Whether in terms of technological convergence, disruption in HMI, process hybridization or data governance, generative AI is already having an impact on company strategies. But the explosion of supply, during this period, is very favorable to making bad decisions which will have been taken under the spell of the formidable marketing of an actor. Then finally, after the first encouraging POCs, there will be nothing very industrial to deploy more widely. Loss of money, but also of precious years.
Because time is the main variable in disruptive generative AI technologies. Today, the strategies of the “pickaxe sellers”, in reference to the California gold rush which first made them rich, are no longer counted in months, but in weeks. Since the beginning of the year, new AI models, new connectors between these models and information assets, or to well-installed tools, have been published every week. Every month, one or more major players in Tech make an announcement of their strategy.
Three months after their introduction last May, there are 920 plugins in ChatGPT, 34 of which are already popular and used by hundreds of thousands of users, or 60 every week! Monitoring teams are on their knees or already overwhelmed by this proliferation, and GreenSI knows something about it!
To those who tell you that they are familiar with the generative AI landscape, ask them when that dates back! But above all, ask them how many people they have full time to really test what comes out each week and not stay on the effects of announcements. So supply is moving much faster than demand. This market is driven by marketing, hence caution before committing massively to a single path, even that of a big name. 😉
The landscape is evolving very quickly since OpenAI on November 30 unleashed this storm of innovation in the Tech ecosystem.
Since February, 2 months after the launch by OpenAI, everyone has noticed the general enthusiasm and the speed of adoption of this new interface in natural language, at least for testing, and everyone is afraid to see this train leave. Those who already had something in the pipeline had to speed up the release of their products, and refine their strategy of occupying a market occupied by OpenAI. This is the case of Google and of Meta, but also smaller players who thought they were in a niche (like AI21 or Anthropic) and woke up in the middle of an ocean full of sharks. 😉
Google Bard was launched to compete with other generative language models, with the key to maintaining its position in the field of AI (it was Google search that produced the key for generative AIs). But it is also that of its dominant position in search engines. Google Bard soon expanded to Europe and Brazil as early as July 2023, suggesting a global language expansion strategy. The strong point is its permanent connectivity to the Web, its content and its databases, when ChatGPT must proceed, connector by connector, with the reliable and acclaimed sources of the Internet.
At the last Google I/O conference, this AI was integrated into other Google products, in photo management and especially in its online collaborative suite.
The strategy of Meta is more difficult to decode. It is less directly linked to its social products.
It is an opening combination (LLaMA2 accessible in open source mode), flexibility and diversification (LLaMA2 is based on smaller models, therefore manipulable by companies with fewer resources). Meta also emphasizes transparency and control. These last two points will be interesting for social networks, and it is perhaps a response to X (ex-Twitter) which evolves its moderation towards a network of qualified users who produce references, and from automatic moderation to AI base (see Progress update on the new Twitter). It will be interesting to see how this strategy evolves as the model is adopted and integrated into Meta’s offering, or even diversification offerings beyond social networks.
Here is the first category of those who were ready. Now those who were further from a product, like Amazon, Apple Or Microsoftthey had to open their purse in order to finance programs or acquisitions.
The most active, the earliest, having clearly been Microsoft, which had entered into an alliance with OpenAI to embed ChatGPT in its products. We have all read that Bing with ChatGPT was going to carve out Google’s market share, and indeed Bing was very downloaded and much more used, as soon as it was released. BingChat in February. But the bellows deflated and six months later, it is clear in the global rankings (like StatCounter) that Bing has not progressed and still holds 3% of the global search market share. Mostly obtained from Edge or Windows.
But Microsoft has a second card to play: getting its business customers to do generative AI – which requires a lot of resources – on its Azure Cloud. Failing to make money with a generative AI service, there is a place to be taken to power global AI, sell resources and try to shake up its big rival number 1 in the Cloud, Amazon. The tension on GPU cards, particularly suitable for learning models, is also rising, in addition to prices, with rumors of shortages.
Amazon therefore announced its strategy at the end of the first quarter. It is thus launching a new platform, with the same objective as Microsoft and Azure, but unlike Microsoft, is also trying to catch up on the models and wants to produce two new artificial intelligences (called Titan). They will also be available on its Amazon platform Bedrock which allows access to multiple artificial intelligences thanks to an API, but also to the next in-house versions.
It is therefore an approach geared towards developers, to create applications embedding conversational LLMs from partners (Jurassic-2 fromAI21 studio and Claude d’Anthropic), and provide them with code generation tools like “CodeWhisperer” (already announced last year) which wants to reduce Microsoft’s dominance in this area. A dominance consolidated by the acquisition in 2018 of Github, which already has a certain lead with its CoPilot tools, but above all total visibility on the codes global sources and therefore a unique learning capacity.
Anthropic, it’s one of those young shoots that was thrown into the deep end from one week to the next. All those who no longer want GAFA hegemony over what is becoming a disruption in the uses of AI technology will support these companies. Other actors deserve to be followed, and in particular Hugging Face founded by French (but American), known for its open source library of Transformers, or LightOna French company which provides turnkey models for businesses.
Antropic was founded in 2021 by Dario Amodei (ex OpenAI), has raised more than 1.5 billion dollars and announces that it is preparing funding rounds for an additional 5 billion. Its objective is to develop an algorithm for self-learning AI, in order to be able to work with dynamic models. In short, a better and more connected ChatGPT which cites its sources. Time will tell if it is superior in terms of services. What catches the attention of GreenSI it is the level of investment of the entry ticket into the world of models (several billions) which in terms of investment is beyond the reach of the majority of companies. Companies will therefore have to turn to “pickaxe suppliers” to adapt general AI with their data. This vision is consistent with the strategy of Microsoft and Amazon, but also Google, who want to bring you to their Cloud for your developments. The first market for generative AI will certainly be that of the Cloud and its infrastructures.
But a second questioning of GreenSI, for both ChatGPT and Claude, it is the ability to finance this service in the future, beyond the first investment, and its evolution with subscriptions at €20 per month. And given the abundance of supply, it is not yet possible to raise prices. Everyone is clearly in deficit right now. But a double Moore’s law, on the increase in power of GPU processors and on the reduction in the need for resources with more optimized models, could soon bring them into a zone of equilibrium of the model. This is what will be essential to monitor, because just as advertising finances many internet services which can collapse if advertising disappears, generative AI will have to find its model… or disappear.
If we did not pay for the platform, which would be subsidized, would the future of generative AI platforms be only in use cases?
This is what we invite you to explore shortly in part 2 of this post on the uses of generative AI. But on this hot topic, don’t hesitate to give your vision in the comments.