AI

AI

New crisis in OpenAI and this one is really on a key point

The two leaders of the department in charge of the cybersecurity policy for OpenAI have resigned, as a consequence the team has been suppressed. From what filters, the resignations of seasoned managers is due to big divergences on the challenges the company has to meet with regard to security.

How to make sure the models are retained in a way that avoids compromission and bias ? The matter is even more touchy as Gen AI moves from text to speech and video, paving way to a whole world of fakes and uncertainty. So far, it seems that the will to be “first to the pole” predominates, whatever the consequences may be. And what looms is not neglectable.

On April 23, 2024, our partner G9+ organized another forum on AI

Among other speakers was Hopsworks, a Swedish company with whom the ECA cooperates. This time, the focus was on the quality of data to be used to train the model. An approach which somewhat differs from the too simple idea “the more data you have in the data set used to train the model, the more efficient this one is”. As Catharina Dember, member of G9+ and of the ECA, points it : this stimulates the reflexion on how to adapt AI to each usage, and especially to avoid energy catastrophe. 

EVIDEN to provide its new server for super-computers BullSequana AI 1200H

This system will be used among others by the new Jean Zay high performance center of the French CNRS, but also by the Danish Center for AI Innovation. It is compatible with the Nvidia servers today commonly used for such high-end computing.

CRÉDIT AGRICOLE to push AI

Crédit Agricole, a French large bank, wants to speed up research and application developments resorting to quantum technologies. Experimentations are conducted with Pasqal et Multiverse Computing regarding financial products valorisation, or credit risks assessment. Other domains are better computation models, machine learning and fraud management.