equestria

Of course, but I'd rather not share it for obvious reasons. It is a nonsensical biography of a non-existing person.

Google does the "ransacking" for them - ContentID.

I exclude all the published LLM User-Agents and have a content honeypot on my website. Google obeys, but ChatGPT and Bing still clearly know the content of the honeypot.

I don't know. The reality is just that most business ideas don't pan out. There were hundreds of well-funded efforts to build online retailers before Amazon happened. If you're well-off, it's obviously easier to try and try again, but you're still likely to fail.

I think the article is right in that if you don't have a clear business idea ("we're building a platform"), the odds are even worse. Except when they aren't, because in some niches, you actually have customers who want a platform. Cloud computing is an obvious example. It's just not the general case for consumer stuff.

That last part makes me a bit nervous. It's dangerously close to the EA belief that it's actually OK to be a ruthless exec for a tobacco company, because you can do good things with your money that you wouldn't be able to do if you quit the job.

I don't think that's the point you're making, but it's good to be careful with that. You can do good after hours, but it doesn't absolve you from what you're doing 9-to-5.

As to your first point: yes, but it's all relative. Most tech workers are "trying to get by" in their minds. Just look at the SFBA rents and the PG&E bills! And wait until you hear about their college loans... most people in the top 1% don't think about themselves as the top 1%.

In the end, making good decisions often requires sacrifice, pretty much no matter how much you make. And we often find ways to rationalize why it's not the right time for that.

The wifi protocol is computationally intensive. The wifi module is effectively a fast 32-bit computer with fairly complex firmware. And then, there's all the RF engineering that needs to happen to make it work.

So, the original thinking was "if you need wifi, we can't price a standalone chip competitively, just buy a SoC". But the genius of ESP32 was that they approached it the other way round: they built a wifi chip, and then figured they can carve out some room for user code. No need to pay for a separate MCU. This worked for a lot of customers, and the economies of scale took care of the rest.

I think it's one of these things that people like to talk about in the abstract, but how many people really want a big CRT taking up space in their home?

Modern OLED displays are superior in every way and CRT aesthetics can be replicated in software, so a more practical route would be probably to build some "pass-through" device that adds shadow mask, color bleed, and what-have-you. A lot cheaper than restarting the production of cathode-ray tubes.

I'm not sure what you're getting at. The point of these (ill-defined) alignment exercises is not to achieve parity with humans, but to constrain AI systems so that they behave in our best interest. Or, more prosaically, that they don't say or do things that are a brand safety or legal risk for their operator.

Still, I think that the original paper and this take on it are just exercises in excessive anthropomorphizing. There's no special reason to believe that the processes within an LLM are analogous to human thought. This is not a "stochastic parrot" argument. I think LLMs can be intelligent without being like us. It's just that we're jumping the gun in assuming that LLMs have a single, coherent set of values, or that they "knowingly" employ deception, when the only thing we reward them for is completing text in a way that pleases the judges.

I doubt it works this way. People at Google use search too, and they don't like what they see.

Part of the problem is that they're fighting against financial incentives that they themselves created. There's plenty of upside and little downside to abusing it, so it's just endless whack-a-mole.

Another issue is just how bureaucratic the process has become. They want it to look good to the regulators and the courts, so they put up with a pattern of abuse for five years, then announce some well-reasoned but narrow policy change (e.g. "product reviews now need to be actual hands-on reviews"), and... a month later, spammers are just adding an extra lie on all the fake review websites.

Oh, come on. Not everything is addiction. I can accept that algorithmic doom-scrolling is somewhat habit-forming, but even there, we have agency. But background music? Yeah, I like it, but I don't get restless or frustrated when it's not playing.

Well, then again: maybe Spotify was hoping you wouldn't notice, but by now, the problem has been exposed publicly a number of times. This article is one of many.

How many of us are canceling their Spotify subscriptions over this? It wouldn't be some huge sacrifice, it's about the least we could do. Most of us won't. The "caring" is just lip service.

If a customer wants endless elevator music, then I don't think that Spotify is wrong to generate endless elevator music for them. The problem is deception. If you want to listen to human performances, then Spotify should give you choice instead of hoping you don't notice.

Free market means you can vote with your wallet. If you don't, then it says less about markets and more about our stated vs revealed preferences. Maybe we just don't care if real artists go away.

Most homes of people who are reasonably well-off - including most people on HN - go through three distinct stages.

The first stage is right after you spend an insane amount of money to buy the property. At that point, you want to validate the investment, so you spend more on making it look nice, but you're probably oblivious to many realities of homeownership. So yeah, this is when people splurge on matched kitchenware, beautiful cutting boards, "smart" kitchen appliances, sleek-looking but impractical veneer / plastic / glass furniture, etc.

The second stage is when you get kids or pets, and you start losing the battle. You eventually throw in the towel, accepting that there are going to be dings on the walls and on stainless steel appliances, holes in window screens, and veneer peeling off after the fifth juice spill accident.

The third stage is when the kids move out and you can actually make the space look nice. Except now, you know that there's some wisdom to old-fashioned solid wood furniture, that cutting boards are for cutting, and so on. So your home acquires more of an "old people" vibe.

I don't think it's a useful way to frame it. Model training is very compute-intensive. Generation isn't. You can run it on locally on consumer hardware. It's just not very monetizable, so we're converging on a black-box "in the cloud" approach.

If we're focusing on energy consumption, using an LLM to generate a newspaper article probably uses less energy than a living journalist would use. The morning commute alone is probably "worth" a dozen articles.

The problem is different. First, that journalist is still alive and consuming resources, just out of a job. Second, because you now have a very cheap way to generate an infinite number of articles, and there are commercial incentives to do so, the "dead internet theory" has a good chance of coming true.