This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
I don't really see what this is supposed to prove one way or the other. You are still stuck in the timescale framing of the most fervent AI bros. Opus 4.6 came out in February, 2 months ago. So what if Opus 4.7 is not a revolutionary upgrade? If AI were truly stagnant, we won't really find out until someone posts in 2028 that Opus 6.7 is only a marginal upgrade over Opus 4.7.
I think you misunderstand my argument. I'm not arguing that AGI is impossible based on this (though I don't believe it's possible). I'm arguing that this is a strong sign that VC money is drying up before they could ever conceivably achieve AGI (even if it is possible).
It’s an interesting sleight of how the tech bros have hoodwinked the finance bros and duped them out of so much money. The train is being driven irrationally with so much FOMO money going out the door, I’m surprised it’s lasted as long as it has without people asking questions.
I wonder if Michael Lewis has already had a draft in the works of the next big story he’s working on. As long as I can pick up a box of GPU’s for pennies on the dollar, I’ll be happy. Although I don’t know how the resellers are going to pop up for such a massive surplus of inventory, but I’m definitely on the lookout. My home lab is about to get even bigger.
More options
Context Copy link
Anthropic raised $30 billion two months ago, their problem isn’t lack of money. All the VC money in the world won’t solve a bad engineering culture.
Sure, but they're on track to burn $11 billion this year in expenses, and more in the future, so that's not going to last too long
...and $14 billion in revenue assuming zero growth. Or closer to $35B if their 10x/yr trajectory continues.
Note that your link says "run-rate revenue" which is a very different thing from actual revenue. Relevant XKCD: https://xkcd.com/605/
Yes, I did note that in my comment. They had 1/12 of that revenue in 1/12 of the year (or some other fraction), and therefore they're on track to $14 billion in revenue assuming zero growth.
Assuming zero growth, and zero decline in revenue, compared to whatever fraction of the year.
Seems deeply unlikely to decrease given their #1 issue is not having enough compute to meet the demand for their compute.
More options
Context Copy link
Speak plainly, please.
This is the second comment in a row that's just literally restating my points, in the style of a rebuttal. What is your point?
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
If that were the end of the story it wouldn't be an issue. It's that it evidently uses significantly more computing power than the performance improvement would suggest, raising the spectre of rapidly diminishing returns.
It seems to me this also has financial implications. If you are paying per token, and the model's benchmark performance increases slightly, but its token cost to reach those higher benchmarks increases tremendously, suddenly you're paying a lot more to do, at best, slightly more.
If Anthropic is making margin on the token cost, then this is an improvement from their financial point of view, right?
I’m not seeing the mathematics on this one. Care to explain?
Toy model to illustrate:
Let's say that I need to make 100 PowerPoints per year, and I use AI for this. And let's say that when I use 4.6, it costs me $1 in token costs to make a PowerPoint presentation based on a prompt. I now have to spend 10 minutes correcting the errors.
Now supposing we bump up to 4.7, and suddenly the PowerPoint a bit better, I only need to spend 5 minutes correcting the errors. But it costs $2 because the token cost is less efficient.
If Anthropic is making margin on the token costs, then the demand for tokens has increased even though the demand for work has not (I still need to make 100 slide decks annually). And while we've saved me some time, we've increased my cost to $200 instead of $100. If Anthropic is making 10% margin, they've now made $20 instead of $10. And since suddenly the token demand has doubled (in this toy world with static demand for PowerPoints which now cost more tokens) Anthropic can likely use the increased demand to raise costs on compute further.
Some disclaimers:
Possibly there's something (else) I am missing here, would be very happy for feedback. I don't use LLMs to code so my lack of experience with the most-common use-case means I have little personal insight into the trade-offs between increased demand for tokens versus higher performance. If people are complaining, though, I assume it's because they feel like they are able to get less done (IOW, the model is less token-efficient). If anyone has a better model for how this works in the real world, particularly in more common use-cases, I would love to be filled in.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link