nimithryn

Oh ok. So I'll just have to apply 4-5 times to every job to be sure I'm considered. Sounds like a good equilibrium!

I think that the problem is, at the end of the day, the engineer must specify exactly what they want the program to do.

You can do this in Python, or you can do this in English. But at the end of the day the engineer must input the same information to get the same behavior. Maybe LLMs make this a bit more efficient but even in English it is extremely hard to give exact specification without ambiguity (maybe even harder than Python in some cases).

Also Christianity was a Levantine religion, growing fast in Greece, Asia Minor, Egypt, Syria, and the Levant. The Eastern Roman Empire at the time was far richer, more urbanized, closer to the grain supply, and closer to Silk Road and Black Sea trade access.

Based on some of his other actions (like moving the capital), I'd speculate Constantine's conversion was part of a deliberate pivot to focus on the East (which paid off, as the centralized Byzantine state went on to survive much longer than its Western counterpart). I'm speculating because inferring his intentions has no historical evidence afaik.

I think there are more possible answers? Jason's mother differs depending on the author...

For example, Jason's mother was Philonis, daughter of Mestra, daughter of Daedalion, son of Hesporos. So Jason's maternal great-grandfather was Hesporos.

I agree that it felt hard to read. It also doesn't make sense that they're fishing in a storm. But from a prose perspective I don't think it's cringe, which is an improvement from my expectation. I'd share some of the writings I think are terrible but I don't like to pick on people.

If R1 one-shotted this then I revise my opinion somewhat. This doesn't give me a gut "this is awful" emotional reaction (although I don't think it's good - it's pretty cliche, and I found my eyes glossing over pretty quickly).

I was somewhat turned off of DeepSeek (the first few questions I gave it, it returned 100% hallucinated answers). But maybe I'll have to look into it more, thanks.

Am I the only one who thinks that R1 is awful at creative writing? I've seen a lot of very credulous posts on twitter that are super excited about excerpts written by DeepSeek that I think are absolutely absymal. Am I alone in this? Maybe people have very different tastes than I do?

(I have no formal training in creative writing, though I do read a lot of literature. Not claiming my tastes are superior - genuinely curious if other people disagree).

Any citation on the additive random noise?

An example:

I'm looking for a new job, so I've been grinding leetcode (oof). I'm an experienced engineer and have worked at multiple FAANGs, so I'm pretty good at leetcode.

Today I solved a leetcode problem 95% of the way to completion, but there was a subtle bug (maybe 10% of the test cases failing). I decided to see if Claude could help debug the code.

I put the problem and the code into Claude and asked it to debug. Over the course of the conversation, Claude managed to provide 5 or 6 totally plausible but also completely wrong "fixes". Luckily, I am experienced enough at leetcode, and leetcode problems are simple enough, that I could easily tell that Claude was mistaken. Note that I am also very experienced with prompt engineering, as I ran a startup that used prompt engineering very heavily. Maybe it's a skill issue (my company did fail, hence why I need a job), but somehow I doubt it.

Eventually, I found the bug on my own, without Claude's help. But leetcode are super simple, with known answers, and probably mostly in the training set! I can't imagine writing a big system and using an LLM heavily.

Similarly, the other day I was trying to learn about e-graphs (the data structure). I went to Claude for help. I noticed that the more I used Claude, the more confused I became. I found other sources, and as it turns out, Claude was subtly wrong about e-graphs, an uncommon but reasonably well-researched data structure! Once again, it's lucky I was able to recognize that something was up. If the problem wasn't limited in scope, I'd have been totally lost!

I use LLMs to help me code. I'm pro new technology. But when I see people bragging on Twitter about their fully automated coding solutions, or coding complex systems, or using LLMs for medical records or law or military or other highly critical domains, I seriously question their wisdom and experience.

Agree. Also, Kubrick's other works strongly emphasize lighting, composition, music, and effects over plot. Just look at Barry Lyndon's candlelit scenes (which look like paintings by the Dutch masters), Eye's Wide Shut's Christmas lights, Clockwork Orange's use of Beethoven, or 2001's match cut from a bone to a spaceship. You can't achieve those in a book, because they are purely audiovisual phenomena. And movies have different constraints: like time limits.

King doesn't have to worry at all about what the score is, or how to film a wall of blood. It's simply not part of the medium. The requirements and constraints are completely different.

Amazing quote. Can you share a link to the article, if it exists?

> It's not about efficiency. I think the current system is more efficient (hence economies of scale). It's just that the rent extraction/profits are concentrated among the platform owners rather than spread out among local venue owners and artists.

Chiming in again. In some ways, this is "efficiency". Serving the globally best stuff is more efficient. The problem is we are trading exploration/variety for exploitation/efficiency. Reasonable minds may differ as to how that should be handled.

> You've ignored the fact that discovery is now a new service component that really didn't exist before. We can tag music and artists and we can automatically discover related music and artists very easily. All the major services do this. Think of an expanded and federated (all the way back to the artist) open source version of this.

This is a good point. We have gone from diffusion-constrained to search-constrained. No clue what to make of that. I would submit to you that the amount of time people have to consume art has not changed, and that improved search now pushes people to consume the globally best art rather than niche content. Are there any domains or platforms that you think do a good job of surfacing specific niche content? Or am I misunderstanding your point?

>There's no reason to think that the previous model somehow produced the most efficient way of "determining who people listen to." It was almost certainly suboptimal. It's no surprise then that this suboptimal model was more or less copied onto the internet and all new players bought out so no new model could be revealed.

It's not about efficiency. I think the current system is more efficient (hence economies of scale). It's just that the rent extraction/profits are concentrated among the platform owners rather than spread out among local venue owners and artists.

> The lack of available artistry and market for it extends through multiple levels, and in this case, right into the software stack that _should_ exist for artists to self publish and for listeners to automatically discover.

I agree it should be possible, the problem is setting up the incentives correctly.

edit: clarity

> My dim view is that it's just a rapacious monopolistic society. Why are there only three (four-ish) cell phone carriers in the US? Why are there only two app stores? Three (four-ish) major music platforms? The internet should have brought a diversity of choice that has failed to materialize because enforcement of monopoly laws in this space has been nascent and aggressively fought against.

I think some of this aggregation is due to economies of scale, though.

> You could literally grab 5 guys from the mall and __force__ them to be a hit in a few months.

This is sort of a corollary to my hypothesis. Since we have a megasociety, everyone has access to music from everywhere, but limited time. There’s too much music being produced. So we now have a signaling game (advertising and marketing) to determine who people listen to. Signaling games will naturally be dominated by capital.

In terms of breeding grounds for local music, where are you thinking? Nashville?

My hypothesis: Lack of geographically local experts.

We've merged more and more into a megasociety. In a geographically distributed society, the power law for quality is more forgiving. You can be the best local band and make a living.

Now that we have ultra-efficient communications, there's less room for "mid-tier" art. Local art gets outcompeted by whatever the top stuff is among a much greater national or international population.

Because of this, there's also less of a breeding ground for maturing artists, or for experimental styles in isolated areas (think California surf rock, or NYC Salsa, etc). There's no place to go if you're not the best.

Also, this sort of thing couldn't be done in the 80s or 90s, because it was much harder to compile that much data.

On the one hand, yeah, this guy is to some degree complaining on the internet because he was rejected. It’s annoying: No one likes listening to complaining.

On the other hand, a) this guy does have a point - it’s very very strange that you can have all sorts of interesting, highly relevant expertise and tech interviews don’t care at all, they just ask you generic stuff that has nothing to do with the job, b) he’s allowed to complain c) you’re complaining about him complaining.

I see a sort of weird logic all the time (usually around interviews or dating) where someone is complaining on the internet about how the system is broken and they are subsequently accused of some moral failing (being “petty”). I don’t think you know anything about this guys attitude or how this guy works on a team - you don’t know him!

I guess the lesson is “don’t complain on the internet”, which I suppose I agree with.

I guess I’m complaining about people complaining about complaining, so I’m part of the problem, too. But seriously, let’s engage with each other in good faith and not just assume someone’s a bad person because they’ve been hurt by rejection.

Edit: grammar

Yeah, this sequence is a meme commonly cited to show "choppy modern editing"

I have chronic insomnia and weed definitely keeps me up

Yeah, also, you tend to need to know the answer before you prompt it.

It's kind of like a Ouija board, in that you get out what you expect.

My experience as well. Heavy GPT-4 use (for a variety of things). Great for boilerplate, great for retrieving well-known examples from documentation, saves a fair amount of time typing and googling, but often completely wrong (majorly and subtly) and anything non-trivial I have to do myself.

Great tool! Saves a ton of time! Not a dev replacement (yet)

Have you tried to build a system like this? It’s not trivial

I’ve used GPT4 (text) heavily as part of my business, including for front end stuff.

The technology is very impressive - but honestly Twitter examples are super cherry picked. Yeah, you can build some very ugly, basic front end web pages and functionality right out of the box. But if you want anything even slightly prettier or more complicated, I’ve found you need a human in the loop (even an outsourced dev is better). I’ve had GPT struggle with even basic back end stuff, or anything even a bit out of distribution. It also tends to give answers that are “correct” but functionally useless (hard to explain what I mean, but if you use it a lot you’ll run into this - basically it will give really generic advice when you want a specific answer. Like, sometimes if you provide it some code to find a bug, it will advise you to “write unit tests” and “log outputs” even if you specifically instruct it to find the bug).

Plus, in terms of capabilities, tools like Figma already have design to code functionalities you can use - so I don’t think this is really a change in usable functionality.

Of course, the tech will get better over time.

It’s about trust - not publicity. Trust is hard to earn back once broken, and there will be multiple offerings eventually.

For example, AWS was one of the first cloud providers. Now there are alternatives, but I still pick AWS because I trust them not to break my dependencies way more than, say, Google

Example? I'm curious

The redistribution is productive, because by redistributing risk (not just among people, but also across time), some ventures that were otherwise not feasible become feasible. For example, you want to build a house - but you don’t have the cash. A bank gives you a loan. They take the risk that you won’t pay them back, you get a house, and return they get a premium. This benefits many stakeholders (you, the bank, the builders, etc). If the bank has too much risk, they can off board it to someone with deeper pockets and a more diversified portfolio.

I worked for a government contractor before moving to FAANG. We had transparent salaries at the contractor (it was a small company). As a senior engineer in FAANG, I make more than the president of the government contractor.

Yes, and that would violate copyright

It has the Super Bowl numbers wrong, too. The last Super Bowl is LVI, which was Rams vs Bengals… the Super Bowl before that one was Tampa Bay Buccaneers vs Kansas City. it has every fact wrong but in the most confusing way possible…

IANAL either, but there are attacks you can use to extract the original training data from the models: https://arxiv.org/abs/2301.13188. I’d guess these attacks will get better with time.

So in some ways, you can argue that Stability is also directly redistributing the original images (albeit in a compressed format).