00:00Think about AI and you are probably thinking about large language models, LLMs, and it is for them
00:05that the huge amount of CapEx is being rolled out this year, some 700 billion in 2026 alone.
00:12A lot is expected of this CapEx. A lot of money is being spent. But LLMs have problems. They are
00:19not deterministic. They cannot learn on the job. They are prone to errors. They are prone to
00:24hallucinations. Ask people who use them and they will say they can't use them for things that are
00:28mission critical. That is not to say they aren't amazing. Of course they are. Look at how the job
00:33market is changing as a result of their existence already. And there is a feeling that it is possible
00:39that all these problems can be developed away. But what if they can't be? What if these problems
00:45are systemic if they are intrinsic to the model and they can't be fixed? It may be that the whole
00:52LLM thing is a bit of a false start and that there is a different type of architecture required for
00:57a different type of AI that is on the way. There's a lot going on in this space, by the
01:02way. But if it
01:02has been a false start, what does that tell us about the money being spent already? Is it possible
01:09that we are seeing one of the greatest capital misallocations of all time?
Comments