the-decoder

Google Cloud enters multi-year partnership with Replit

Google Cloud has signed a multi-year partnership with the AI coding startup Replit as it looks to strengthen its position against competitors like Anthropic and Cursor. Under the agreement, Replit ...
the-decoder

DeepseekMath-V2 is Deepseek's latest attempt to pop the US AI bubble

Chinese startup Deepseek reports its new DeepseekMath-V2 model has reached gold medal status at the Math Olympiad, keeping the company in tight competition with Western AI labs. According to Deepseek, ...
the-decoder

Pinokio 5.0 turns local machines into personal AI clouds

Pinokio 5.0 wants to make running open-source AI models on your own hardware as easy as using a web app. Developer "cocktail peanut" positions the free tool as a local alternative to cloud platforms ...
the-decoder

Google rolls out Gemini 3 "Deep Think" for Gemini Ultra subscribers

The technology "builds on" on the Deep Think variant of Gemini 2.5, which recently posted impressive scores at the International Mathematical Olympiad and a major programming competition. To try it ...
the-decoder

Leaked "Soul Doc" reveals how Anthropic programs Claude’s character

In a lengthy LessWrong post, a user says he managed to extract an internal training document from Claude 4.5 Opus that defines the model's personality and ethical guidelines. Anthropic has confirmed ...
the-decoder

Deepseek V3.2 rivals GPT-5 and Gemini 3 Pro, reaches IMO gold level as open source

Chinese AI lab Deepseek has unveiled V3.2, a new language model that matches GPT-5 and Google's Gemini 3 Pro in key benchmarks and reasoning tasks. The team identified three main flaws in current open ...
the-decoder

OpenAI publishes prompting guide for GPT-5.1

OpenAI has released a prompting guide for GPT-5.1, its latest model designed to follow instructions with more precision. The guide walks developers through how to update existing workflows and adjust ...
the-decoder

Microsoft unveils Fara-7B, a compact model for running AI-driven computer control locally

Microsoft's new Fara-7B model is a compact AI system built to operate user interfaces purely through visual input. Despite its small size, it aims to keep pace with far more complex systems while ...
the-decoder

OpenAI CEO Sam Altman feels like he's "on the side of the angels" working on AI models

OpenAI CEO Sam Altman made religious references about artificial intelligence at a recent developer event in London, according to Hugging Face engineer Aymeric Roucher. "I don't pray for God to be on ...
the-decoder

Black Forest Labs secures $300 million to accelerate its vision for advanced image models

Black Forest Labs has raised USD 300 million in a new Series B round, bringing its valuation to USD 3.25 billion. Salesforce Ventures and Anjney Midha (AMP) led the round, with existing investors like ...
the-decoder

AI data centers could soon consume as much electricity as one-third of all U.S. households

Barclays estimates that data centers for AI giants like OpenAI, Meta, and Amazon could need as much as 46 gigawatts of electricity. That's about the same amount of power used by 44 million US ...
the-decoder

GPT-5 generates the "most impressive LLM output" yet, says OpenAI researcher

OpenAI researcher Sebastien Bubeck says GPT-5's math skills saved him a month of work. In a post on X, Bubeck reports that GPT-5 tackled a highly complex mathematical task for him. The model designed ...