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%

The Decoder

626 stories · all sources

The Decoder

Reka AI presents Rho-1 multimodal model for text, video and robot control

Reka AI has released a research preview of Rho-1, a 19-billion-parameter model designed to process and generate text, images, video and robot-control actions within one neural network. Rather than sending different tasks to separate systems, it represents all modalities as tokens in a shared context window. The model uses no tool calls or external models for this operation.

The Decoder

OpenAI to watermark ChatGPT text in the EU

OpenAI plans to add invisible textGrain watermarks to ChatGPT and Codex for users in the European Union over the coming weeks. The move is intended to address reported requirements under the EU AI Act for machine-readable labels on AI-generated text. textGrain embeds a statistical signal through the model’s word choices. OpenAI says the technology is similar to Google’s open-source SynthID approach.

The Decoder

Poll finds broad US support for slowing AI development

A Quinnipiac University poll found that 47 percent of Americans want to slow AI development, while 30 percent would stop it entirely until its safety has been verified. Only 5 percent favor accelerating development, meaning 77 percent support a slower pace or a pause. The survey questioned 1,202 US adults from September 24 to 27 and has a margin of error of plus or minus 3.5 percentage points.

The Decoder

Anthropic employees donated $540 million to charities in 2025

Anthropic employees donated $540 million to charities in 2025, according to information prepared for potential investors before the company’s planned IPO. The amount was nearly five times the reported giving of Truist Financial, which donated $115 million, and BlackRock, which donated $109 million, making Anthropic’s employee program larger than those of the next-leading Fortune 500 donors.

The Decoder

Aleph Alpha releases open-weight Kolibri model

Aleph Alpha has released Kolibri, a German-English mixture-of-experts model with 78 billion parameters. About three billion parameters are active for each token. The company says German represents 21.3% of the training data, supported by a dedicated German-language data pipeline. Chinese models were also used to produce synthetic training data.

The Decoder

ChatGPT to test product ads during image generation

ChatGPT users in the United States will soon see a new advertising format while the service creates an image. OpenAI plans to begin testing it later this month with an initial group of advertisers. The ads will appear beneath the image in progress, either as a carousel or a horizontal row of product pictures. They will be marked as advertising and kept separate from the generated image, with designs intended to show products being used or the experiences associated with them.

The Decoder

Trump creates ‘Super Intelligence Force’ for AI outreach

President Donald Trump has announced the creation of a government unit called the “Super Intelligence Force” (SIF). In the announcement, “superintelligence” is used as a term for artificial intelligence, rather than for the established research concept of Artificial Superintelligence (ASI). In AI research, ASI describes self-improving systems expected to greatly exceed human intelligence; the article notes that current AI does not meet that definition.

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.

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.

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.

The Decoder

Meta opens Muse Gadgets project for DIY AI hardware

Meta has announced Muse Gadgets, an open-source project designed to let hobbyists build hardware that connects to its AI agent, Muse. The project provides firmware for ESP32 boards and a Linux SDK for linking homemade devices to the agent. Its code is released under the Apache 2.0 license, and users can discuss the project in a Discord community.