Artificial Intelligence – Scepticism’s Turn

The doomers are wrong, but not as much as the cool-aid crowd.

• The last week has seen the AI sceptics jumping up and down about a coming crash, but I think that the data they are citing is misleading, making them wrong, but not nearly as wrong as those who think that super intelligent machines are nearly upon us.
• Two pieces have caused angst in the last week or so,
  • First, the US Census Bureau: which found that AI adoption has been declining among firms that have 250 employees or more.
  • Firms with 100-249 employees have also seen a dip, while smaller firms continue to see adoption but at lower levels of penetration.
  • It is important to note that the level of penetration being observed here is low, with large firms dipping from 14% to 12% over the summer, and this could easily just be a blip.
  • I think that the reason for this decline is contained in the MIT study (see below).
  • Second, the MIT study: which triggered a sell-off of the AI stocks as a result of a finding that 95% of generative AI pilots were failing to deliver any tangible benefits.
  • This was immediately jumped on by the media as a sign that generative AI is nothing but hot air, but as usual, the reality was quite different.
  • The MIT researchers found that most of these pilots were failing because they had not been implemented properly, and in the few instances where implementation had been carried out properly, tangible benefits resulted.
  • The general view appears to be that I can give my employee a ChatGPT prompt and he or she will instantly become far more productive, but the reality is quite different.
  • This aligns with RFM Research, which has concluded that the best productivity gains will be seen with a model which has been fine-tuned on the company’s data as opposed to a generic chatbot prompt.
  • It is this realisation that I suspect has caused companies to cut back on using generic chatbots, which caused the dip observed by the US Census.
• The net result is that I think that there is tremendous value to be earned from using generative AI in a company, but only if it is implemented properly.
• This means fine-tuning the model on the company’s proprietary information so that it can surface pieces of information that almost everyone in the company will not know exists.
• I have long thought that all companies do not make anything like full use of the data that they have, and using a generative AI to ingest, cross-reference and surface this data will deliver enormous value to the enterprise.
• This most likely means having a 3^rd party expert supplier come in, implement the model and then periodically update and refine the model to ensure that it is working optimally and safely.
• This is pretty much what the MIT study concluded, and so I do not see this as a sign of doom but more a reflection of the fact that just giving an employee a generative AI prompt is not going to deliver very much.
• However, at the same time, the idea that OpenAI will have revenues of $200bn by 2030 remains fairly preposterous as there is no evidence whatsoever that super-intelligent machines are about to be created.
• Hence, there is still a correction coming, but it is not going to be a crash on the scale that the sceptics are predicting, as generative AI can deliver enormous value when correctly implemented and used.
• In investment terms, this means that the hyper valued companies that are burning cash like there is no tomorrow are likely to get into real trouble, but the picks and shovels like Nvidia, Qualcomm, Nebius, Samsung, Micron, SK Hynix and so on are likely to continue to do well.
• Hence, this is where I would put my investments into AI, and in that regard, I continue to hold Samsung and Qualcomm.