I'll steer clear of the code attribution/copyright questions. That's too complex for my brain. I'll also steer clear of the fact that one should understand the code they submit. That's obvious, surely.

What I'm primarily confounded by is Linus's "AI is a programming tool like a compiler is" analogy. I first heard him make it here (the video starts where his remark does). His recent mailing list email seems to be an extension of that line of thought of his, together with the "editors are programming tools too" remark for effect and bluster. The analogy seems to be, AI is a tool that takes the process

[High level code] -> [Compilers] -> [Machine code],

and augments it to

[Code narration] -> [LLM] -> [High level code] -> [Compilers] -> [Machine code].

I sort of see the analogy, but I am not as smart as Linus so I don't really see it. Unless you're working on the compiler, you're not going to, nor are you expected to, check the low-level code it generates from high-level code for correctness. Isn't it also the logical conclusion of that same analogy that we don't need to check the high-level code an LLM generates, either? This conclusion isn't far fetched. It is what seems to have happened with the race condition issue behind his recent 10 patch series he used an LLM to produce.

I actually do understand Linus's pragmatic approach to LLM usage declarations, but as a leader of an enormously important project, he appears to be setting an awfully bad programming example by pushing and/or pursuing this compiler analogy.

Oh, yes, terrific news. Wouldn't want to curtail the destruction of physical books as they get ripped up, scanned for LLM training, and discarded.

[...] congrats on becoming a comaintainer of NextBSD! Now, can you do me a solid and look into fresh name options for the project that, oh, I don't know, don't seem virtually indistinguishable on print from the name of your cousin project, NetBSD? How about NeXTBSD? Make sure to run everything by your lawyer buddy, Claude, Esq., to make sure we don't find ourselves in any trademark disputes. Thanks!"

I'm half-kidding. I think I'm grumpy because every. single. instance. of "NextBSD" on this story initially registered in my mind as "NetBSD". Maybe my brain has turned lazy with age and opts for shortcuts when reading text.

On a more serious note, cool story, and good luck. Who knows, maybe somebody at Apple has unwittingly shared all the good proprietary stuff with Claude, and now this Claude can magically make NextBSD run well on M* Macs.

I used the Carvana app a couple of years ago to buy a car and trade in my own. I paid for the difference in cash. I was a little uneasy buying a used car without having driven it first (yes, I know there are return policies, but it's a hassle) but the purchase and trade-in process was really very smooth. I drove off with my new (well, used) car after 30' at the Carvana location and seriously wondered why I'd ever go back to a dealer, at least for buying a used car.

Then I was told there was a title transfer problem with the car I'd bought. The registration wasn't coming through because of faulty documentation on Carvana's end. I went through two 30-day temporary renewals, the maximum allowed in my state. Fun fact, the last temporary renewal lapsed, too, while I was out of town for a long work trip. Thankfully the problem got resolved by the time I got back. So, after around 3 months from the purchase date. The most frustrating aspect, for me, was how difficult it was to speak with someone to get updates. In the middle of all this, I also discovered a whole Class Action lawsuit about exactly this. I hadn't done my due diligence well enough, I guess.

Would this have happened if I had gone through a dealer? I think it's less likely.

Will I use Carvana again? Also less likely.

To be fair, the University of Pittsburgh Medical Center you are pointing to is not publicly traded, and cannot be publicly traded (so no IPO). Because it is not-for-profit.

The GP's point is presumably that there is an incongruity in combining "not-for-profit" with "IPO." That's true. Of course, in OpenAI's case, it is technically not the parent non-profit foundation conducting the IPO, but a for-profit entity it spun off and has a controlling stake in. I imagine the GP knows this and was being sarcastic about exactly this contorted structure.

By the way, if you're saying that the label "not-for-profit" has been stretched left and right beyond recognition, then yes I think I agree...

Having an AI avatar take meetings for you sounds like a good way to ensure plausible deniability.

"I do not recall ever personally directing subordinate X to do action Y."

What are the chances this has factored into his thought process?

Good question. My understanding is this: we're being nudged toward an era where there are only a few powerful computers, we don't own them, but we rent computing power from them. To use in our dumb terminals. Not too much unlike an era we had a few decades ago.

100% agreed.

Right. I was thinking maybe with a scannable QR code content identifier, or a verifiable watermark. One that only 'you' can create for the content. And which people can trace back to 'you.' Like private/public key pairs. Except user friendly.

I think we diverge here, but possibly because we're thinking of credentials differently. If credentials allow me to trace back the name/etc. of the entity behind them, then, sure, corporations can have one too. One per entity. I can choose to react differently to corporate credentials than to personal ones.

A big question: who hands out the credentials? Do they do identity confirmation? The picture gets murky for me here.

Last paragraph in the article:

Way to bury the lead. Company with a history of anti-competitive practices sues competitor. News at 11.

PS. For what it's worth, Amazon's point that Perplexity inadvertently messes up Amazon's contracts with advertisers "who pay only for legitimate human impressions" does seem interesting. I guess they're inadvertently over-billing their advertisers. I'll be curious to see how this develops.

Thank you! +1 if I could. I did know of Po-Shen Loh, but only as the US IMO coach. The extra dimensions to him that I could glean from the video are awesome.

[...] without the burden of understanding. Reality will provide feedback shortly." This phrase is on a sticker in my office. (Look it up, there's a cute sloth on the sticker too.)

I'm currently teaching proof-based math to college freshmen. In the honors track. My syllabus does not disallow the use of AI because I haven't been able to wrap my head around how to best do that yet. Students end up with homework scores with averages in the very high 90s. Standard deviations in the low single digits. This is, of course, way more impressive than past semesters.

Then came the first midterm, a month into the term. In class, closed book, closed notes. It covers basic things, simpler than homework, and is preceded by a practice midterm they can do at home with questions of a similar nature to the real thing. The result: an average of 70 and a big standard deviation. It looks like the usual distribution, just way to the left compared to the past.

A few students have come into my office to discuss their performance. I see them see the sticker. I think they then start to feel the burden of understanding. I'll know more after the next midterm.

We're all still learning how to deal with this new reality.

This reminds me of Cluely, whose manifesto concludes with this (emphasis theirs):

I know they're trying hard to be slick and catchy with their wording, but boy does this framing make me uncomfortable.

PS. Here's a wayback link, in case the manifesto goes kaput soon.

The situation you describe here is actually quite common. From the Texas Department of Public Safety:

I've encountered similar wording in every US state I've been in.

But in any case, I think they're precisely trying to maneuver themselves out of this situation altogether. Quoting from the article:

"Providing guidance" ... Slick. Let's see how that goes.

"Is the AI going to go to jail for you?"

Fun fact: we don't even seem to be able to give "the AI" traffic tickets yet in CA.

On a more serious note, the likelier scenario is: an AI company roll out an AI tax prep tool that waives all liability and is, say, 50% cheaper than all the (non-free) alternatives. I can see people going for that. Some will say, well, I'll do it if the expected value of my refund minus my cost and penalty dollar value is > 0. I'm not one of those people. But they're out there.

Aha! That makes a lot of sense. Thank you for the clarification.