Sakhanda Wire
NVDA $150.96 -2.89% MSFT $506.79 +0.11% GOOGL $283.77 +1.27% META $615.78 -0.50% AMZN $266.86 +3.48%

Latest news

Older stories

MarkTechPost

HarnessDev finds most agent-harness changes fail to generalize

Researchers from ByteDance Seed, Singapore University of Technology and Design, Georgia Institute of Technology, M-A-P, and TokenWave.AI introduced HarnessDev, a benchmark that evaluates the runnable software surrounding a language model rather than only its answers. The harness includes execution, tools, context, state, recovery, and verification. Six models started from the same minimal seed, which scored 0 without a loop, planner, verifier, retry, or stopping rule, and built harnesses for five benchmarks covering four domains.

MarkTechPost

Anthropic Adds Plugin Evaluation Workflow to Claude Code

Anthropic has introduced a plugin evaluation workflow for Claude Code that tests plugins with realistic prompts, grades the resulting work, and compares it with a run where the plugin is absent. The workflow is available in Claude Code v2.1.269 and later for plugins containing a plugin.json or .claude-plugin/plugin.json manifest, as well as skills-directory plugins. Model calls used for evaluations and judging are billed to the relevant plan or API account.

TechCrunch

Garry Tan backs open-weight distillation by U.S. AI labs

Y Combinator CEO Garry Tan said U.S. regulators should not prevent American open-weight AI developers from using distillation techniques on domestic frontier models. He told CNBC that he would favor an “American distillation regime,” and explained that smaller labs should be able to learn from leading U.S. systems to expand the country’s supply of open-weight models rather than leaving that field dominated by Chinese alternatives.

TechCrunch

Fields Medalists warn AI race threatens openness in mathematics

Twenty-five mathematicians who have received the Fields Medal signed an open letter warning that competition among AI companies could endanger the way mathematical research is conducted. The signatories said rapidly announced AI-generated solutions may leave too little time to verify results, explain new methods, identify earlier contributions, and establish proper credit.

The Verge

New Mexico lawyer fined $5,000 over fabricated AI testimony

New Mexico’s Supreme Court fined lawyer Stephen Aarons $5,000 and held him in contempt after finding that an appeal brief for a client convicted of murder included material generated or fabricated by artificial intelligence. The action was reported by Reuters, citing a court filing issued on Wednesday.

TechCrunch

Exhibit tables for TechCrunch Disrupt 2026 available until September 18

Exhibit tables for TechCrunch Disrupt 2026 can be booked until September 18 at 11:59 p.m. PT, although the remaining spaces may sell out earlier. The exhibitor program offers a dedicated 6′ × 30″ table in the Expo Hall for all three days of the event, allowing participating teams to demonstrate products and speak with attendees.

MIT Tech Review

AI extinction fears to be examined in upcoming roundtable

An upcoming online roundtable will examine claims that advanced artificial intelligence could destroy humanity. Employees at leading AI laboratories have said that this outcome is a real possibility, prompting debate over whether the concern is justified or amounts to scaremongering and hype.

TechCrunch

Moonshot AI targets $2 billion in annualized revenue

Moonshot AI is aiming for $2 billion in annualized revenue by the end of the year, Bloomberg reported on Friday. That would be twice the company’s reported revenue run rate for August and follows the strong adoption of its open-weight K3 model, released this summer. K3 usage has fallen slightly in recent months, but OpenRouter data currently indicates that K3 models generate as many as 300 billion tokens daily on the platform.

Ars Technica

New Mexico lawyer held in contempt over fabricated ChatGPT testimony

The New Mexico Supreme Court found defense attorney Stephen Aarons in direct contempt of court after he submitted a brief containing false testimony attributed to entirely fabricated witnesses. The filing included invented police testimony and other factual and legal errors. The court also referred Aarons to a disciplinary board for additional proceedings.

TechCrunch

Anthropic researcher resigns with warning about self-improving superintelligence

An Anthropic researcher resigned this week and posted a warning on X, saying the company was “racing straight to self-improving superintelligence and gambling with our lives.” Anthropic’s alignment lead co-signed the message instead of withdrawing or softening it. The warning reflects concerns that have appeared before in the artificial intelligence industry about the risks of increasingly capable systems.

The Decoder

Vinyals expects gradual AI self-improvement, not an intelligence explosion

Oriol Vinyals, until recently Google DeepMind’s vice president of research, expects AI systems to improve themselves over time but considers a sudden intelligence explosion unlikely. Speaking at the Agentic AI Summit 2026, he said AI could accelerate some research and engineering work by a factor of ten or more, while remaining constrained by physical limits such as hardware performance and the speed of light.