In the mythology of tech start-ups there is the garage. It all starts in a garage. The garage where, in 1947 at 367 Addison Avenue in Palo Alto, the two young engineers David Packard and Bill Hewlett had "startuped" HP - the mother of all innovative IT companies - is today one of the most legendary symbols of entrepreneurship. The same goes for Steve Jobs' garage where Apple was born. The origin of Google is also not very different.
The genesis of these companies destined to last has entered the imagination of entire generations of entrepreneurs, inspiring them and also showing them that something great is built on an idea.
Many have begun to see the return of the garage in the unexpected and surprising development of generative artificial intelligence.
The success of Open AI, a startup unknown until a few months ago and today even in the crosshairs of the antitrust, seems to represent a turning point in the path of high technology such as the landing of the personal computer in the second part of the seventies with the Apple 2 or the access to the Internet with the Netscape browser in the mid-nineties may have been.
analogy
The juxtaposition can make sense up to a certain point. In today's new wave of generative AI we find features common to those moments, but also completely new features.
A common trait is that the initiator of a revolution, as often happens in history, was not an incumbent, i.e. a large entity with a position of pre-eminence such as IBM was when it appearedapple 2 or Microsoft when Netscape Navigator started getting people on the internet. It was start-ups founded by visionary and courageous young people who triggered the change and fixed it in the habits of the mass of consumers.
However, it happened that, after a few years of disorientation, the incumbents have regained control of the situation: Big Blue with the IBM PC and Microsoft with Internet Explorer. However, it happened that the resumption of control lasted a San Martino summer because the challenge launched by the disrupters had set in motion a process of proliferation of investments and innovative businesses that escaped the control of any entity.
Perhaps the similarities between the aforementioned turning points in the history of technology and the current one of generative AI essentially end here.
New conditions and new strategies
The huge economic resources and computing power required by start-ups attempting to establish their solutions based on the trendiest technology of the moment could make the epos of the garage a siren song.
in the race ofGenerative AI the entry threshold is very high and the bill is salty. The creation of applications or platforms requires investments, infrastructures and organizational resources that surpass the possibilities of a start-up even in the case of a strong contribution of venture capital.
For generative AI start-ups, there are two paths: develop an innovative technology and sell it to the incumbent who does not have it, or, keep control of the company and team up with the latter using the resources, services and infrastructure that the incumbent has already built in the size it serves.
Therefore building a successful start-up is an almost impossible mission without the support of a company that has large, widespread and expensive data centers. These companies are few in number and are usually powerful multinationals feared even by governments.
Arm in arm with the enemy
It is therefore understandable why it is the option of the alliance rather than the competition that is most accredited in the strategy of startup growth. Thus a conjunction of objectives is verified between the start-up that wants to mess up the games and the company that already runs those games. Something unthinkable at the time of the garage: can you imagine Steve Jobs' Apple going to IBM to offer its technology in exchange for support in scaling his business? Unimaginable.
Today, however, this is exactly what is happening. Indeed it has already happened with thestrategic alliance between OpenAI, who developed the ChatGTP4 and DALL-E algorithms, and Microsoft which possesses the cloud infrastructure and computing power to bring these applications to the masses of consumers.
OpenAI has, in fact, received 10 billion dollars from Microsoft. Most of this sum will go back to Microsoft to pay for the time consumed by AI applications on the computers of its server farms. Thanks to clusters of thousands of specialized chips, Chat GPT and other similar technologies are able to adequately serve the ever-growing public who question them. And OpenAI is not the only one to have embarked on this path of so-called "unnatural" alliances.
Furthermore, OpenAI's choices also end up influencing possible competitors. In fact, the latter must be able to get their hands on a similar amount of processing.
Looking for computing power
Cade Metz, technology reporter of the New York Times”, in a recent article reports some cases of startups that are moving in the wake of OpenAI.
Among others there is Cohere which recently raised $700 million which it will use largely to access the server farms of Google. The three young founders of Cohere are former employees of the Mountain View company. The CEO of Google, Sundar Pichai, has personally approved this agreement which provides decisive support to a potential competitor.
Indeed Cohere is building an artificial neural network model, an LLM (Large Language Model), capable of competing with Bard, Google's proposal. In reality there will be no collision because Cohere's business model is B2B.
The idea of the founders of Cohere intact is to make their technology available to other companies, in the sense of providing them with a sort of AI engine to build and manage your own applications whether they are chatbots, search engines, personal tutors or specific contents. All these applications will be cloud and hence voracious of computing power.
The tech giants are therefore in a strong position because they have the in-house resources to push these systems further than anyone else. Google also owns the patent on Transformer, the technology behind the AI systems that Cohere and many other companies are building.
Transformer is an example of what is called a neural network, ie a mathematical system capable of learning from massive data. Neural networks have been around for a long time and are the basis of voice assistants like Siri or instant translation services like Google Translate.
The wild card of open source
In all this ambaradan there is a stone guest, theopen source. An option that Meta he embraced unreservedly.
Meta, another giant with the computing power to support the next wave of Ai, recently released the source of its large language model, LLaMA, into the public domain. A step that puts anyone in a position to reuse it and build their own version of AI on it.
An approach, that of Meta, different from competitors such as OpenAI or Google, whose applications, GPT-4 and Bard, are a "black box" in the sense that the data and code used to build the model are not available to third parties.
Many believe that open source will allow anyone to enter the generative AI competition with similar opportunities.
Not entirely, however, because even for startups that are moving in the direction of open source, the problem of computing power and server farms remains. Will Meta provide him with all this? Who knows if the real business isn't right here. Also, putting something open source doesn't alienate the intellectual property of that resource.
One could say that Meta is a bit trying to do what Google did with Android, i.e. transferring the use of the tool to third parties in exchange for data and information collected with the massive circulation of the tool itself.
We know that AI applications improve with the influx of data and if open source brings data to the Meta system so much the better.
Other AI companies, such as the French startup Mistral, are considering releasing open source versions of their technology. OpenAI, which has already made open-source AI models for speech and image recognition available, said its team is only considering developing an open-source LLM if it can reduce the risks of abuse. below an acceptable threshold.
Good Luck!
Sources:
Cade Metz, In the Age of AI, Tech's Little Guys Need Big Friend, The New York Times, July 3, 2023
Nick Clegg, Openness on AI is the way forward for tech, The Financial Times, 11 July 2023
Christine Criddle, Madhumita Murgia, Hannah Murphy e Leila Abboud, Meta to release commercial AI model in effort to catch rivals, The Financial Times, 1 July 2023
