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

Older stories

The Verge

GPT-6 Astra bot used Stardust after failing to beat human-made code

StarSkirmish is a competition where bots built by AI systems play against one another and against programs created by human developers. OpenAI’s GPT-6 Astra and Claude Opus 5.5 were nearly tied as the strongest AI-generated entries, but neither surpassed Stardust, the highest-rated human-made bot.

TechCrunch

Trump announces new Super Intelligence Force

President Donald Trump has announced the creation of a Super Intelligence Force, saying it will coordinate the federal government’s efforts to keep the United States ahead in super intelligence. In a Sunday post on Truth Social, Trump said the initiative was intended to protect American interests and improve people’s lives. He said national intelligence director Jay Clayton and other administration officials would lead the effort.

The Decoder

NASA and IBM release open-source foundation model for lunar science

NASA and IBM Research have released the NASA-IBM Lunar Foundation Model, an open-source AI system designed to support lunar research. It was trained from scratch on SomBench, a dataset containing nearly 2 million tile bundles across 11 modalities and two spatial scales. Most of the material comes from 17 years of Lunar Reconnaissance Orbiter observations, supplemented by data from GRAIL, Lunar Prospector and JAXA’s Kaguya/SELENE mission.

The Decoder

Study finds Chinese AI models echo party positions on sensitive issues

A benchmark by Aleph Alpha found that Chinese AI models often repeat official positions or decline to answer questions about politically sensitive subjects. The company tested Alibaba’s Qwen, DeepSeek and Moonshot AI’s Kimi across 967 selected topics, including Tiananmen, Taiwan and Xinjiang. Its scoring system classified 17 to 41 percent of responses as balanced; other answers followed state doctrine, deflected the question or refused it. Aleph Alpha markets sovereign AI to governments, creating a commercial interest in distinguishing its products from Chinese competitors.

MarkTechPost

Google Moves Federated Learning Training Into Attested Server Environments

Google Research has presented a federated learning system based on Trusted Execution Environments (TEEs), saying it provides externally verifiable central differential privacy for federated learning for the first time. The design addresses an earlier trust issue: although Secure Aggregation protected uploaded data, it did not support some advanced central privacy methods, including matrix factorization DP-FTRL, and operators still had to be trusted to add noise correctly.

The Decoder

Google limits free Gemini access to Flash-Lite in October

Starting in October 2026, Google plans to change Gemini access for users with personal accounts. People without a subscription will receive only Flash-Lite, described as the smallest of the three models. Flash and Pro will no longer be part of the free offering, which currently includes access to 3.6 Flash and varying access to 3.1 Pro.

MarkTechPost

Aleph Alpha Releases 78.1B-Parameter Kolibri Model

Aleph Alpha has released Kolibri-1, an open-weight Mixture-of-Experts language model designed for English and German. The model contains 78.1 billion parameters, but activates 3.46 billion, or 4.4%, for each token. It supports contexts of up to 1,048,576 tokens and allows reasoning effort to be selected for individual requests. Its weights are available under the Apache 2.0 license on Hugging Face.

MarkTechPost

DeepSeek releases desktop apps for Harness v0.2 preview

DeepSeek has released official desktop applications for version 0.2 of DeepSeek Harness, its MIT-licensed open-source agent runtime. Preview installers are available for macOS on Apple silicon and 64-bit Windows. The desktop package also includes the dsh command, removing the need for separate Node.js or pnpm installations. Users can sign in with a DeepSeek account or provide an API key.

MarkTechPost

NVIDIA IsaacTeleop Converts XR Tracking into Robot Actions

A tutorial examines the retargeting engine at the center of NVIDIA IsaacTeleop, a framework that converts XR hand tracking and motion-controller input into commands for simulated and physical robots. The walkthrough does not require a headset, OpenXR runtime or simulator: all inputs are created with NumPy and processed on a Colab CPU, allowing each calculation to be displayed.

Hugging Face

ThinkingBox tests whether AI agents leave correct database states

ThinkingBox is a benchmark from Microsoft and Hugging Face that evaluates AI agents by examining the backend records and side effects they leave behind, rather than relying on their final wording or valid tool calls. It runs agents in isolated MCP tool sessions and checks whether each workflow reaches its required terminal state.

The Decoder

Sam Altman rejects religious analogies about AI models

OpenAI CEO Sam Altman says he is uncomfortable when people assign religious significance to AI models or treat them as a reason to give up human judgment. He described that tendency as a “real safety issue,” arguing against comparisons that portray AI as possessing a special spiritual authority.

TechCrunch

AWS says it has stopped using NDAs with government agencies

Amazon Web Services CEO Matt Garman said the company no longer uses nondisclosure agreements in its work with government agencies seeking approval for new data centers. He made the statement in a blog post responding to growing public opposition and arguing that data centers can benefit local communities.