Ternary Computing Died in 1965. AI Inference Just Reinvented the Only Part That Mattered.
A Soviet machine called Setun ran on minus one, zero and plus one from 1958 and was killed off in 1965. Ternary LLM weights,…
A Soviet machine called Setun ran on minus one, zero and plus one from 1958 and was killed off in 1965. Ternary LLM weights,…
Adreno Neural Fusion adds AI matrix units to Qualcomm's GPU a year after Apple and level with Arm, but the 18MB of HPM is…
Sony's PSSR 2.0 significantly reduced training time from four GPU-months to under one week by simplifying its neural network's functions. It switched from a…
OpenAI paused internal work on its Astra model due to concerns it might enable autonomous cyberattacks, highlighting safety issues in AI development. Meanwhile, Meta's…
D-Wave's recent paper in Nature reveals a significant advance in error correction, showing a 99.9% fidelity using dual-rail erasure qubits. This architecture aims to…
The article discusses the evolution of NVIDIA's AI technologies, particularly the Tensor Cores and their role in enhancing graphics rendering through techniques like DLSS…
AI inference, a crucial aspect of enterprise computing, involves a single forward pass through a neural network without learning or adapting post-training. The costs…
A team from Copenhagen and CU Boulder developed a harmonic parity qubit, enabling control over odd harmonics in a superconducting circuit. This advancement aims…
AMD and Anthropic announced a partnership involving 2 gigawatts of AMD GPUs in Helios systems, supported by a $5 billion equity investment. The collaboration…
Alibaba's Qwen team launched the Qwen3.8-Max, a 2.4 trillion parameter model, outperforming competitors in key benchmarks. While it promises "open weights" soon, skepticism remains…
On August 1, OpenAI unveiled "Astra," a model that solved ten long-standing mathematical problems, including group theory's non-sofic groups. However, concerns arise over Astra's…
PLA Unit 96941 developed a method to summarize sensitive military code by distilling outputs from GPT-3.5 into a domestic model, avoiding direct API risks.…
In 2026, AI liability laws' landscape shifted dramatically. California eased court access for AI harm victims, while Europe withdrew an extensive liability directive, complicating…
The AI Liability Directive (AILD), designed to address the black-box problem in AI accountability, has been withdrawn due to lack of consensus among EU…
The EU AI Act's enforcement begins Sunday, but key regulatory components face delays. Originally intended for full implementation, many high-risk requirements are postponed until…
One maker's tech digest, weekly. Chips, AI, gaming, EVs — no spam.