On the Trump 2.0 disaster scene, a reflection on the dangers of the Genesis Mission, which directed by the US Department of Energy to create AI scientists–and give AI companies access to federal datasets. Trump’s attempt to have his own (obviously Big & Beautiful) Manhattan Project?! Unsurprisingly, sounds very messy at this stage.
ICLR 2026 being exposed having 21% of the reviews are fully AI generated using tools from Pangram Labs, in New York. (And around 10% of the submissions being mostly written by AIs as well.) The inevitable evolution of the machine learning conference reviewing process?! Plus a review of the impact of Alphafold for its fifth anniversary. And a discussed paper by Oh & al. on meta-learning, conceiving reinforced learning algorithms that learn how to create learning algorithms. Without engaging with the paper per se I wonder at the degree of incrementality of the proposal. And hence find the catastrophic warning reproduced below rather over-the-top.
“In conclusion, it seems probable that AI will have an increasing role in the design of AI algorithms, a trend for which Oh et al.’s work is a harbinger. It is both exciting and worrying; the potential for intellectual discovery is vast, but the possible acceleration of a technology that already has an outsized societal impact is concerning in a world that is almost certainly not ready for the field’s dizziest possibilities to be realized ahead of schedule.” Joel Lehman
Also a book review of Belluz & Hall’s Food Intelligence, on the harms of processed food and the many myths that come with obesity and diets. Along a comment on diminishing investments on agricultural science being partly to blame for high food prices. And a concentration of the countries involved in R&D. But not helping (me) with the paradox that developed countries struggle to keep farming profitable enough for the farmers. As shown by the perpetual unrest of heavily subsidised farmers in the EU.
Xi'an's Og 