Meta invites to AI Day: Program for start-ups and opportunities for Germany
AI has moved from the back seat to the front seat at Meta. The company hosted a roundup on AI in Berlin.
Naila Murray at Metas Day of AI.
(Image: emw / heise online)
Meta already has AI in its DNA - that's what Naila Murray, Head of Research at FAIR (Fundamental AI Research), said at a roundup on the topic of AI, which Meta hosted in Berlin. The program includes panels, keynote speeches and workshops in Red Teaming. This is the name given to the testing of AI models for sources of error and risks.
AI has moved a bit from the back seat to the front seat, says Murray. So instead of playing in the background to offer Meta's services as they are, people can now use AI directly themselves. Murray explains, for example, how Meta has worked step by step to develop automated recognition and cropping for images. You know, you can turn image sections into stickers in Meta's applications, and the outlines of an object are automatically recognized. That's what AI does. However, it often has to be said that it is not always generative AI that has triggered the hype since ChatGPT. Such applications have been in the works for a long time. Meta has been investing and researching in the field of AI for years.
Support program for start-ups and a new infrastructure
The keynote will also be about how you can take part in a PhD program at Meta. There is also a brand-new AI accelerator program for start-ups. Meta is working together with Hugging Face and Scaleway on this. According to Meta, the core of the program is access to the latest open models and resources in the field of AI as well as mentoring and access to the startup campus in Paris, Station F. Applications can be submitted until mid-August.
Meta has decided to offer its own AI models as open source. This was also an issue. Open source has advantages for the security of AI systems, is the reason given time and again. Competition and progress would also benefit from open source. Basically, Meta is a company that stands for openness. And AI is an area of research in which there has always been a close exchange between researchers, says Murray. An announcement that remains relatively thin on the ground. The scientist talks about the need for a new infrastructure to be able to use AI on a large scale. More information on this will be available soon.
German AI research
The participants on the podium changed; Meta invited Jonas Geiping from the Ellis Institute in Tübingen, for example, which was founded just last week. Prof. Georg Rehm from DFKI is there. He says, for example, that the German Research Center for AI was founded back in 1988 and that it is often hard to imagine, given the current hype, that this is such an old AI institution. Murray emphasizes that Germany is a location where a lot of important basic research in the field of AI has been carried out for a long time. If you look at the panel, you can actually hope that Germany and Europe have not been left behind, as is often feared.
Read also
AI models learn by heart and do not reason
AI: French antitrust authority opens investigation against Nvidia
OpenAI: Whistleblowers seek contract audit by stock authority
Photo News: Sony's vlogging update and Nikon's test balloon
Amazon launches AI shopping assistant Rufus in the US just in time for Prime Day
However, Rehm also points out that the funding of projects in Germany is always very specific and rather short. In the USA, on the other hand, there are projects that are set up over ten or more years and receive a lot of money despite a high risk, i.e. an idea that is still really abstract. Specifically, he cites Apple's Siri as an example, which basically emerged from such a project. Funded by Darpa, the US Defense Advanced Research Projects Agency. Metas Public Policy Director for Europe, Dr. Julia Reuss, also talks to Richard Würz from the venture capital firm Redstone about Germany as a location. He assures Germany of great potential.
When asked about artificial general intelligence, everyone agrees that a lot of research is still needed. However, it is not even clear how AGI can be achieved. Rehm explains that he believes that large language models are not the right way to go. And Jonas Geiping from the Ellis Institute also says that large language models are language models, which means that they can handle language.
Multi-agent systems are currently the focus of development at Meta when it comes to AGI. So the focus is not really on general intelligence, but on agents that initially learn from each other. What does not appear, however, are social networks, which are still Meta's main source of income. Of course, they contain AI and a lot of other research that has simply not been as visible as AI is now in focus.
(emw)