throwaway0123_5

If the program being cut is valuable, I would say absolutely. Ignoring that this website is obviously going to have massive bias, a huge amount of the savings listed are for programs I would generally support: Health services, health research, USAID (we saved a billion dollars on "polio immunization," yay), environmental research, education, etc. Maybe the specific contracts cut were wasteful, but they certainly don't provide enough information to determine that and I'm not going to assume the cuts were good because "money saved."

Also, the savings are pretty minor overall. If we trust the website, we get $1.3k saved per taxpayer. The vast majority of the programs cut would have to be completely useless for me to think it is worth it to save $1.3k.

There doesn't seem to be much consensus on defining what intelligence is. For the definitions of at least some reasonable people of sound mind, I think it is defensible to call them intelligent, even if I don't necessarily agree. I sometimes call them "intelligent" because many of the things they do seem to me like they should require intelligence.

That said, to whatever extent they're intelligent or not, by almost any definition of intelligence, I don't think they're achieving it through the same mechanism that humans do. That is my main argument. I thing confident arguments that "LLMs think just like humans" are very bad, given that we clearly don't understand how humans achieve intelligence and the vastly different substrates and constraints that humans and LLMs are working with.

In Firefox my tabs have text, and I frequently rearrange tabs with my cursor. I think this is a pretty common usage pattern (I do it on a daily basis). It would be an enormous pain if most of the area on the tab turned my arrow cursor into an I-beam cursor that I couldn't move the tab with. I checked Chrome and it looks like the tabs work the same way.

While having the text in the tabs is very useful to know what is under them, I don't think I've ever needed to actually copy the tab text. It would be a huge UX downgrade for me (and I think most people) if the tab text was selectable.

Some people might need it to be selectable for accessibility reasons and there should be a toggle for that, but I don't think "absolutely all text everywhere is selectable" is a good default.

I think saying that "LLMs can produce outcomes akin to those produced by human intelligence (in many but not all cases)" and "LLMs are intelligent" to both be fairly defensible.

> I see no reason whatsoever to believe that what your wet meat brain is doing now is any different from what an LLM does.

I don't think this follows though. Birds and planes can both fly, but a bird and a plane are clearly not doing the same thing to achieve flight. Interestingly, both birds and planes excel at different aspects of flight. It seems at least plausible (imo likely) that there are meaningful differences in how intelligence is implemented in LLMs and humans, and that that might manifest as some aspects of intelligence being accessible to LLMs but not humans and vice versa.

> The early narrative was that companies would need fewer seniors, and juniors together with AI could produce quality code.

I don't remember this being the narrative at all. Almost as soon as LLMs could produce useful code the narrative I was hearing among devs was "Seniors + LLMs means you'll need less juniors."

And the narrative from companies (sometimes requiring the smallest amount of reading between the lines) is "You'll need less human employees, period."

Also I think it is plausible (if not likely) that LLMs reduce the demand for other kinds of software. They're (relatively) general purpose and can handle many tasks that previously would've needed bespoke software (and a team of people to create/maintain it).

I don't know much about the science of colors and I imagine this is going to be very subjective but I would at "worst" describe the rightmost OKLCH color as blue-green (but probably just "light blue"). Meanwhile the two rightmost HSL colors I would not describe as blue at all. Second from right I would call "light purple" and the rightmost honestly looks grey to me (which is weird because the caption says the darker HSL values "muddy out towards greyish" but the leftmost values for OKLCH and HSL both look pretty blue to me).

Also if I use the macOS app "Digital Color Meter" I get essentially the same green value for the rightmost OKLCH and HSL colors (226 and 227 for "display native values" and 228 and 227 for sRGB).

I think that is probably true now (in the US), but the hypothetical is that it is made easier to lose your license.

DC recently banned "right on red" but it is routinely ignored and the penalty is apparently a $100 fine. If the penalty was loss of license (maybe not on first offense) I think there's a lot of people ignoring the current rule that would not be willing to ignore the possibility of losing their license.

Doubtless some would ignore it, but you can go to jail for driving on a suspended license. I suspect there are a lot more people willing to risk a traffic ticket and a few $100 in fines for speeding, bad lane changes, etc. than there are people willing to risk jail for driving on a suspended license.

> I would adore to see obtaining a drivers license ratchet up in difficulty in order to remove dangerous human drivers from the road.

I think it would be far more effective to make it easier to lose your license than it would be to make getting the license more challenging.

The absolute most dangerous drivers I see on the road aren't bad drivers in the sense that they're unskilled at controlling their car. I can't weave between cars at 120 mph or cross three lanes of traffic to make an exit I didn't see until the last second without killing myself, but I routinely see people do that. Sure they don't care about driving safely and/or following the law, but they're probably sane enough to pull it together for a brief driving test.

The other big category of dangerous drivers is drunk/distracted (texting) drivers. Again, most of the people engaging in these behaviors are probably smart enough not to do them during a driving test.

> You can say that because it is universal, it should level the playing field just at a different starting point, but you are still creating a situation where even incredibly intelligent people will choose to pursue leisure over labor, in fact, the most intelligent people may be the ones to be more aware of the pointlessness of working if they can survive on UBI.

This doesn't seem believable to me, or at least it isn't the whole story. Pre-20th century it seems like most scientific and mathematical discoveries came from people who were born into wealthy families and were able to pursue whatever interested them without concern for whether or not it would make them money. Presumably there were/are many people who could've contributed greatly if they didn't have to worry about putting food on the table.

> The most intelligent will prefer a system where their superior intelligence provides them with sufficient resources to choose a high-quality mate.

In a scenario where UBI is necessary because AI has supplanted human intelligence, it seems like the only way they could return to such a system is by removing both UBI and AI. Remove just UBI and they're still non-competitive economically against the AIs.

But if AI can do anything that human labor can do, what would even be the incentive for AI owners to farm wheat and sell it to people? They can just have their AIs directly produce the things they want.

It seems like the only things they would need are energy and access to materials for luxury goods. Presumably they could mostly lock the "human economy" out of access to these things through control over AI weapons, but there would likely be a lot of arable land that isn't valuable to them.

Outside of malice, there doesn't seem to be much reason to block the non-technological humans from using the land they don't need. Maybe some ecological argument, the few AI-enabled elites don't want billions of humans that they no longer need polluting "their" Earth?

> Sam recently said only 7% of Plus users were using thinking (o3)

Thinking or just o3, and over what timeframe? There were a lot of days where I would just rely on o4-mini and o4-mini (high) b.c. my queries weren't that complex and I wanted to save my o3 quota and get faster responses.

> That means 93% of their users were using nothing but 4o!

Also potentially 4.1 and 4.5?

It seems like you might need less output tokens for the same quality of response though. One of their plots shows o3 needing ~14k tokens to get 69% on SWE-bench Verified, but GPT-5 needing only ~4k.

I thought so too, but I gave it a screenshot with the prompt:

> good plot for my presentation?

and it didn't pick up on the issue. Part of its response was:

> Clear metric: Y-axis (“Accuracy (%), pass @1”) and numeric labels make the performance gaps explicit.

I think visual reasoning is still pretty far from text-only reasoning.

Right, I get that they make most of their money from sources other than their salary. But that seems like something we should aim to discourage, as it is pretty ripe for abuse. It is clear that many of them trade on non-public information, and for those that own businesses there's a lot of incentive to push legislation that helps the business. So if we were to ban most of that and have them live just on their salaries, I think the salaries should be higher given the challenges/responsibility of the job and the area they're living in.

Yeah $150k in DC is comfortable but far from luxurious (if you have a family, which it seems like most congresspeople do). It would not be enough to incentivize me to do the job of a congressperson.

iirc 4090s have 24GB

If a company in [insert almost any sector] explicitly advertised and provided: "Human customer support agents, no voice menus/AI whatsoever."

I'd almost certainly buy their service over any other company. Seriously, that could be the whole ad. By the time I'm calling customer support, its because the answer isn't easy to find on their website/emails/whatever else. I'm calling because I need to escalate to a human. Sometimes it isn't even about getting information, but rather that I need a human who actually has the authority to help me (which I assume nobody is delegating to AI anytime soon for fear of jailbreaks). Making people jump through ten layers of "I know you want to speak to an agent, but I need more information about your case" is just infuriating.

As someone who rarely does front-end but occasionally needs to hack something together, ClojureScript is a godsend. I get to hijack the react ecosystem for nice UI components while also getting to use a pleasant language and avoiding JSX (I hate anything XML-y).

I doubt there's any age bracket where most people have manually coded a website tbh.

I think they were talking about total time spent working rather than remote vs. in-person. I've seen more than a few studies over the years showing that going from 40 to 35 or 30 hours/wk has minimal or positive impacts on productivity. Idk if that would apply to all work environments though, and I don't recall any of the studies being about research productivity specifically.

> LLMs are the single most general-purpose tool in existence.

Wouldn't this award have to go to computers? They're a prerequisite for using LLMs and can do a lot more besides running LLMs.

From the second paragraph before any mention of Phoenix or Arizona:

> let us go back in time and space, and journey to tropical Brazil in the distant time of 2007

It seems fair to say that it is ~never the overall bottleneck? Maybe once you figure out what you want, typing speed briefly becomes the bottleneck, but does any expert finish a day thinking "If only I could type twice as fast, I'd have gotten twice as much work done?" That said, I don't think "faster typing" is the only benefit that AI assistance provides.

Maybe they're "learning" how to use AI [1], but in at least some cases they're not learning what they're supposed to be learning from the class they're enrolled in [2]. The point of the assignments was never to produce the the answers as fast as possible by any means necessary [3], but to build strong foundations so that you can later do much more challenging/useful things.

If your assignment is "program X in assembly," and you just copy/paste the prompt into ChatGPT, you aren't learning assembly, you aren't learning important concepts relevant for low-level performance optimization, etc. etc. That was the point, not to produce X.

What would be useful is specific classes with assignments that are designed to be worked on with the assistance of AI. "AI-enabled SWE" etc. Lessons on how to effectively utilize it for research and learning. But those things aren't replacing learning the fundamentals, and probably can't unless AGI is achieved and actual knowledge is just economically irrelevant.

[1]: For a lot of undergrad assignments you can probably just copy/paste the description into ChatGPT, which I don't think qualifies as being difficult enough to merit "learning" how to do).

[2]: I guess this is "fine" if you're in the "AI will imminently replace all white-collar work" camp, but then its just dumb to be enrolled in college in the first place, unless maybe you're going for free.

[3]: In some sense, fully outsourcing your thinking to AI is the same as just paying someone to do your homework, which has always been possible but less accessible.

For cities and landscapes in the US it is scary good. I tried about twenty photos from (mostly rural) Central and South America and it is much less effective there w/o a well-known landmark to go off of (likely unsurprisingly). In four cases it got the wrong continent entirely, and in one case guessed as far away as India.

It doesn't seem like it is intended for a manager to use to get information about their direct reports, but rather org-wide information?

NASA apparently has ~18k employees, it seems like it might be useful to be able to query "Who at NASA has X, Y, Z skills that can help us with this project." Then you can speak to some of those people face-to-face. It won't be perfect but certainly sounds like a useful tool in principle.

> The SWE-Bench result carries a bit more weight

Although I have issues with it (few benchmarks are perfect), I tend to agree. Gemini's 63.8 from Sonnet's 62.3 isn't a huge jump though. To Gemini's credit, it solved a bug in my PyTorch code yesterday that o1 (through the web app) couldn't (or at least didn't with my prompts).

I suspect it would be somewhat challenging to do, but I'd love to see something like this where the contributions are bucketed into different levels of difficulty. It is often the case for me that a small percentage of the lines of code I write take a large percentage of the time I spend coding (and I assume this is true for most people).