Artificial Intelligence – Race to the bottom

Price and feature erosion are bad signs.  

• One would have expected Anthropic’s latest innovation to be part of its premium tier, but the fact that it is free from the start is a sign that the race to the bottom kicked off by Meta is already beginning, which in turn may herald the bursting of the AI bubble.
• Anthropic is one of the now many AI companies that have a foundation model and are attempting to make a business from selling generative AI services built on top of it.
• Its latest innovation is called Artifacts which is a real-time rendering of the content that Claude is generating.
• This is directly analogous to the emulator in a software development kit that runs the software being created exactly how it will run on the targeted software platform.
• This is a big-time saver because without the emulator, developers would have to compile the code they write and run it on the target device to see what their app will look like resulting in endless back and forth as they tweak and debug.
• The emulator is standard for software developers and seeing that one of the big use cases for generative AI is generating code for apps, websites, documents and so on, it makes complete sense to create something similar for LLMs.
• This is a feature that I have not seen any other LLM competitor add to its offering yet and so this should be a differentiating move that Anthropic should be using to drive users to its premium offering.
• However, this feature is being made available to all tiers of its service which I take as a sign that Anthropic is struggling to bring users onto its platform and that competition is already tough.
• There have been other signs of this such as Meta’s move to make its biggest and best model available to anyone for free and the fact that one no longer has to have an account to use ChatGPT.
• There is no sign of the super intelligence that we were promised with generative AI powered by LLMs, and as a result, I remain concerned that the expectations that have been set will not be met.
• LLMs have some large, and I think lucrative, use cases but they are not about to take over from humans in a large number of tasks and this is where the disconnect lies.
• This is precisely what happened with the Internet in 1999-2000 and I suspect that a similar level of reset is likely to happen here.
• The Internet went on to change the world as we know it but following the bubble, there were several years of depression and despondency while the Internet developed and evolved to be able to live up to its full potential.
• Some are predicting a reset by the end of the year, but the truth is that there is no real way of knowing when or what the catalyst will be that triggers the reset.
• The only thing I am sure of is that current expectations and valuations are unrealistic and that the reset is likely to begin in the venture capital space as start-ups fail to meet their targets and go to their backers for more money.
• Against this backdrop, everyone is going to take a hit but the least pain is likely to be felt by Nvidia and TSMC as they are already making good money from this sector and, compared to other AI companies, they are attractively valued.
• That being said I continue to prefer the adjacencies of inference at the edge and nuclear power as the best way to get exposure to AI.