincts.

Wouldn’t this enable, for example, Trump claiming he didn’t make the “bloodbath” comment, calling it a deepfake, and telling Youtube to remove all the new coverage of it? I mean, more generally, what stops someone from abusing this system?

The author does have a way with words lol. I love this paragraph in particular, emphasis mine:

As we speak, the battle that platforms are fighting is against generative spam, a cartoonish and obvious threat of outright nonsense, meaningless chum that can and should (and likely will) be stopped. In the process, they’re failing to see that this isn’t a war against spam, but a war against crap, and the overall normalization and intellectual numbing that comes when content is created to please algorithms and provide a minimum viable product for consumers. Google’s “useless” results problem isn’t one borne of content that has no meaning, but of content that only sort of helps, that is the “right” result but doesn’t actually provide any real thought behind it, like the endless “how to fix error code X” results full of well-meaning and plausibly helpful content that doesn’t really help at all.

And he describes exactly what I have to deal with on the regular, “content that only sort of helps” that “steals your attention from the content you actually want.” Even moving from Google to DDG has only mitigated this problem, it hasn’t fully gone away.

But yeah, one of his conclusions seems to be the Death of the Hyperlink? Which, I mean, not even LLM’s can kill that. I doubt <a href is going away any time soon.

I’m kind of impressed by the amount of research they did to figure out why this guy’s bill was so high, then immediately offered a resolution, and then immediately offered another avenue if the resolution wasn’t good enough. Shout out to the customer service rep.

Ich bin ein kbinner?

No

There are far more important facets to truthfulness and semantics than yes/no questions. If this is the only way you evaluate LLM’s, you will quickly fall for confirmation bias.

The car or the account?

Por que no las dos?

Poor Duolingo. Once upon a time I used it to learn Japanese, but by the time I could start reading kanji and noticed that duolingo was still constructing sentences entirely out of hiragana, I knew I had outgrown it and moved on to Anki.

Using AI to learn a new language has to be incredibly frustrating - you can either tell where’s messing up, or you can’t tell at all and then you learn incorrect information…

None? I don’t debate that Blue Sky is corporate-owned while Bitcoin and the Fediverse aren’t. Rather, I’m saying the thing they all have in common is that they like to think of themselves as “decentralized” federations of independent systems and users, but in reality they are all “centralized” systems with shared weaknesses. This is the “ideological contradiction” I thought you were referring to.

Venture Capital

This is the same criticism that was made of cryptocurrency’s claim to fame regarding decentralization, consensus, and resilience to authoritarian takeover.

“If you take all these different parts of your identity, all the games you play, all the things you buy, all the groups you join, and stick them into one system, that’s a central system. It doesn’t matter how many servers that system spans, you’ve pooled all that data in one place.”

And ultimately we can make the same criticism of the Fediverse itself. It’s nice that there are different platforms, different instances, different communities… but it’s still just one entity at the end of the day. This is especially apparent with the spam wave we just saw. Misskey, Mastodon, Lemmy, even kbin was not invulnerable. You don’t need to attack them individually, you can attack them all at once, and then they will naturally spread your attack to other instances for you.

Super bowl ads don’t mean anything. The crypto world also ran ads before/at the super bowl, look how they’re doing now.

Air Canada had to shut down their AI chatbot because it was making up refund policies and costing them money.

New Hampshire is pissed that someone used an AI Biden voice to spread voter disinformation.

If it’s not already there, AI is quickly approaching the FO phase of FAFO.

Almost makes me nostalgic for the way clothing used to work in Cyberpunk 2077.

“We have Fandango at home.”

Fandango At Home at home: