durumu

I think that short film is AI generated. I only watched like 30 seconds of an office scene in the middle but it spontaneously changed from daytime to nighttime with zero explanation.

Even if most of the code you write is solving repetitive plumbing tasks, today's models are incredibly bad at API design taste. IMO designing software in a way that minimizes side effects and is easy to change and test is more than 1% of software engineering.

Lately most of the code I write has been through LLMs and I find them an enormous productivity booster overall, but despite the benchmarks they're not expert human level quite yet, and they need a LOT of coaxing to produce production quality code.

As far as things LLMs are bad at, I think it's mainly the long tail. I'm not sure there's one singular thing that >1% of programmers work on that LLMs suck at, but I think there are thousands of different weird sub-specialties that almost no one is working on and very little public code exists for, thus LLMs are not good at them yet.

Yes, LLMs are currently useful and are improving rapidly so they are likely to become even more useful in the future. I think inevitable is a pretty strong word but barring government intervention or geopolitical turmoil I don't see signs of LLM progress stopping.

I think that's more reflective of the deteriorating relationship between OpenAI and Microsoft than an true lack of demand for datacenters. If a major model provider (OpenAI, Anthropic, Google, xAI) were to see a dip in available funding or stop focusing on training more powerful models, that would convince me we may be in a bubble about to pop, but there are no signs of that as far as I can see.

I like those vehicles, honestly -- delivery trucks are going to park in the bike lane regardless and these are much smaller and safer to maneuver around. I want to see more of them and hope it leads to more bike lanes being built in NYC.

Neovim is fully backwards compatible, no? I'm not sure what the downside of switching is.

Most LLMs do this due to the proliferation of ChatGPT-generated content in the training data.

I doubt there is a service that bundles a bunch of API access for one subscription fee and works with vim. But there are a few plugins that provide cursor like functionality and let you bring your own API key. Avante and code-companion are the most widely used ones. Magenta.nvim looks promising.

I think AI capabilities perception in general is being greatly damaged by the Google search AI summary. Whatever model they use is so cheap and crappy, yet I can't opt out of it or even get my eyes to skip the box... Claude or Perplexity or whatever can comfortably and concisely answer questions about Auckland holidays without hallucinating, yet the Google search AI thinks you can eat rocks and put glue on pizza, and I see people trot similar examples out all the time to prove that "AI is dumb".

In Manhattan ebike access is excellent -- there are tons of bike lanes and bikeshare stations. They are typically as fast as Ubers for getting around the city since traffic is so bad here, and much cheaper. The main issue is that it's not very safe. Probably this does not generalize to most other US cities.

I agree NYC is not wisely spending its $100 billion per year, but I think the congestion tax makes sense as a way of pricing in externalities. As a non-car-owner in lower Manhattan I dislike passenger cars -- they make it much less safe for me to bike around, and less pleasant for me to walk around. I think most people here benefit if we have way fewer large vehicles in the city, so the limited spots should be reserved for people who get immense economic value from them, like truckers or movers, not random people from the suburbs who want to have dinner in the city.

Reminds me of competitive programming, a la Codeforces or IOI, where you solve incredibly challenging algorithmic problems that are wrapped in some silly story about a cow in a garden or something. (In my opinion, that is part of the challenge and fun!)

I believe Reddit in particular actually has gotten much more optimized in the past 15 years. I don't think this is rose colored glasses, the content really is much more engaging and addictive, with more short form videos and content that can be understood immediately at a glance.

I think that has more to do with being a kid vs being an adult. Kids are probably still buying gfs on Roblox and Minecraft today (disclaimer: I have no idea what kids play these days lol)

Sure, but the current 2-3% annual growth rate is probably not going to hold if we invent actually powerful AI in the next decade. I imagine a step change in the exponent.

I agree the average tutor is better than the average lecturer. But if I watch a YouTube lecture, then I might have access to the best lecturer in the world, or at least a 99%ile lecturer, e.g 3blue1brown. This only works because of the scale -- it wouldn't be possible for millions to get access to tutors this good 1 on 1.

I think it's not clear which way that effect goes -- the Cold War was also the closest humanity has ever come to destroying itself. If all nations depend on each other, there's less nuclear conflict risk and less risk of war in general.

I feel like this comes up with me in programming too! Like if I write some really beautiful function as part of solving a problem, I will be a lot sadder if it doesn't make it in, sometimes to my detriment. Similar energy to "cattle not pets".

IMO using a lowercase l as a numeric variable is cursed. It is way too easy to read it as 1 in many monospace fonts.

I think it's i & 32. Agreed on that being more readable.

I've used NOCOMMIT. Less verbose, equally clear.

I hope that at some point most people will acknowledge implicitly that the fewer the number of measures the more probable is that it is a simplification that hides stuff.

But the larger the number of measures, the more free variables you have. Which makes it easier to overfit, either accidentally or maliciously.

That seems way too high. I live in Manhattan and I pay $2.50-$2.75 for onigiri depending on the filling.

I think there's a useful distinction between plausible-seeming text that is wrong in some subtle way, vs text that is completely fabricated to match a superficial output format, and the latter is what I wish people used "hallucination" to mean. A clear example of this is when you ask an LLM for some sources, with ISBNs, and it just makes up random titles and ISBNs that it knows full well do not correspond with reality. If you ask "Did you just make that up?" the LLM will respond with something like "Sorry, yes, I made that up, I actually just remembered I can't cite direct sources." I wonder if this is because RLHF teaches the LLM that humans in practice prefer properly formatted fake output over truthful refusals?

Which came first, the computers or the code?

The people who like living in dense cities already live there.

One would then think that dense cities would build more housing, but that doesn’t seem to happen. NYC, SF, and Boston are the densest cities in the country and build barely any housing compared to areas like Houston and Austin.

I also just think it isn’t true that everyone who wants to live in a dense city does. My experience is that a lot of people would absolutely love to live in NYC but can’t afford it, and I think this is primarily because there is not enough housing.

Was this the one you were thinking of? https://eieio.games/nonsense/game-12-stranger-video/

That graph is productivity growth, and it's consistently positive.

Here is the model here:

- Apple's software (particularly macOS) is extremely high quality. Best game in town.

- Their hardware is also high quality, but it is often overpriced.

- You must use their hardware to get the best experience using their software.

- Therefore, people might end up buying, say, the Apple XR display or AirPods Max even though they are overpriced compared to other monitors/headphones, because it gives you a better experience using macOS.

I am a pretty huge Apple fan but I think this is pretty unfortunate, not gonna lie -- it kills me a little that macOS doesn't look sharp on my 4K monitor. But it isn't an iCult, it's just that the cheaper thing doesn't work as well with macOS.

I'm not sure what you mean -- I would think the ISP is analogous here, so Comcast, AT&T, Verizon et al?