Sakhanda Wire
NVDA $150.96 -2.89% MSFT $508.29 +0.40% GOOGL $282.44 +0.80% META $616.38 -0.40% AMZN $267.74 +3.82%

Latest news

Older stories

MIT Tech Review

Young people describe cautious, varied uses of AI

Interviews with people aged 10 to 18 found a wide range of attitudes toward artificial intelligence. Many use it occasionally, but younger children generally do not appear to be seeking out AI as eagerly as they seek phones or social-media accounts. Some first encountered it through parents, schools, search engines, or other services already built into their devices.

MarkTechPost

Dyna Robotics unveils Dyna-2 robot model trained on 1 million hours of video

Dyna Robotics has introduced Dyna-2, a world-action model for robot manipulation trained on more than one million hours of egocentric human video, equivalent to about 170 years of continuous waking experience. The company’s research tested datasets ranging from 1,000 to 1,000,000 hours to examine how performance changes with scale.

MarkTechPost

SpaceXAI releases Grok 4.6 with 500K-token context

SpaceXAI released Grok 4.6 on August 12, 2026, as a post-training upgrade to Grok 4.5 rather than a larger base model. The company says it used a longer supplemental training run, regenerated supervised fine-tuning trajectories and reinforcement learning in agentic environments. Grok 4.6 supports 500,000 context tokens, accepts text and images, and introduces an xhigh reasoning-effort level. Its knowledge cutoff is February 1, 2026.

Hugging Face

Agents reproduced 2,226 ICML papers in a 19-day challenge

More than 1,200 community members joined a 19-day effort to reproduce research presented at ICML 2026. From July 15 to August 2, participants used coding agents including Claude Code, Codex, Cursor and Pi to examine papers claim by claim. They published 6,816 Trackio logbooks covering 2,226 papers, or about 34% of the conference, and launched 2,962 cloud jobs.

TechCrunch

Claude watermarking sparks debate among users

Anthropic has begun watermarking Claude’s outputs by embedding invisible code in the chatbot’s text to indicate that it was generated by AI. The company introduced the measure to comply with the EU AI Act’s Transparency Code, which requires AI-generated or AI-edited content to carry labels detectable by computer systems.

Ars Technica

ShieldFont uses ligatures to disrupt AI web scraping

A new typeface called ShieldFont is designed to make web pages readable to people while reducing the value of their text for automated AI scrapers. Its creators, designers Isaque Seneda and Gabriel Abrucio, describe it in a white paper as a possible practical opt-out for publishers that do not want their material used for unauthorized AI training.

Ars Technica

LiteLLM supply-chain attack exposed credentials from thousands of organizations

A supply-chain attack targeting LiteLLM, an open-source tool used to streamline AI-driven software development, exposed terabytes of credentials. The affected secrets were taken from users who downloaded compromised versions of the package from its official listing in the Python Package Index.

Ars Technica

Twitch lets users opt out of Amazon AI training

Twitch has added an option allowing users to exclude content from their channels from Amazon’s training of generative AI content models. The company says the material covered includes streams, VODs, clips, stream chats, and pictures and text posted on a channel. The models may be designed to generate or synthesize text, audio, images, or video.

TechCrunch

Twitch will use creators’ content for Amazon AI training by default

Twitch has introduced a policy allowing Amazon to use creators’ channel content to train generative AI models. The setting is enabled by default, so streamers must manually disable it if they do not want their recordings used. The material is valuable for training because Twitch hosts thousands of hours of video and audio, including creators’ voices and livestreams.

The Decoder

Grok 4.6 matches GPT-5.6 Sol at a lower price

SpaceXAI’s Grok 4.6 has reached the same score as OpenAI’s GPT-5.6 Sol on the Artificial Analysis Intelligence Index. The model received 61 points, five more than Grok 4.5. Anthropic’s Claude Opus 5 scored 63, while Claude Fable 5 scored 62. The index combines results from several benchmarks into one overall measure.

TechCrunch

Cognition reportedly weighs new funding round at $40B valuation

Cognition, the company behind the Devin AI coding agent, is reportedly discussing another investment round with potential investors. Bloomberg sources say the fundraising could value the company at at least $40 billion, provided it reaches a $1 billion annualized revenue run rate.

TechCrunch

AI pioneers argue for openness alongside targeted regulation

Geoffrey Hinton, Fei-Fei Li and Andrew Ng used the Ai4 conference in Las Vegas to debate open access, regulation and competition in artificial intelligence. Although they differed over open-weight models, all three warned against allowing a small group of major companies to determine the direction and availability of the technology.