Hacker News Reader

Comment by thundergolfer | Hacker News Reader

I just can’t handle this guy’s “it’s a tulip bubble!” approach when just yesterday Thinking Machines released a demo which would have been taken as literal magic if it was shown 20 years ago.

The tech is astonishing and Ed spends 98% of his writing acting like the industry is a scam.

View on HN

This entire article spends a lot of time hiding the ball and conflating what I'd distinguish as "data centers" vs. "data center capacity". Existing data center capacity is being allocated to AI because that's where the demand is, this is in addition to future data center capacity still in the process of being built. Microsoft and others are playing word games and combining both numbers for PR purposes, which I think Ed knows and is being obtuse about.

It's also annoying how he keeps using the term "bubble" as a sleight-of-hand synonym for, I don't know, "useless and fake". Tulip valuations vastly outgrew their actual value and never made it back up. The Dot Com bubble was just early and investment outpaced real-world adoption. Nobody looks back on dot com and thinks claims like "nearly all commerce and social interaction will move online" are silly because that's exactly what happened. AI investment might be at outsized levels right in this moment, it remains to be seen, but even if it doesn't get any better from here it's already insanely useful. That won't "pop" in any meaningful way that he predicts.

EDIT: To say nothing of the overwrought 2023-era comparisons to cryptocurrency. It was obvious at the start that crypto was a solution in search of a problem, shaded by the fact that most advocates benefited strongly from convincing everyone else how valuable of an idea it was. Anybody has been able to spend 20 minutes with free ChatGPT for years now and immediately start to grasp the real-world applications of tech that genuinely replaces a substrate of knowledge work.

"Insanely useful" doesn't mean hundreds of billions of dollars of upfront investment from our largest corporations is prudent especially when proponents of AI don't seem to approach the risk honestly. The argument isn't just that AI is going to take over the world, it's that the consequences of massively over-investing in it could lead to a broader economic collapse more like the Great Depression than the Dot Com bubble. I'm not going to predict what might happen, but I don't think the Dot Com bubble is necessarily predictive of any outcome here, either.

This is not money that would otherwise be going to agriculture or energy; one of the few salient points that Ed makes is that a lot of these deals are circular and based on invented money. I think the comparison to the dot com bubble is sound purely because basically every DC investment that exploded would have been legacy-defining investments if they'd been able to stay solvent until like 2010. Some companies and some people are going to lose, but they're not going to have been wrong, just badly timed.

Tech companies sitting on godlike amounts of money and being able to invest such huge sums in AI at a nascent stage is worrisome. There is absolutely nothing to indicate that this is anything close to an optimal distribution of investment in the economy. It's not a plus that some of these companies are playing an investment shell game. They don't hand out awards for badly timed investments, they hand out bankruptcy rulings. It's fine and good if a few speculative companies go bust. Nobody will think it's fine when companies ingrained into the fabric of the economy do. At the end of the day it's clear we've learned very little from previous crashes and I think it's a false assumption to think we'll suffer equivalently for it.

As someone who's admittedly anti-AI, what knowledge work does it replace? It seems to me it supplements some knowledge work, while outsourcing actual intelligence to the human operator.

IMO it seems like most AI intelligence is just a Clever Hans situation: the AI produces a stream of responses, and the human selects the one that is correct, then they conclude that the AI is intelligent.

Let me prefix by saying that I'm solidly in some kind of AI middle ground. I think people who are fully outsourcing everything they do to AI are insane, and I think people who have planted their feet and are pretending AI is useless are also insane.

The way I think about it is that a lot of what we considered knowledge work isn't anymore. In "the before times", I would have considered it knowledge work to know how to dig into an unfamiliar code repo or long document and produce a useful summary of the information within, or identify which parts of a codebase are applicable to a given problem I'm trying to solve. AI turns semantic search up to 11; you can point it at an unfamiliar repo and say "what do I have to touch to make this work" and get a 90% accurate result. That's insane magic. I think if the bar is to consider it not a replacement for knowledge work as long as there is a human in the loop, then we're not there yet, but it keeps eating away at more and more of the basic pieces.

> I think if the bar is to consider it not a replacement for knowledge work as long as there is a human in the loop.

That's where I put it personally, because of humans' limited amount of useful focus during a work day.

Anything that requires human attention will take some of that resource, and don't think models' rate of improvement will be fast enough to overcome that in the near future. Reviewing an output that is 99%, 99.9%, or 99.99% correct all take about the same amount of time, so the output needs to be correct enough not to need review before any knowledge work is replaced.

I’m afraid your numbers, all over 99%, are anchoring the conversation to an unreasonably high quality level.

I would have personally gone for 75%, 85% and 95%, which are all still best case scenario answers.

Had I taken on chatbot advice on electronics or chemistry I’d have died every couple of weeks (doing some hands-on real world R&D in my basement as a distraction from software).

it-bubble investor here. I think you're missing the point. Take a look at a 30 year graph of something from that era let's say Ericsson B. Phone networks are still around and very useful that's totally besides the point as far as valuations goes.

https://www.tradingview.com/chart/?symbol=NASDAQ%3AERIC

Reply on news.ycombinator.com