Google Gemini – More of same.

It will soon be much cheaper.

• Google has launched a new version of its generative AI and followed everyone else down the $20 per user per month business model, but as it appears to be pretty similar to the other models available, I expect that price erosion will set in pretty quickly.
• Google has made several changes involving its AI services.
  • First, Bard becomes Gemini: Google has rebranded its freely available chatbot to Gemini to match the model that started powering the chatbot a few months ago.
  • Gemini is the successor to LaMDA which was announced in December 2023 and is available in three versions.
  • These are Gemini Pro (powered Bard), Gemini Nano (for smartphones) and Gemini Ultra that has just been made available as a paid subscription.
  • The free service is unchanged since December as this is merely a change in name from Bard to Gemini and it still refuses to draw images.
  • Second, Google ecosystem services, which are unchanged on the surface but will now be powered by Gemini Pro replacing the older models that they were using.
  • For example, Duet AI will now become Gemini for Workspace and Google One subscribers will be able to use Gemini in Gmail, Docs, Slides, Sheets and Meet.
  • This is Google’s answer to Microsoft’s Co-Pilot for Office 365 and looks to be coming at a slightly lower price given all of the other things that come with the subscription.
  • This is precisely the route that I have been expecting Google to take where it uses AI to make its existing services better rather than launching a large range of new services.
  • Samsung and Google have also integrated Gemini Nano into the Samsung Galaxy s24 which again is about improving Google ecosystem services making the hardware more attractive to buyers.
  • Third, Gemini Advanced – This is a new service that competes directly with Open AI’s premium service that gives access to GPT-4.
  • Gemini Advanced gives access to Gemini Ultra 1.0 which Google claims can outperform human experts in 57 subjects which is tested on multiple choice, standardised testing.
  • This is not a great test of cognitive ability as each field has a syllabus that the machine memorises and then takes a multiple-choice test.
  • Where the human experts destroy the machines is when they are tasked with using their expertise in situations or circumstances for which they have not been explicitly trained.
  • This comes out in the early testing where despite being an expert in knowledge, Gemini Ultra 1.0 fails to answer questions that my 8-year-old can do easily.
• The net result is that Gemini finally puts Google in the AI race with a simple range of models that serve to both improve existing products and power stand-alone products to compete with everyone else.
• Immediate testing indicates that Gemini Ultra 1.0 is roughly equivalent to GPT-4, Claude, Mistral and so on meaning that there is no longer very much that separates the contenders in this field.
• Furthermore, the ever-growing number of contenders in conjunction with Meta’s enablement of the open source community sets the stage for a huge growth in the supply of available services that are all fairly similar.
• This is what I think will trigger a series of price cuts that will end up with the price for most of these services going to something closer to $2 per month as opposed to $20 per month.
• This will mean that all of the assumptions upon which billions of dollars have been raised will be out by around a factor of 10x.
• The result will be start-ups coming back to VCs for more money before they previously said that they would resulting in down rounds and a fall in valuations and enthusiasm.
• This is the pin that I suspect will prick the AI bubble and there is plenty of scope for this to happen this year or early 2025.
• For those who are not in the race, this will imply that there will be perfectly good models available to buy when the bubble bursts without having to pay the crazy valuations that the likes of Anthropic and Open AI are currently demanding.
• Google will survive this unscathed as I suspect that its services will be improved as a result of Gemini leading to greater loyalty and an improved ability to monetise its ecosystem.
• Of all the big tech names and AI in particular, Google’s shares offer the best value and is the one I would buy if I were going to go down this route.
• The other AI name to consider is Baidu which is by far the cheapest AI stock in the world, but this requires one to take a view on the geopolitically fraught situation between China and the USA.
• Investing in China at the moment is not for the faint of heart although there is great value to be had.