Toggle light / dark theme

The team turned AlphaEvolve loose on Google’s Borg cluster management system for its data centers. The AI suggested a change to the scheduling heuristics, which has been implemented to save Google 0.7 percent on its computing resources globally. For a company the size of Google, that’s a significant financial benefit.

AlphaEvolve may also be able to make generative AI more efficient, which is necessary if anyone is ever going to make money on the technology. The internal workings of generative systems are based on matrix multiplication operations. The most efficient way to multiply 4×4 complex-valued matrices was devised by mathematician Volker Strassen in 1969, and that held for decades, but DeepMind says AlphaEvolve has discovered a new algorithm that’s even more efficient. DeepMind has worked on this problem before with narrowly trained AI agents like AlphaTensor. Despite being a general AI, AlphaEvolve came up with a better solution than AlphaTensor.

Google’s next-generation Tensor processing hardware will also benefit from AlphaEvolve. DeepMind reports that the AI created a change to the chip’s Verilog hardware description language that dropped unnecessary bits to increase efficiency. Google is still working to verify the change but expects this to be part of the upcoming processor.

As searches for the leading dark matter candidates—weakly interacting massive particles, axions, and primordial black holes—continue to deliver null results, the door opens on the exploration of more exotic alternatives. Guanming Liang and Robert Caldwell of Dartmouth College in New Hampshire have now proposed a dark matter candidate that is analogous with a superconducting state [1]. Their proposal involves interacting fermions that could exist in a condensate similar to that formed by Cooper pairs in the Bardeen-Cooper-Schrieffer theory of superconductivity.

The novel fermions considered by Liang and Caldwell emerge in the Nambu–Jona-Lasinio model, which can be regarded as a low-energy approximation of the quantum chromodynamics theory that describes the strong interaction. The duo considers a scenario where, in the early Universe, the fermions behave like radiation, reaching thermal equilibrium with standard photons. As the Universe expands and the temperature drops below a certain threshold, however, the fermions undergo a phase transition that leads them to pair up and form a massive condensate.

The proposed scenario has several appealing features, say Liang and Caldwell. The fermions’ behavior would be consistent with that of the cold dark matter considered by the current standard model of cosmology. Further, the scenario implies a slight imbalance between fermions with different chiralities (left-and right-handed). Such an imbalance might be related to the yet-to-be-explained matter–antimatter asymmetry seen in the Universe. What’s more, the model predicts that the fermions obey a time-dependent equation of state that would produce unique, potentially observable signatures in the cosmic microwave background (CMB) radiation. The researchers suggest that next-generation CMB measurements—by the Simons Observatory and by so-called stage 4 CMB telescopes—might reach sufficient precision to vet their idea.

Google DeepMind’s AlphaEvolve AI system breaks a 56-year-old mathematical record by discovering a more efficient matrix multiplication algorithm that had eluded human mathematicians since Strassen’s 1969 breakthrough.

HUGE AI breakthrough: Absolute Zero Reasoner deep dive. Self-improving AI that learns with no data! #ai #aitools #ainews #llm.

Sources:
https://arxiv.org/abs/2505.03335
https://github.com/LeapLabTHU/Absolut… Thanks to Tavus for sponsoring this video. Try Tavus for free https://tavus.plug.dev/T4AQw5K 0:00 Absolute Zero intro 0:50 Traditional methods of training AI models 4:00 Absolute Zero algorithm 5:01 How Absolute Zero Reasoner works 7:19 Types of training tasks 9:00 How good is Absolute Zero 10:47 Tavus 12:11 Adding Absolute Zero to existing models 13:01 Interesting findings 15:43 Uhoh… 16:50 Ablation study 18:15 More interesting findings Newsletter: https://aisearch.substack.com/ Find AI tools & jobs: https://ai-search.io/ Support: https://ko-fi.com/aisearch Here’s my equipment, in case you’re wondering: Dell Precision 5690: https://www.dell.com/en-us/dt/ai-tech… Nvidia RTX 5,000 Ada https://nvda.ws/3zfqGqS Mouse/Keyboard: ALOGIC Echelon https://bit.ly/alogic-echelon Mic: Shure SM7B https://amzn.to/3DErjt1 Audio interface: Scarlett Solo https://amzn.to/3qELMeu.

Thanks to Tavus for sponsoring this video. Try Tavus for free https://tavus.plug.dev/T4AQw5K

0:00 Absolute Zero intro.
0:50 Traditional methods of training AI models.
4:00 Absolute Zero algorithm.
5:01 How Absolute Zero Reasoner works.
7:19 Types of training tasks.
9:00 How good is Absolute Zero.
10:47 Tavus.
12:11 Adding Absolute Zero to existing models.
13:01 Interesting findings.
15:43 Uhoh…
16:50 Ablation study.
18:15 More interesting findings.

Newsletter: https://aisearch.substack.com/
Find AI tools & jobs: https://ai-search.io/
Support: https://ko-fi.com/aisearch.

Here’s my equipment, in case you’re wondering: