It always seems smart to be a contrarian/bear, but historically, betting against technology and technological progress is dumb.
Not only is he not a shareholder, but based on his opinions, he clearly isn't a user of modern models either. Somehow he still thinks they are party tricks and that "They will never be AGI," with no elaboration on _why_ they _can't_ be AGI, and why that actually matters for these companies (hint: it doesn't really matter much. As it turns out, you could replace a lot of humans with a modern LLM and have a reasonable comparative advantage in doing so, even though the models are ostensibly not "AGI.").
His predictions have also historically just been bad (https://danluu.com/zitron/). Even AI 2027, which is basically semi-fictional, is more accurate and worth paying more attention to than this guy. And the superforecaseters which actually made AI 2027 and actual futurists are generally higher signal.
Not being a 'party trick' and being 'AGI' are wildly different assumptions. Apart from AGI being a poorly defined term, I don't see any reason why the LLM methodology has the potential to scale to AGI.
There's other avenues which are probably needed to make the cohesive AGI picture come true, such as World Models. Simply learning from text and being able to generate language is not, in my mind, the end-all-be-all for intelligence.
> His predictions have also historically just been bad (https://danluu.com/zitron/). Even AI 2027, which is basically semi-fictional, is more accurate and worth paying more attention to than this guy. And the superforecaseters which actually made AI 2027 and actual futurists are generally higher signal.
As I noted upthread, those predictions don't cover the financial side, which is where a lot of the issues currently are.
Like, LLMs/AI are definitely going to have large impacts, but those impacts will take a long time to materialise (outside of software, and even there we'll need much better tools to make good use of agents), and the financial investment are predicated on near-term returns from the tech that seem wildly implausible to me (and others).
The bunch of crypto miners getting multiple billion dollar valuations should be a clue here that all is probably not well.
And fundamentally, I don't see model building as a very good business. It's incredibly expensive, and it's hard to maintain the margin in a very competitive space. This is probably good for the world, but very bad for the investors into these model companies.
I thinking most people who follow and like Zitron are in it for the AI pessimism, and only some of them care about the financials. He has shifted his focus to the financials, and they are seeing his predictions of failures for these AI and data center companies failures as validation that AI is all hype.
To me the AI stuff is scarily good now, don't see many software jobs holding up for long and other fields will follow, and he might be right about the financials but it doesn't change anything to me the AI is here, getting better and better and is going to take our jobs.
He is arguing financials and companies failing, but his followers see it as AI sucks and is all hype and nothing to worry about.
Idgaf if model building is profitable, the genie is t going back in the bottle and hoping it back in is definitely not the solution
> don't see many software jobs holding up for long
They'll definitely change, but I'd be really surprised if people stopped hiring software engineers/product/data people.
The agents are useful to the extent that you can sense-check them, and that's a big part of what the humans do in this loop.
> the AI is here, getting better and better and is going to take our jobs
As above, I feel it probably won't take software jobs, but the situation in the rest of the economy is far worse. The labs simply cannot get the kind of data that they would need to train the models to do this.
Barring super-intelligence (which I'm not even sure is a coherent construct), I don't see how that happens.
Loads of stuff is gonna change, but it'll take longer than a lot of people around here expect (rather like the Internet did change the world, but the dot com bubble was still a bubble).
They are a party trick, provided the party is software development. If your job does not have you sit behind a computer all day, their utility is minimal.
I would stick to parallelizing first principles. The issue you're describing is when individuals cannot offload responsibilities to agents and other humans because of humanistic ideals. First principles are by definition axioms, and this is the only way to build. It's amazing how quickly heuristics are conflated with first principles when efforts are spread thin.
It doesn't matter how fast you're going if you're going in the wrong direction. Programming can still be a challenge with LLM's, but everyone is panicking and skipping the research portion.
