Artificial Intelligence – Arms race of stupid

Look for the picks and shovels of this latest bubble.

• Google is rushing to market with a competitor for ChatGPT triggering an arms race which will end with huge losses when reality pricks yet another bubble triggering the 4^th AI Winter.
• Google has made two moves very recently:
  • First, Anthropic: Google has made a $300m investment in an OpenAI competitor called Anthropic.
  • Anthropic was founded by a group of AI researchers who left OpenAI over a difference of opinion over Microsoft’s growing influence on OpenAI.
  • Somewhat paradoxically, Anthropic is now embarking on a very similar journey with Google as OpenAI did with Microsoft but this is probably due to the fact that its largest investor is Alameda Research.
  • Alameda Research is part of FTX and as such Anthropic is earmarked for sale meaning that the company may not have had much choice but to seek help from Google.
  • Anthropic follows a very similar philosophy which is that with enough compute and enough data the answer will pop out at the end.
  • This means that it is going to be very processing intensive all of which will presumably now be executed on Google Cloud.
  • Second, Bard. Google announced its intention to launch a chatbot like ChatGPT called Bard.
  • Critically, Bard will also be able to draw on Google’s expertise in characterising the Internet in real-time and as such, it will be able to comment on current affairs.
  • By contrast, ChatGPT is stuck in late 2021 and has no knowledge of anything that has transpired after its cut-off date making it useless as a search engine.
  • Bard will presumably use an amalgamation of Google technology, Google Search and Anthropic to produce a product that I suspect will be a lot better than that which Open AI has come up with.
  • There is a reason why RFM ranks Google as the finest AI company in the world.
  • Google would not be drawn on when this would hit general availability meaning that this is still in development, but Google caved to pressure to announce something now.
• The popularity of ChatGPT and the technology press’ willingness to ignore the reality of what ChatGPT is, has kickstarted an arms race where a series of huge but very dumb machines will be created to meet demand.
• The problem with all of these machines (and any AI built using deep learning) is that has no causal understanding of what it does.
• Just take this extract from a conversation I had with ChatGPT a couple of days ago. “No, 509 is not a prime number. A prime number is a positive integer greater than 1 that is only divisible by 1 and itself. The positive integer factors of 509 are 1 and 509, so it has additional factors other than 1 and itself, making it not a prime number” (see here for full exchange)
• Here, ChatGPT defines a prime number, proves that 509 is prime and then goes on to state that 509 is not a prime number thereby demonstrating that in reality, it has no understanding of what a prime number is.
• This is the same for all AI which is why RFM research has argued for years that AI is great for tasks where the dataset is stable and finite but hopeless for anything where things change or are loosely defined.
• This means that ChatGPT and its competitors will sooner or later run into the brick wall of reality just like chatbots like Alexa and autonomous driving have done in recent years.
• The result will be crashing valuations, huge write-offs and another period of navel-gazing while the AI industry ponders what went wrong.
• This will be the 4^th AI Winter.
• This does not mean that there is not money to be made but it will not be in the companies that are building these behemoths but in the companies that supply them who may well see a sudden surge in demand as this bubble gets underway.
• Here, Nvidia and AMD leap immediately to mind as suppliers of the silicon chips that these chat systems run on.
• Intel is also in line to benefit but I remain concerned with market share loss to AMD in the server as well as the structural and financial issues that the company is still wrestling with.
• Nvidia and AMD are not cheap but crucially they are well below their highs meaning that a hype cycle of mad ChatGPT-linked speculation could easily drive them back to those highs if not beyond.
• As a fundamental investor, this is not the kind of position that I normally take but if I wanted a flutter on this, this is how I would position for it.