None of this sounds credible.
Most "news" is just a dramatized press release these days. The conditioning of the public to viewing media as authority creates a mental trap where you're told something is important/credible, despite having zero hallmarks of credibility. And we perpetuate this because it's familiar, not correct or accurate.
This is exactly why humanity continues to stumble, despite centuries of evidence as to what doesn't work. We just can't accept that most of what we tell ourselves is reality is just a for-profit (or for-influence) product that masquerades as knowledge or important information.
Anthropic raises money from IPO -> How long will that sustain them given the immense cash burn? Forget about gross profit - that's the number they want you to focus on and can be played with. Keep the focus on the cashflows -> if a firm runs out of money its over, who cares what your revenues and gross margins are. We are not talking Amazon level loss's which were paltry AND they were able to raise more than enough to keep going for awhile.
Management of firms who pay rent to LLM models need to reign in costs -> they will use a bundle of chinese +american models (this is already happening look at the data) -> affecting revenues, earnings and cash flows of frontier labs. Given the losses in absolute terms, any stunting of the growth rate would be catastrophic for the story re. valuation.
The frontier model story rests on one giant assumption: lay-offs must happen en masse at some point. It's not happening and in my view won't happen - there simply isn't enough trust. Moreover, do you think firms are going to trust these model providers, given its in their interest to steal IP? lol
Not rocket science mate.
First, a lot of usage can switch to open models (like is happening, I agree), and yet frontier model usage can still grow (like is happening). As long as the market for frontier level intelligence keeps growing, the cost of frontier intelligence is marginally profitable to the user, and the open weight models don’t catch up, I don’t see any reason for frontier level usage to fall. Indeed, we haven’t seen any of that so far, despite the rise in open weight usage. My own product has started using open weight models, but my usage of frontier models also continues to grow, because they keep unlocking new capabilities. Maybe that’ll end, but I’m skeptical. I think the problem space that can benefit from the best intelligence money can buy is unbounded.
Plus the labs can still offer cheap non-frontier models to compete with the open weight models (like Luna and Haiku), and they have a massive advantage in terms of available compute.
Secondly, I don’t think that the big labs need mass layoffs to justify their valuation. That’s perhaps one path, but in my view a more likely one is that we get greater increases in productivity and economic growth, and they capture a large share of that. If global GDP grows 2% faster than it otherwise would over the next decade, that’s an extra $175 trillion in economic value created over the next decade. If it grows 5% above baseline growth, it’s half a quadrillion. GDP in 2036 would be more than $100T higher per year than if GDP grows at historical rates. AI companies don’t have to capture very much of that to justify a $2T valuation.
Now, will any of that happen? Who knows, it’s definitely unrealistic and unprecedented, but then again, so is much of what I see happening in the world today. Time will tell.