Trending AI Posts & Hot Research Catchup - Page 6
Huawei’s Tau Scaling Strategy: Why Chip Performance May Depend on Speed, Not Smaller Transistors
16 sources0Huawei is pursuing performance through faster signal and data movement, tighter chip integration and interconnected accelerator clusters—not by relying only on smaller transistors.
How is Huawei Technologies attempting to overcome the physical and geopolitical limits facing conventional processor development—where trans
14 sources0Huawei’s proposed alternative is to treat performance as a system-time problem rather than solely a transistor-size problem. Tau Scaling aims to reduce the delay with which signals, data and work move throug...
Why BESIII Says X(2370) Is the Best Glueball Candidate Yet
22 sources0At ICHEP on August 5, 2026, BESIII presented its strongest case yet that X(2370) is dominated by the lightest pseudoscalar glueball.
Huawei Mate 90: Kirin 9030 στο βασικό μοντέλο και LogicFolding στα Pro — όλες οι φήμες
32 sources0Οι διαρροές τοποθετούν τη σειρά Huawei Mate 90 για τον Σεπτέμβριο του 2026, με το βασικό μοντέλο να φέρεται να χρησιμοποιεί οκταπύρηνο Kirin 9030 και τα Pro να κρατούν το νέο Kirin 2026/9050 με LogicFolding·...
Kog's Software-First Strategy: How a French Startup Claims 30x Faster LLM Inference on Standard GPUs
21 sources0Kog achieves 3,000 output tokens/s per request on an 8× AMD MI300X node using a purely software driven inference engine — no custom hardware, no quantization, and no speculative decoding — by co designing a...
CERN’s ALICE Experiment Captures the Sharpest View Yet of Gluons Inside Atomic Nuclei
16 sources0The ALICE collaboration at CERN has made the first multidimensional measurement of incoherent J/ψ photonuclear production in ultra peripheral lead lead collisions, published in Physical Review Letters in Aug...
1,000x More Sensitive: KTH and Southampton's Flat Optical Fiber That Turns Glass into an Ultra-Sensor
13 sources0Researchers at KTH Royal Institute of Technology and the University of Southampton developed a high aspect ratio flat fiber (HARFF) that demonstrates pressure sensitivity up to 1,000 times higher than standa...
Meta's Muse Glimmer: The 30B Open-Weight AI Model That Runs Entirely on a Single Consumer GPU
40 sources0Meta's Muse Glimmer, released August 10, 2026, is a 29.6B parameter open weight agentic model (Apache 2.0) that runs entirely on a single consumer GPU — no cloud required.
Two Genetic Edits That Supercharge CAR-T Cells Against Solid Tumors
13 sources0The two genetic edits are knockout (loss of function) of P2RY8 and GNAS , discovered through the first ever genome wide CRISPR screen performed in human T cells inside living mice bearing solid tumors.
A New Study Wipes Out Decades of Dark Photon Dark Matter Constraints — Here Is What It Means for Experimental Searches
11 sources0A paper posted to arXiv on October 15, 2025, by Anson Hook, Junwu Huang, and Mohamad Shalaby demonstrates that decades old cosmological constraints on dark photon dark matter are invalid across roughly ten o...
Gaia Data Just Killed the 'Universal' Stellar Initial Mass Function—Here’s Why That Changes Everything for JWST
24 sources0Gaia DR3 data analyzed across 2 billion Milky Way stars shows the stellar initial mass function (IMF)—the distribution of stellar masses at birth—varies systematically between star clusters, overturning the...
JWST Spots Three Supermassive Black Holes in a Single Early Galaxy—A Cosmic First
15 sources0Using the James Webb Space Telescope, astronomers have made the first ever detection of three actively accreting supermassive black holes in a single distant galaxy, J0148 4214, seen just 1.2 billion years a...