[mechagodzilla]

But the open models create a (rapidly rising!) 'floor' - models that are worse/less capable than the best open-weight models effectively have zero economic value to their creators, even if they just spent $10B to create them. What fraction of queries require a better-than-R1 answer? Every time that floor rises, it puts even more pressure on whatever narrow, temporary advantage these closed companies can achieve to actually justify the valuations.

[lanstin]

All the interesting benefits of LLM/generative AI grow exponentially with the intelligence of the model. There's no point in a 90% capable model, except something like replacing phone trees or extremely bad customer service.

A model that is 50% less likely to try to prove false theorems tho is much more useful than than one that happily tries.

I'm not sure that HN folks appreciate this, but making commercial software is way easier than math and science (all the evidence you need, except the experience of understand QFT or advanced mathematics) is found in HN itself.

[nilkn]

I think for at least the next several years at minimum we’re going to see people continuing to integrate large language models more deeply into more aspects of their workflow. Many people will become used to what’s only possible with the absolute best and will simply lose interest in second rate models. I have no doubt that at some point this process will stabilize, but we’re definitely not close to that point yet.

[ndiddy]

The problem I see is that all the current AI leaders are spending way more than their revenue in training costs to stay ahead of the open source models. As an example, in 2024 Anthropic made $908 million in revenue, but spent $5.6 billion on training. If they cut down on training costs to become profitable, it would only take a year or two before open source models outperformed Anthropic's models, essentially making them worthless. With this in mind, I don't see how it's possible for Anthropic to generate a return on investment.

[somanyphotons]

I don't see how it's possible for the open models to continue to get better without similar levels of extreme money expenditure on training hw. Someone is paying