1. Progress in generative AI
Welcome back! In this video, we'll discuss progress in generative AI, its players and driving factors.
2. A collaborative effort
Progress relies on a variety of players, each bringing unique contributions.
Universities lead fundamental research and produce researchers.
They are often funded by governments.
Open-source communities build and iterate on datasets and models.
Startups and large companies hire researchers and integrate generative AI into products.
Let's take a deeper look at each.
3. Universities
Universities play several key roles in generative AI progress.
First, they produce research and theory that is a foundation for progress in the field.
For example, the GAN architecture was originally proposed by a group from the University of Montreal.
Universities also train researchers in generative AI and related fields. Students often go on to found startup companies or lead research in other institutions.
Finally, universities collaborate through partnerships with industry or government grants.
4. Governments and civic organizations
Governments set the regulatory environment for development, but may also provide funding.
For instance, the US government's DARPA (Defense Advanced Research Projects Agency) has funded research into making artificial intelligence more explainable.
Civic institutions contribute independent analyses, resources, and project funding. For instance, the Canadian Institute for Advanced Research (CIFAR) created CIFAR-10, a widely used image dataset.
5. Open-source communities
Open-source projects provide open access to the tools used to develop generative AI, as well as some generative AI models themselves.
This lowers barriers to deploying and experimenting with generative AI technology, speeding up innovation through collaboration.
Many foundational models, like the image generator Stable Diffusion and large language model LLaMA, are open-source projects used broadly across the space.
However, there are risks.
Given the decreased ability to monetize, open-source projects are often not maintained for long.
Also, giving open access to such powerful tools raises the risks of misuse by malicious actors.
6. Startups and large companies
Companies large and small develop generative AI to gain a market advantage.
They bring generative AI to the broader world through their products.
They also show off their latest technology to entice more support for their work.
Companies with large funding resources additionally publish research themselves. For instance, Google published the first research introducing transformer models.
Large companies also acquire startups and hire researchers to gain access to fresh talent and intellectual property.
They also produce much of the hardware and cloud resources required by other contributors to do their work.
7. The openness challenge
Companies are faced with a challenge when considering generative AI research publication.
On one hand, publishing openly will win support from developers, attract new talent, and provide more opportunities for feedback from a broader set of stakeholders.
On the other hand, releasing too much risks losing competitive advantages and possibly allowing malicious actors to use the work for destructive ends.
8. The boundaries of generative AI development
As we look to the future of generative AI development, several components will spur its progress while others will slow it down.
Accelerators include continued decline of hardware costs, further research development, and pressure from market and geopolitical competition.
Decelerators include hitting the physical limits of technology, regulation that inhibits progress, companies keeping their research behind closed doors, and a lack of funding, data, or talent to fuel the field.
The ideal is collaboration among players that balances openness, competition, and responsibility.
9. Let's practice!
Let's see what we've learned!