Wow I hope this is true. This can only be good for the world.
I'm sure AI will still come but it is too disruptive now. It needs a slowdown. As usual all the greedy investors are to blame.
The AI bubble is popping; we just don't know it yet
111–120 of 154 posts
Re: The AI bubble is popping; we just don't know it yet
#112Earlier quoted context omitted.
No, cost per task is probably not going down on average. The underlying issue is that LLMs are still bad at most tasks they could be used for, so as their contexts get larger and they're able to run longer without becoming incoherent, more tokens are spent on tasks to improve the quality of output. So cost per task is probably going up on average, but so is the quality of the output.
I obviously mean cost per fixed task. Do you agree that if you fix a task, the cost is going down? Meaning you get more work done from the same cost over the years.
Yes, obviously. The same task at the same quality and the same speed got cheaper. That's not what anyone is disputing. The problem is that overall expenses are going up.
Re: The AI bubble is popping; we just don't know it yet
#113Earlier quoted context omitted.
I obviously mean cost per fixed task. Do you agree that if you fix a task, the cost is going down? Meaning you get more work done from the same cost over the years.
Yes, obviously. The same task at the same quality and the same speed got cheaper. That's not what anyone is disputing. The problem is that overall expenses are going up.
If a fruit company developed fruit that was cheaper but people ended up buying more fruits, would you then call fruits more expensive or more cheap?
I assume cheap. Then why would you call AI more expensive?
Re: The AI bubble is popping; we just don't know it yet
#114The sooner it pops, the better. Later on it will be even worse.
Re: The AI bubble is popping; we just don't know it yet
#115Outside a relatively small world of circular investment and FOMO feeding FOMO the general consensus seems to be “let it burn.” It appears very unlikely we will ever see an IPO of OpenAI. Anthropic appears less doomed, but still iffy at best. Tons of other large, but little discussed, AI startups are just dead-companies-walking at this point. The likes of AWS are showing good headline numbers but are taking out massiv…
Anthropic is almost purely a model company. They own close to no data centers. If models are becoming commodities, and the main bottleneck is actually serving them at scale, Anthropic does not appear particularly well positioned. If AWS had a ~60-80% margin for decades, I see no reason why inference can't have a ~60-80% margin for quite some time. The problem is, if costs continue to drop ~90% for the same level of q…
> I see no reason why inference can't have a ~60-80% margin for quite some time.
Same as anything with low margin: Viable competition and low switching costs.
Many sysadmins lazily do 100% AWS because they don't know any alternatives. CFO's might enforce using more economical inference providers if the cost is even 10% less.
Re: The AI bubble is popping; we just don't know it yet
#116Earlier quoted context omitted.
TSLA's 400 PE ratio (along with Elon Musk's trillionaire status) was never sustainable. Where I disagree is with companies running market average PE ratios supposedly being doomed as well. But also, don't count out OpenAI and Anthropic, no matter how shaky the numbers. The market has also demonstrated how to price SPCX so if they IPO, they will see their true FMV and I'm sure it's >0 and much much less than what they…
> But also, don't count out OpenAI and Anthropic, no matter how shaky the numbers. The market has also demonstrated how to price SPCX so if they IPO, they will see their free market FMV and I'm sure it's >0 and much much less than what they believe. For even in the worst case scenarios, they have valuable personnel, experience, and IP deploying AI at scale for what is to come, even if it's based on Chinese open weigh…
I think it is very likely that OpenAI stop operating as an independent entity, but more likely that Microsoft would get full control.
Re: The AI bubble is popping; we just don't know it yet
#117Earlier quoted context omitted.
Anthropic is almost purely a model company. They own close to no data centers. If models are becoming commodities, and the main bottleneck is actually serving them at scale, Anthropic does not appear particularly well positioned. If AWS had a ~60-80% margin for decades, I see no reason why inference can't have a ~60-80% margin for quite some time. The problem is, if costs continue to drop ~90% for the same level of q…
Anthropic is perceived as the leading AI company in the world. Even if they started selling inference alone, people would prefer to pay them a premium for it rather than figuring out how to operate opencode or other replacements. Add to that government contracts, enterprise and other markets, and I think Anthropic would be totally fine even if models plateau. Transistor count has increased exponentially for decades a…
My company has started asking people to convert to open code all the sudden to avoid vendor lock in with anthropic
Re: The AI bubble is popping; we just don't know it yet
#118There is something I have been pondering recently. If we compare the cost of AI subscriptions (let's say Claude's 100/month) to a median developer salary (let's say 100k/year to 200k/year), the difference is orders of magnitude. This fills like a gap that needs to close. I suspect llms are too cheap right now but will raise their prices to a point where only big companies will be able to afford subscriptions to use t…
Why is this a gap that needs to close? Can you elaborate more than feelings? Right behind these big models are local inference with AMD/Apple having a great hardware start and a lot of local sized models making great progress.
Re: The AI bubble is popping; we just don't know it yet
#119I would argue we still have not even really gotten started. What do we have in the decade ahead? Robotics in every household, models 10x+ faster and more intelligent than today. Really no significant impact in life sciences, R&D, and 'offline' world / robotics today as of yet, which is where most of the value will live.
Even if we agree with this take (and I do think it's a likely take that you're right on the long term), it doesn't change that it seems likely we're in a bubble, and it probably will pop. We see a similar paradigm with lots of revolutionary technology. The initial promise is high, people get very excited, lots of money pours in, and.... 15-30 years go by before we start seeing real impact across the economy at large.…
> So take robots: I can promise you that you won't see robotics in every household in the next decade (especially so if we exclude the current market of robot vacuums).
People in 2016 were adamant that the current frontier LLM capabilities would not be achievable in the next decade.
Modern home robots prototypes look awkward and mostly useless the same way GPT-2 was looking like a curious but mostly useless thing in 2016.
I can totally see a capable and affordable home robots that are worth buying for the majority of the population being a reality by 2036. Maybe not _every_ household, but a good number of them.
Re: The AI bubble is popping; we just don't know it yet
#120Earlier quoted context omitted.
Yes, obviously. The same task at the same quality and the same speed got cheaper. That's not what anyone is disputing. The problem is that overall expenses are going up.
If a fruit company developed fruit that was cheaper but people ended up buying more fruits, would you then call fruits more expensive or more cheap? I assume cheap. Then why would you call AI more expensive?
I think you're arguing with the wrong person. Please note what I actually said:
> This is poorly stated in the podcast, but the underlying point is correct: while cost-per-token is going down, overall token use is way up. This is causing the cost of using LLMs in corporations to skyrocket.