From the paper:
'AlphaEvolve was able to find a simple code rewrite that removed unnecessary bits, a change validated by TPU designers for correctness. ... Integrated into an upcoming TPU, this improvement represents Gemini's first direct contribution to TPU arithmetic circuits'
'Firstly, the FlashAttention kernel for the configuration of interest was sped up by 32%. Secondly, AlphaEvolve found improvements in pre- and postprocessing of kernel inputs and outputs, resulting in a 15% speed up'
'We applied the system to over 50 open problems in mathematical analysis, geometry, combinatorics and number theory. In roughly 75% of cases, it rediscovered state-of-the-art solutions ... And in 20% of cases, AlphaEvolve improved the previously best known solutions'
OpenAI submitted their policy proposal to the US government this morning. They directly link fair use with national security, and said if China continues to have free access to data while 'American companies are left without fair use access, the race for Al is effectively over.'
'The federal government can both secure Americans' freedom to learn from Al, and avoid forfeiting our Al lead to the PRC by preserving American Al models' ability to learn from copyrighted material'
They also argue for banning the use of PRC-produced models within Tier 1 countries that 'violate user privacy and create security risks such as the risk of IP theft.'
'designed to mirror the reasoning process underpinning the scientific method ... the AI co-scientist system is intended to uncover new, original knowledge and to formulate demonstrably novel research hypotheses and proposals'
It uses a team of six distinct agents 'inspired by the scientific method itself' using 'automated feedback to iteratively generate, evaluate, and refine hypotheses'
This morning the White House released the Interim Final Rule on Artificial Intelligence Diffusion.
If you live in one of the tier one nations, there are no restrictions on the amount of chips you can buy, or the size of the datacenter you can build.
This is new, a carveout for 'clearly innocuous purposes'. Any order less than 1700 GPU's will not require a license, and does not count against the national chip caps. This is probably a result of the backlash leading up to this.
NVIDIA has built a 2.5 billion parameter audio model called Fugatto that generates music, voice, and sound from text and audio input. Sound inputs become completely mutable. It can change a piano line to a human voice singing or make 'a trumpet bark or a saxophone meow.
Using a feature called temporal interpolation, Fugatto can 'create the sounds of a rainstorm moving through an area with crescendos of thunder that slowly fade into the distance. It also gives users fine-grained control over how the soundscape evolves'