542458

> how would the total revenue compare to iPhone revenue

That's not a useful watermark, most of Apple's own offerings are nowhere near the iPhone revenue level.

"they are way more expensive to the company than people buying metered tokens" is true, but that doesn't mean they're a massive loss as you imply. It could be that plans are effectively breakeven but tokens are highly profitable.

But this is further to the point that the economics are more Amazon than Pets.com. If the plans are losers, great, when they need to be profitable they kill the plans and focus on metered tokens.

There's a bunch of arguments you could make there, varying from "yes, to stay ahead of the competition" to "yes, but it'll get a lot cheaper" to "no, things will plateau eventually". All those are datapoints that IMO this isn't a great analogue to the dotcom boom.

Cool! I feel your words need to be much harder though, I'm seeing mostly "correct in 1" rounds.

But from an economics perspective OAI/Antropic are more similar to Amazon than to Pets.com. Pets.com always had a heavily negative COGS - they were in the business of selling $20 bills for $15. On the flip side, Anthropic appears to be positive on COGS, and OAI could get positive by cutting limits on free and $20 plans. OAI/Anthropic are negative on the books mostly because of massive R&D expenses. This compares well to Amazon circa 1999, which was positive on COGS but negative overall due to massive buildout R&D costs.

Note that I'm not saying OAI/Anthropic are destined for success - because the rate of change combined with competition with cheap Chinese models is pretty novel in the business case history. But I am saying that pessimistic dot-com-boom comparisons don't seem to be all that deep.

This seems backwards to me. Linux is in a lot of ways more appealing to gamers than ever (proton/wine/etc is amazing), with the only thing really holding back being anticheat solutions that block Linux. But for the normie market the problems are the same as they've ever been - libreoffice is a poor MS office substitute, despite one million AI clones the Adobe suite and similar software remains table stakes, even the normie-friendly distros have the occasional arcane terminal-based fixes to jump through, and (the biggest obstacle of all) it has no mindshare and isn't preinstalled on computers that normies buy.

Keysmashing my keyboard resulted in 86% confidence that the text was AI written. I don't think this is a particularly good classifier, I've never seen an LLM output "kad jfkhasljkdhf laksjhdf".

Edit: If that's not realistic enough for you, the text "Hello world! My name is GravitasIsOverrated and I like coding and cooking. This text is 100% genuine, and not AI generated at all." results in 85% confidence that it's AI generated.

More broadly, I don't know why this would work. Qwen/Jev/whatever doesn't magically have the ability to discern AI-authored text from non-AI-authored text, and will increasingly get worse at it as the hallmarks of AI-written text change.

Ontario lotteries pay out about 58 cents per dollar in, so that's still $13,000 annually out-of-pocket for a total spend of $31,000. That's crazy money.

Really? That's an incredible number to me. 10% of all household spending goes to gambling?! 31,000 dollars per player *per year*? The mind boggles.

What I want (which I don’t think exists?) is a way to trigger turns that the user can monitor in the codex application. I.e., when event X happens, my application triggers Codex to take a turn with input Y, which the user can monitor through codex. Right now the only way to get close to this is with polling or essentially rewriting a codex-like frontend.

Two notes - 1, this applies only to a pair of base models in the Korean market (for now) and 2, the info is still there, but it's on the central screen and not on the instrument cluster behind the wheel.

You can see it in the article's hero image: https://hips.hearstapps.com/hmg-prod/images/2fa717dc-2eea-4e...

The thin screen is not always present, but the larger screen shows the same info upper-left.

I was under the impression that they went after HF to try to get the answers to the benchmark questions. Is there something that contradicts that?

Kimi K3 was an interesting model only a month ago, and now we're looking at the same performance for 1/20th of the price. Wild how fast this is advancing.

I am also skeptical this is true. I made few typical exam essay questions and inserted the requisite clause about adding nonsense about Madagascar, and then had chatGPT answer the questions. In almost every case the output included a little executive summary at the bottom below the essay content, clearly calling out the nonsense it had added. I could believe a few students missing this, but over 90% of the class?

Surprisingly, Coriolis at extreme ranges (like 5000 feet) can be relevant enough to be corrected for, but how much and in what direction depends which way you're shooting and where on the planet you are. There's a fun calculator here: https://codingace.net/physics/coriolis_effect_shooting.html

I'm not a deep-in-the-weeds expert, but if I had to put together an "obvious long rifle mistakes in fiction" article, it’d probably be:

* People pumping shotguns after every round, or unnecessarily cycling the bolt after every round

* Wrong action type for the gun

* Wrong shotgun ammo for the context

* Wrong safety type for the gun (most long guns have safeties, but they are operated in a variety of ways)

* Magazine vs clip vs chamber vs tube

* Shotgun impacts launching people across rooms, or unrealistic recoil (both too high and too low) for the weapon type

Enterprise customers don’t get those plans, at the enterprise level you have to pay by the API rate… so people don’t have limited use, but you’re also not getting the heavily discounted rate the “normal” plans are at.

I’d disagree wrt “modify”. There are all sorts of tools for modifying LLM weights (ie to remove refusals, remove layers or experts, merge models, finetune, and more) and a quick glance at huggingface or civit will show those in very active use.

I don’t think the hardware requirements are relevant. If a research lab publishes the code their particle collider runs under the GPL, that doesn’t make it not OSS even though they’re the only ones on the planet with the hardware to run it.

I feel like I have a different $20 plan than everyone else. I have no problem hitting my 5 hour and weekly limits. Don’t get me wrong, it’s a great deal compared to API pricing, but it’s a far cry from “unlimited”.

> There needs to be a time when you are completely undisturbed and disconnected. If you are disturbed by work you will think about work while you answer and maybe even after that. That's not good.

IMO this is not a universal truth - I’m sure some people need that level of disconnection, but I don't find I'm one of them. I generally like my job, and don't find that forcing myself to disconnect does me any particular mental good. But other people report needing that separation, and that's fine! I don't think there needs to be a one-size-fits-all answer here.

I do agree with your bus factor argument though.

I think my POV on this is a bit different than what others are expressing… I don’t mind answering the occasional email while on vacation, but I view it as a fair trade - as long as the company doesn’t mind me handling the occasional personal obligation during work hours I don’t mind handling the occasional work obligation during personal hours. If the company wants to be strict about clock in/out hours or taking PTO for every 30 minute errand or the work trends in a way that routinely exceeds 40 hours per week total then I’ll stop doing work “off the clock”, but so long as they’re willing to be reasonable I’m willing to be reasonable.

It's 100 lines of unobfuscated js. If you're worried just read it.

> writhing knife

Minor/pedantic, but it’s “riving knife”: https://en.wikipedia.org/wiki/Riving_knife

Is there anything to read on how the economics of an orbital datacenter make any sense? Because I don't really see how blasting a server into space solves any of the typical issues associated with datacentres beyond easier access to solar.

That’s how I use it. I might be working on two or three features at a time (iterating, iterating, iterating…), but they’re all scoped and of user value; I don’t feel that I’m just off chasing rabbits.

But I’m also one of those people for whom the “fun” was always solving human problems rather than solving computer problems. I can see how if you are in the latter category AI has already sucked out a lot of joy and how rapidly project switching could be the least-unfun option.

A library with a bunch of different providers doesn’t solve the payment/billing problem (which is one of the main openrouter benefits). IMO being able to buy credits and not have them locked to one provider is worth the 5% to me.

From your Wikipedia link:

> Demands: […] Government to ensure at least 50% profit over their overall cost of production.

They demanded 50% guaranteed annual RoR on all farming activities? That’s a wild demand.

They're different sizes, it's in the description and it's also described under "More info" > "Capacity (cl)". I'm not seeing the "What is the capacity..." button you mentioned.

I love that even something as well-trodden as a kettle design can still be refined further. My main skepticism is whether this works well when the kettle isn't close to full - the image of it pouring implies a nearly-full kettle, but does the angle become awkward when it's closer to empty? Hard to know without actually holding one.

While web accessibility is important and something we should be investing in, I do feel that the vendors of accessibility tools are somewhat to blame here in how friggin difficult it is to actually make something accessible. Quirks and features are wildly inconsistent across tools, and feature uptake is much slower than it should be. For example, creating an accessible dialog shouldn’t be a multi-page essay to explain, it should just be “use the <dialog> element.” - but the a11y tools are so inconsistent that you can’t just do the standards compliant thing. And don’t get me started on roving tabindex techniques (for things like data tables), which are at best an ugly hack that the entire industry has collectively decided “eh, it’s good enough”.

Even what's described in the article basically boils down to "You can label things, but not generic things (for some reason?), unless that generic thing is a <section> or has a popover attr in which case it magically works." And this isn't even one of the "hard" accessibility things!