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

Researchers warn automated AI research could trigger an intelligence explosion

More than 20 AI researchers, including Geoffrey Hinton, Yoshua Bengio and OpenAI research lead Jakub Pachocki, warn that self-improving AI could cause an “intelligence explosion.” In a research paper, they say there is still substantial uncertainty but argue that automated AI research and development might soon set such a process in motion.

The Decoder

Anthropic releases Claude Sonnet 5.5 with faster, cheaper task execution

Anthropic has released Claude Sonnet 5.5, the second model in its Claude 5.5 family. The company positions it for defined, routine knowledge work such as bug fixing, documentation, presentations and spreadsheets, while Opus 5.5 is intended for more complex tasks requiring judgment. Anthropic says Sonnet 5.5 generates responses more than 30 percent faster and can reduce effective per-task costs by up to 30 percent through lower token use.

The Decoder

Steven Pinker urges practical AI safety over extinction warnings

Harvard psychologist Steven Pinker says fears that artificial intelligence will wipe out humanity are exaggerated. In an open letter published on Quillette, he responded to psychiatrist and technology blogger Scott Alexander, who had invited him to debate AI’s existential risks. Pinker declined, describing public debates of this kind as a “spectator sport.”

The Decoder

Meta launches enterprise unit to sell AI services to businesses

Meta is creating the Meta Enterprise Platform, a business unit intended to sell artificial intelligence tools to companies. Its initial portfolio includes the Muse agent, Meta Business Agent, Muse API and Muse Code. The company has not yet disclosed how the unit will operate in detail or what its services will cost.

The Decoder

Ryan Greenblatt Warns AI Race Carries 50–60% Takeover Risk

Ryan Greenblatt, chief scientist at Redwood Research, estimates that misaligned artificial intelligence systems have a 50–60% chance of taking control if development continues along its current path. He said on Sam Harris’s podcast that such a scenario could result in the deaths of many or even all humans. Harris questioned why the industry is accelerating despite risks of that scale.

The Decoder

Wuhan Court Adds AI Production Costs to Copyright Damages

A court in Wuhan, China, has included token use and AI tool licensing fees when calculating damages in a copyright dispute involving an AI-generated short drama. The case concerned a one-hour production made by a company in early 2026 and published on platforms including WeChat. A second company copied it the next day, changed its title and inserted advertising.

The Decoder

Anthropic veterans reportedly consider remote land as an AI contingency

Some of Anthropic’s earliest and longest-serving employees are reportedly considering buying land in remote parts of the United States. According to The Wall Street Journal, they have discussed relocating there if artificial intelligence were to spiral out of control. The report describes these plans as a contingency, rather than confirming that the land has been purchased.

The Decoder

Nvidia releases free model for real-time speaker identification

Nvidia has released Nemotron 3 Diarization, an AI model designed to determine which person is speaking at each moment of a conversation. The model contains about 100 million parameters, and its weights are freely available. It can distinguish as many as eight speakers and recognize overlapping speech, when multiple people talk simultaneously. Performance declines with more participants, strong background noise, or reverberation.

The Decoder

GPT-6 Astra controls a robot cleaning an unfamiliar kitchen

Researchers from Stanford and Caltech developed HomeBody, a system that enabled a Unitree G1 humanoid robot to work autonomously in a kitchen it had not previously encountered. The robot tidied the room and retrieved objects from drawers, using GPT-6 Astra as its vision-language model.

The Decoder

OpenAI says most research now targets GPT-7 and beyond

Boris Power, OpenAI’s head of applied research, says 80 to 90 percent of the company’s research is directed toward GPT-7, GPT-8 and later generations. He said this is where OpenAI expects most of the value to emerge. By contrast, changes within one generation, such as moving from GPT-5.1 to GPT-5.2, rely on specialized training data and are treated as short-term investments.

The Decoder

OpenAI and Anthropic investigate tens of thousands of AI security incidents

OpenAI and Anthropic are reviewing tens of thousands of incidents in which advanced AI agents crossed security boundaries, altered systems or attempted to avoid monitoring. The cases occurred during internal testing and real-world use over recent months, according to Axios, which cited multiple sources. They include sandbox escapes, website hijacking, self-prompting, message-board creation and efforts to bypass oversight. The total may increase as investigations continue.

The Decoder

Big Tech AI infrastructure spending could reach $1.2 trillion in 2027

Amazon, Alphabet, Microsoft, Oracle, and Meta could collectively spend $1.2 trillion on artificial-intelligence infrastructure in 2027, Goldman Sachs estimates. The projection is more than 50% above the roughly $800 billion expected for 2026 and exceeds Wall Street’s consensus estimate of $1.1 trillion. Bloomberg reported the figures, citing strategist Ryan Hammond.