the-decoder

Here is what an LLM that knows nothing after 1930 thinks our world looks like in 2026

"Talkie" is a 13B-parameter language model trained only on texts written before 1931. It doubts a second world war will happen and pictures 2026 as a world of steamships, railroads, and penny novels.
the-decoder

Google signs AI deal with the Pentagon, ignoring protest from over 600 employees

Google has signed a contract letting the Pentagon use its AI models for classified tasks, on the same day over 600 employees protested the deal in an open letter. Legal experts say the contract's ...
the-decoder

Meta wants to power AI data centers with solar energy from space

Meta has signed a deal with startup Overview Energy for up to one gigawatt of space-based solar power to feed the massive energy demands of its AI data centers. Satellites would collect sunlight in ...
the-decoder

China blocks Meta's $2 billion acquisition of AI startup Manus

China's economic authority NDRC is forcing Meta to unwind its already completed $2 billion acquisition of AI startup Manus. Reversing the deal is complicated since Meta has already integrated the ...
the-decoder

OpenAI kills its dedicated coding model Codex again, folding it into GPT-5.5

With GPT-5.4, OpenAI rolled its separate Codex model for programming tasks into the main model. There's no dedicated coding line anymore, explains Romain Huet, OpenAI's Head of Developer Experience.
the-decoder

OpenAI says old prompts are holding GPT-5.5 back and developers need a fresh baseline

OpenAI has released a prompting guide for GPT-5.5 with one major takeaway: don't reuse your old prompts. Start fresh with minimal, result-focused instructions. And role definitions, once dismissed as ...
the-decoder

Qwen3.6-27B beats much larger predecessor on most coding benchmarks

Alibaba has released Qwen3.6-27B, a new dense open-source model with 27 billion parameters. According to Alibaba, the model outperforms its much larger predecessor, Qwen3.5-397B-A17B (397 billion ...
the-decoder

Anthropic says stronger AI models cut better deals, and the losers don't even notice

In a week-long experiment, Anthropic let Claude agents buy and sell goods for employees. The result: stronger models negotiated better prices. The catch: the people stuck with weaker agents had no ...
the-decoder

The UAE wants half its government run by autonomous AI agents within two years

The United Arab Emirates is rolling out what may be the world's most ambitious government AI overhaul. Within two years, the country wants 50 percent of all government sectors, services, and processes ...
the-decoder

Anthropic confirms Claude Code problems and promises stricter quality controls

Following user complaints about declining quality in Claude Code, Anthropic identified and fixed three separate bugs affecting reasoning depth, caching, and text length restrictions. To prevent ...
the-decoder

China moves to block tech firms from taking US money without government approval

China plans to block tech companies from accepting US capital without government approval, according to Bloomberg. China's National Development and Reform Commission (NDRC) has reportedly told several ...
the-decoder

As agentic AI pushes rivals to raise prices and cap usage, Deepseek ships a good-enough model for almost nothing

Chinese AI lab Deepseek has released V4-Pro and V4-Flash as open-weight models with up to 1.6 trillion parameters and a one-million-token context window. A new architecture dramatically cuts the ...