Abzu

A supercomputing center with 15,000 Nvidia H100s! Impressive, and a major boost for Danish AI research. As a skeptic about the growing use of black box AI in scientific research, I am also concerned. In drug discovery, and in most other fields, understanding the science is just as important as predictions. We need to understand pathologies, pathways, and modes of interaction, not just throw data at a supercomputer and get predictions that may be correct but which nobody understands. Otherwise, we will end up like ignorants who visit an Oracle of Delphi, hoping to get answers without understanding. I am happy that this amazing machine is being built, but don't give up on good old science, please. Incidentally, there are types of AI that can help with real scientific discovery, too.

Hot Mistral over an Azure sea Late in February, the European AI darling, Mistral, announced a partnership to bring their "Made in Europe" generative AI to Microsoft's Azure cloud services. All good, no? Now, enterprises have a European alternative to the dominant OpenAI on this widely used cloud. But, European politics being what they are, the backlash was immediate and predictable. Lawmakers in the European Parliament accused Mistral of having cheated them, and the European Commission launched an antitrust investigation. What would have been an ordinary and beneficial business deal under normal circumstances turned into a political storm, likely damaging the startup company, the reputation of Europe as an industry-friendly jurisdiction, and the confidence and motivation of other European-based AI companies. Should @AbzuAI think twice if an equivalent opportunity arises for us?

One after another, we are modeling the ways RNA therapeutic medicines interact with the body. With our scientific AI platform we are constantly making progress. It will still be a long time before RNA therapeutics designed in silico just work in humans. But by solving biological puzzles one by one, we are moving ever closer a future when every disease with a genetic origin can be simply and efficiently cured. If you want to discuss collaborating on scientific problems that advance RNA therapies, please feel free to connect and reach out to me or @AbzuAI.

RNA Therapeutics. Together with the rest of @AbzuAI, I am throwing all my energy into using out AI to develop extremely targeted, safe, and effective new drugs against all the rare diseases that have no cure today. Expect me to be posting a lot about that in the future😀

A worksession on organisational keystone habits today @AbzuAI. Rarely have I felt such a strong energy and enthusiasm. I love working with these fantastic people❤️

A quick reference sheet to help determine high-risk use cases as defined by the EU AI Act

Gathering week at @AbzuAI - every quarter we get the entire company together for a week of planning, working together and quite a bit of partying. The three first Abzoids from Barcelona arrived in the Copenhagen office this morning.

I've been following the development of the EU AI Act with apprehension and concern, but in the end, it seems to have turned out well. My initial thoughts: I wholeheartedly support the risk-based approach adopted in the Act. It ensures that AI systems posing potential harm to health, safety, and fundamental rights are subject to necessary requirements. An important aspect of the Act is its focus on regulating the use of AI, rather than research. By regulating use, we ensure that AI benefits society without stifling the creativity and exploration that drive our field forward. During the discussions, the idea of putting AI research under government control was a major debate. I'm relieved to see that this did not end up being the case. Indeed, the AI Act appears to promote free and open source and unhindered sharing of research. Lastly, the Act recognizes the importance of responsible use of AI in high-risk areas. My view has long been that symbolic AI is the best way to get the benefits of AI while ensuring safety, trustworthyness, and human oversight. The EU AI Act supports this stance with its emphasis on transparent and explainable AI. All in all, I am happy with the outcome and, honestly, quite relieved.

First thoughts on the firing of Altman as CEO of OpenAI: It appears the cause was Altman's tampering with the OpenAI statutes, which are to promote a free and open AI for the benefit of humanity, favoring collaboration with Microsoft instead. Elon Musk has been raising his voice about this recently. I wonder if his hand is behind this? I would also note that the board of OpenAI consists of people who have no ownership in OpenAI, and hence no strong commercial interest in violating the statutes. Another point worth noting: Greg Brockman was demoted as well, although not fired. Interesting times ahead.