I was lucky to upgrade my GPU before the price increases, but the open weight models are quite alright for the moderate stuff, and electricity is much cheaper than the cloud token costs at the moment. So, unless HW prices will be forever high and that bleeds into the 2nd hand market too, I don’t think there will be at least some economical viability.

I think most of the time with these kind of problems we are comparing apples and oranges. Sure we are both using LLMs to make fruit, but what kind of fruit? I’ve found models are best at what they are most trained for but as soon as you want Kiwi and they aren’t as popular as apples you need to be on your A game. I think people who say LLMs are really good are doing trivial stuff or the most stock standard thing you can think of, like making a new portal web app with the most ubiquitous technologies that have been around forever.

I think it’s great that they set it on open math problems for this reason. It’s relatively straight forward to prove/disprove if it did anything novel or just combined a bunch of things it’s trained on like i’d expect.

Ah, that’s probably it. And yeah, governments need to figure out how to manage this. I think it’s destined to be bound by whatever is actually economical post hype. Not pending efficiency breakthroughs of the likes of mores law.

I am a software engineer by trade, so biases’ being what they are. But what I was more trying to demonstrate is that it is provable. Not that making the proof is easy.

hmmm, I just copied from Wikipedia as my primary source. That was the faux pas that was cool before AI.

I know you said all, but I’m curious which part do you think is horse shit? Do you think Torvalds is lying about using AI all together? or just how useful AI actually was? The commit with the alleged AI use is in the an RC for the next release, and it is authored by Torvalds.

Code is probably one of the easiest things to verify, for that reason I struggle to see any other viable uses for LLMs tbh.

user: write code that correctly asserts what 1+1 equals.
LLM: assert_eq!(1 + 1, 4);
LLM: hmmm, that crashed.
LLM: assert_eq!(1 + 1, 3);
LLM: hmmm, that also crashed, I must be doing something wrong, let me try one more time.
LLM: assert_eq!(1 + 1, 2);
LLM: that worked!
LLM Response: assert_eq!(1 + 1, 2);

user: what does this sentence mean? "Buffalo buffalo Buffalo buffalo buffalo buffalo Buffalo buffalo"
LLM Response: This doesn't mean anything, or someone really likes buffalo's

I can always run the code to check, I still don’t know how that sentence makes sense.

Exactly. I think npm and cargo just make Dependency (mis) management more accessible. You still need to choose worthwhile dependencies, and if you don’t want to update frequently, dependencies with stable long term support.

I think the best analogy to highlight the drawbacks of Rust is to say it’s like “Good test coverage.”. If you read that as 100% test coverage you can spend a lot of time doing some trivial overhead that probably isn’t doing a whole lot. But, if you have half a brain it’s probably not particularly harmful. If you read that as effective test coverage, you probably know the problem quite well and it’s probably some overhead, but you get to feel a bit more confident and have some executable documentation.

Neither of which I have heard to be useful or converted in the game industry so it sounds like a waste of time.

That said, if you’re like me and that overhead sounds nice, then you might like Rust :P

The problem of not teaching the less experienced and just doing the work yourself has always existed. Perhaps to a lesser degree in some industries that mandated it by necessity/demand. AI makes it more convenient/viable or reduces necessity/demand but the lesson is the same and will be learnt again, at least eventually.

The edge case is software development, for the bog standard stuff I write at work it probably helps me get work done about 15%-30% faster and with costs of $200-$400 USD a month the math actually works out. At current prices anyway. Also the electricity costs of running open weight models is even cheaper and if there is a pop HW prices may make that an obvious choice too. The worst case IMO would be a slow boil of prices to recoup the unsustainable growth then a slow decline to wherever normal ends up. In that case everyone is left indignant.

Agreed, However I also wonder how or if LLMs could help with what we don’t know we need. With the processing pace of LLMs and the massive context the can processes at once. It seems like it could make the broader connections that are usually the harder to find more accessible.

Pretty sure all my repos are MIT licensed for the betterment of everyone, but I’m not on the list! So I guess I’m not good enough, or they are failing to follow the attribution clause of it.

I didn’t know until recently that the US had such an effective blockade, so enemies of the US are Cuba’s only hope, this is great to see.

https://youtu.be/5Dvu7qiM5rg

If you cut out buffer bloat it’s easy to forget that everything is only slightly slower, it’s the inconsistency in the latency that makes you really notice. With a base of ~130ms when it doubles for those first couple of packets of every page load you really feel it, as long as it’s smooth, your brain can adjust.

I have it for my whole house and no one notices/complains.

https://www.waveform.com/tools/bufferbloat https://github.com/lynxthecat/CAKE-autorate

That said, it’s kind of niche then more niche to run it on the VPN too, I wish it were more accessible.

I wouldn’t take it as far as paying allows demands. If I decided to pick up some litter in my way in public and some sees it. Just because they shoves a $20 in my face, doesn’t mean they can demand and expect I’ll pick up all the other trash around. There would need to be contract (even just Social) negotiations at the least!

In the last couple of months I just bought 12 months through a voucher that is no more so doubt I’d get a refund, so I’ve got some time to decide. I swear everything I look into deep enough has something going wrong with it, but I guess all we can do it take the battles we can handle.

Sure, that’s fine, I tend to agree, I was drawing out the picture of the collateral, so it’s considered as much or as little as it is worth, but at least considered.

I read it as he implicitly does understand that and is the reason he suggests getting a refund if that is your priority. He is just urging people to see nuance that there is collateral damage for the rest of the people that make mullvad too. A loose analogy would be, I disagree with Trump and I wish my country would divest from the USA, which is all well and good, but the workers don’t have great protections and would likely be the first to hurt. Sure a majority of them voted for him I could rationalize, but that’s not the same rationalisation companies use to layoff before they actually start to hurt.