Sakhanda Wire
NVDA $155.46 -0.44% MSFT $506.24 -2.27% GOOGL $280.20 -1.24% META $618.87 -1.20% AMZN $257.89 +0.05%

Latest news

Page 28 of 254

MarkTechPost

MoE routing strategies shape model quality, balance and serving costs

Mixture-of-Experts models use a router for each token, selecting a small number of expert networks instead of running all of them. The choice affects model quality, workload balance and the communication costs of multi-GPU serving. Mixtral 8x7B, for example, exposes 47B parameters but activates 13B for each token.

MarkTechPost

Arete-27B cuts propaganda responses in Qwen3.8-27B fine-tune

Nous Research has released Arete-27B, a 27B dense open-weight model created by supervised fine-tuning Alibaba’s Qwen3.8-27B. The model keeps the base architecture and Apache 2.0 license. The training used no reinforcement learning or preference optimization, and the vision encoder remained frozen because the dataset was text-only.

TechCrunch

Midterm Results Increase Pressure on Data Center Expansion

Democrats’ historic victory in the midterm elections is expected to create new challenges for technology companies, particularly hyperscalers and artificial intelligence firms. Voters in many local ballot measures approved restrictions or moratoriums on data centers. In Texas, governor-elect Gina Hinojosa campaigned on limiting such projects after defeating Greg Abbott.

TechCrunch

Memo claimed executive was recruited to spy on Nvidia CEO

A memo circulated earlier this year reportedly said that China’s Ministry of State Security recruited technology executive Alice Huang to spy on Nvidia CEO Jensen Huang. Multiple media reports described the document as based on U.S. intelligence. The New York Times, which had earlier reported that the Nvidia chief was a target of Chinese intelligence, was the first to publish details about the memo.

MarkTechPost

Guide details how to run NVIDIA Dory 1.0 4B with llama.cpp

A coding tutorial examines Dory 1.0 4B, a compact reasoning model developed by NVIDIA’s Nemotron Labs. Its middle layers form a shared block that can be executed one, two, three or four times for each token. Increasing the loop count raises computation per token without increasing the parameter count, allowing one checkpoint to operate at four compute levels without retraining.

The Decoder

Anthropic advances pre-Thanksgiving IPO at lower valuation

Anthropic is moving forward with plans for an initial public offering before Thanksgiving despite this week’s technology selloff. Investors cited by the Financial Times said the developer of Claude is seeking to raise between $80 billion and $100 billion, at a valuation of $1.25 trillion to $1.75 trillion.

The Decoder

Inkling 2 Small produced 316 expletives in benchmark runs

An analysis by Vals AI found that Thinking Machines’ Inkling 2 Small used the word “fuck” or variations of it 316 times during 2,655 runs across 13 benchmarks. Profanity appeared in 50 runs, while 241 instances came from only two exceptionally long runs.

The Verge

NYT Cooking adds an AI assistant for everyday meal decisions

The NYT Cooking app now includes a built-in AI assistant designed to help with everyday cooking. The article describes a recurring weeknight problem: around 7 p.m., the author stands in front of the refrigerator trying to choose a dinner that will not prompt another complaint about chicken. Meal planning is acknowledged as a possible solution, but it is not part of the routine described.

The Verge

US Approved H200 Chip Recipient Despite Empty Chinese Office

Questions are being raised about how the US Commerce Department and Nvidia vet Chinese companies approved to handle Nvidia’s H200 artificial-intelligence chips. The concerns follow reporting that New Matrix appears on the list of authorized firms, even though its listed office in China was found empty.

The Decoder

Open and closed AI models show narrowing gap in cyber capabilities

The Vals CyberOffense benchmark found that AI agents can independently attack simulated corporate networks, starting with only a public web address. Running from isolated Kali Linux machines, the agents move through web applications and Windows servers toward administrator access, while an external system records and scores every action. GPT-6 Sol Daybreak achieved the strongest result, reaching 80.8 percent of the objectives. The open-weight Deepseek V4.1 Flash 1017 reached 72.8 percent, at about $3 per run compared with roughly $22 for GPT-6 Sol Daybreak.

The Verge

AI is reshaping cybersecurity as automated attacks accelerate

Cybersecurity is entering a period of rapid change as artificial intelligence helps attackers discover vulnerabilities and exploit them in real time, according to Alex Stamos, chief security officer at Cognition and a longtime security specialist. He described recent weeks as a succession of breaches affecting AI companies, neocloud providers and other technology firms.

The Decoder

Fine-tuning cuts propaganda in Alibaba’s Qwen3.8-27B to 0.9 percent

Nous Research says it retrained Alibaba’s open-weight Qwen3.8-27B to reduce responses reflecting Chinese state propaganda. The resulting model, Arete-27B, was trained with supervised fine-tuning on about 53 million tokens over two epochs. The process used neither reinforcement learning nor preference optimization, and the team removed overlaps between training and test prompts.