Jeez, I got some autocorrect failures in there, sorry.

See, in my experience those kinds of numbers are not realistic even on the face of it. I posit that it's not humanly, cognitively possible to read, properly think about, review, and really deeply understand that amount of code.

When I wrote code manually, I'm not just slower because I type slower than an LLM (barely -- I go fast when I know what to type) or I need to look stuff up. During that time, my brain is also super busy looking for and analysing edge cases, data flow, performance, potential security issues, etc.

If I go too fast, those are the kind of issues I'll find with that code days, months, or years later. If I do, the stuff lasts and saves so much time down the line. Even the best, latest, biggest models cannot write code with that amount of oversight by someone who knows the requirements, context, environments, and codebase. Even the LLM coffee I feel I reviewed carefully will have those kinds of issues, because my mindset when "orchestrating" and reviewing agent cover is fundamentally different than when I'm in the designer, architect, and builder mindset. This is also a direct consequence of the amount of code that is passing over my desk.

I'd say that for any productivity gain that exceeds 30%, maybe 50%, on production code, you're inherently going to have that kind of quality loss. It just doesn't scale that way.

My completely unfounded theory is that the Security Response Center is now fully "agentic" and no human ever even looked at their reports, or knew they were banned.

I tried posting a technical question about a Graph API feature to Microsoft's user forums one day. After about twenty times of my post being automatically deleted for violating the community guidelines, I gave up trying to figure out which part of my question tripped up the absolutely inane, useless Microslop moderation bot.

"DDG is worse than Google" used to be my opinion forever. But this has been completely flipped over the past 12-18 months. And I don't think it has anything to do with Bing having gotten better, just with how absolutely abysmal Google results are now.

There was one key moment where an "authoritative" source on some topic was no longer in my Google results. I added the exact name of the website, and it ended up being somewhere on page 3. WTF?

I then did one of my occasinal quick comparisons with DDG/Bing. First result. I've switched to DDG/Bing that day and stayed.

Now, it's exactly the other way around. After some DDG search, I will sometimes curiously do the same search on Google and confirm that the first useful result tends to be way beyond the first page, if found at all.

Google Search is dead.

It wasn't an Internet bubble, it was a "dotcom" bubble. People didn't stop using the Internet because it's very useful technology. The various dotcoms imploded because they were overvalued garbage with no plan on how to reach profitability.

LLMs have their uses, which will survive, but these billions are largely about overvalued garbage with no possible way to profitability.

Read Ed Zitron. He has the numbers. This crash will be absolutely devastating.

The PR guy was already replaced, it seems. That statement is so offensively LLM-generated, it might as well have been satire.

I'm sure their revolutionary new operating model will allow them to not just disrupt new markets, but delve into a symphony of synergies.

The API exists. I'm not sure whether it was unavailable for a while, but it certainly looks like it's available right now:

https://www.last.fm/api

I still use the service. I switch a lot between what I listen to music with (Qobuz, foobar2000, YouTube, Roon...) and they all support it. It's great to have a cross-service listening history, statistics, and favourites.

I think the use case is still cool and valid, although I wonder how many non-veterans know about it. And I visit the site itself very rarely these days, because the social activity died off along with that on pretty much any other non-social-media website 15 years ago, so I'm worried about their revenue model. I would be willing to pay for the service, but also assume I'm in the minority.

It was mostly the other way around for me for the past 15, maybe 20 years. I would sometimes go for an Nvidia card when ATI did something to upset me, or I felt like I wanted a cutting-edge card again for once for some new game -- as you say, they're usually ahead of ATI/AMD technology-wise, although most of the time I ended up feeling that I fell for their marketing, that the new features are an overhyped dud and ATI was lagging behind more in marketing than technology.

But Nvidia drivers, my god. It usually didn't take me long to wish back the stability and out-of-your-way-ness of ATI/AMD drivers. And don't even get me started on trying to use them on Linux.

It's like that to this day for me. My desktop has an (older) AMD card that just works. I never have to bother with it or fiddle around. Meanwhile, my laptop with a 3080 gets weekly driver updates that seem to only make things worse. It still can't handle waking up from hibernate. It still gets confused about HDMI screens getting attached or disconnected and needs several retries. It still causes a bluescreen or a multi-minute system freeze almost every time I use that laptop.

Who would even want to steal such utter crap, though? To what end?

It may have started as a technical limitation, but turned out to have other, nice effects down the line.

Luckily, I'm both a retro gamer and a motorcycle rider :)

Scanlines and other CRT effects were technical artifacts, but artists adapted to them. As a consequence, games from that era were meant to be played on those kinds of screens, with those effects. They look better, sometimes mindbogglingly so, than on a pure, sharp, 1:1 pixel translation to a modern panel. There are some good YouTube videos and blog articles demonstrating this. These fake effects often restore details and effects that otherwise would simply be missing.

Maybe similarly, the whole clutch and gear shebang may have started as just a necessity with ICEs, but it turns out that controlling a vehicle like that is just massively, amazingly fun. As I suspect is the case for most riders in these parts, I don't ride motorcycles as a pure method of transportation. I do it for fun. And riding an electric motorcycle, without clutching and shifting, about 90% of what's so fun and exciting about riding is just missing for me. I don't think anyone without experience riding both could understand. It's not about nostalgia or habit, it's about a physical activity we love doing, and an electric drive removes most of the elements that make the activity fun.

Maybe compare it to drawing vs. generating a drawing with Nano Banana. You can get a nice picture either way, and the GenAI is definitely the more modern, faster way. Depending on your drawing skills, you may even get an objectively much better result. But you'll never have the same fun and satisfaction as when you draw it yourself.

I can't believe anyone still uses the "will fall behind" meme in earnest. At this point it's only really used to mock AI bros. But, apparently, not everybody got the memo. You could say that they fell behind. Badum tss.

Honestly, those idiots can go ahead and leave me behind with all the other luddites who would rather read a news story that's true because somebody researched it, than made-up LLM slop. It will be funny to see the reaction if/when the AI bros realise that what they "left behind" is all that is real and interesting about life.

Start making them again in the first place. I checked my dad's PC a week ago, when he thought his Blu-ray burner was broken (luckily was just a loose power connector), I had already looked for a replacement just in case and realised that (a) LG was the last remaining manufacturer, with all other brands being rebadged LG drives, and (b) LG has stopped production mid-2025. There's only leftover stock or used drives, both at inflated prices (the cheapest I found was $260, compared to the original store price of $80).

I know a lot of archivists use M-Disc Blu-rays as a non-magnetic additional backup media, as do I, so I'm worried about the future of that.

The Chinese Room thought experiment isn't about consciousness, though. It's about understanding Chinese.

It's a poor analogy, but to get it closer to the questions about LLMs: ask the guy in the room, in his own language, what he just said in Chinese. Tell him that the answer he gave was about how old he was, and ask him whether the answer he gave was correct.

That's the thing Dawkins doesn't understand. The point (or one major point) of consciousness is a working model and understanding of the world based on experiences. A token generator can never do more than fake that. Sure, if you ask the LLM about its understanding of the world, it will give you answers that read like it has one. But that's what it's designed to do, and doesn't change anything about the fact that there is no understanding behind the words. You can dig deeper and deeper, and the model will never be able to come up with anything else but a blend of what others have previously said about in a context where similar words have been said.

I think the discussion about determinism is a red herring, and for sure we don't know a lot about how consciousness emerges, but what we know for sure is that LLMs lack pretty much every precondition for even making it possible.

I think it needs to be somewhat targetted so that oxygen is actually pushed away from the fire, instead of more oxygen being pushed toward it? But yes, everything with an "if" or "switch" statement in its code is "AI-powered" these days. It even goes for the double-whammy with its "NASA technology", just like all those plug-in heaters, ACs and phone signal boosters from scam ads.

Yeah, I don't think people would lose a lot of sleep over billionaires becoming mere millionaires.

I have no clue about modern financial markets, but from everything I'm reading, some economies are almost entirely propped-up by these imaginary numbers, and people's retirements and insurances are tied to those economies. I'm hoping for the "AI" crash as much as the next guy, but also fear that what comes after will be a far cry from business as pre-2022-usual.