"François Chollet posits that intelligence is not about how well a specialized model performs, but how good it is in learning new things, or in other words,"

"How well it can generalize across tasks."

"In this sense we are probably still far away from general intelligence and this is one of the reasons why I personally don’t like the term artificial intelligence."

"We don’t really understand and agree on what intelligence is."

"In large language models such as ChatGPT, the key to their success was the scale, being able to create very large models that can be trained on huge amounts of data."

"We don’t yet have anything of comparable scale in biology, and probably the main limitation is the amount of data we have."

"It is very expensive to obtain experimental data."

"It will be possible to overcome the lack of experimental data with simulation and it’s an interesting question how to combine simulated data."

"The range and diversity of problems in biology is significantly bigger than in language."

"It could be that in some applications we don’t necessarily need the kind of scale we find in ChatGPT."

"Biotech and pharma companies try to do it and successfully in many cases."

"One example is Recursion that scaled-up cell-painting technologies, allowing to image hundreds of millions of cells and see what happens to the cells. *When you perturb them either chemically or genetically."

"What I would like to do is to take a step back and look at the next generation of data sources where the consumer of the data will not be a human but a machine."

"If we say that the data does not necessarily need to be viewed by human scientists, we can come up with completely new experimental data sources."

"I am honoured and delighted to join Oxford, which has unparalleled expertise in AI and related fields and amazing students."

"I am looking forward to forging new collaborations and synergies within the Department and beyond that would allow us to develop the next generation."

"Learning methods that solve real-world problems and at the same time have the trust of domain experts and the broader public."