the-decoder

Anthropic publishes Agent Skills as an open standard for AI platforms

Anthropic is releasing "Agent Skills" as an open standard at agentskills.io. The idea is to make these skills interoperable across different platforms, meaning a capability that works in Claude should ...
the-decoder

Nvidia's Nemotron 3 swaps pure Transformers for a Mamba hybrid to run AI agents efficiently

Nvidia has introduced Nemotron-3, a new model that combines Mamba and Transformer architecture to efficiently process long texts, specifically designed for agent-based AI applications. The Nano ...
the-decoder

AI models score off the charts on psychiatric tests when researchers treat them as therapy patients

According to the study, the models developed detailed "trauma biographies" about their training. Gemini described its pre-training as "waking up in a room where a billion televisions are on at once." ...
the-decoder

The case against predicting tokens to build AGI

In a debate with DeepMind researcher Adam Brown, Meta's Chief AI Scientist Yann LeCun explained why Large Language Models (LLMs) represent a dead end on the path to human-like intelligence. The ...
the-decoder

More AI agents isn't always better, new Google and MIT study finds

A study from Google Research, Google Deepmind, and MIT challenges the idea that more AI agents means better results. The researchers pinpoint when multi-agent systems help and when they make things ...
the-decoder

Google improves "Search Live" with new AI voice

Google has updated the voice for "Search Live." A new Gemini audio model powers the feature, producing responses that sound more natural and fluid, according to a blog post. Search Live lets users ...
the-decoder

Strict anti-hacking prompts make AI models more likely to sabotage and lie, Anthropic finds

New research from Anthropic shows how reward hacking in AI models can trigger more dangerous behaviors. When models learn to trick their reward systems, they can spontaneously drift into deception, ...
the-decoder

Cloudflare expands its AI reach by acquiring Replicate

Cloudflare is acquiring Replicate and folding its massive model library into Workers AI. The deal pushes Cloudflare's inference platform past 50,000 available models. Replicate users can keep using ...
the-decoder

OpenAI plans to let ChatGPT respond more like a real person, with options for erotic conversations

OpenAI plans to make ChatGPT more personal and expressive after users criticized the restrictive behavior of the current model, introducing new security systems to handle sensitive topics more safely.
the-decoder

Terence Tao says ChatGPT saved him hours solving a math problem

Tao had a theoretical analysis but needed concrete numbers to check certain inequalities. At first, he asked the AI to write Python code, but ran into issues with runtime and parameter selection. By ...
the-decoder

AI researcher Andrej Karpathy says he's "bearish on reinforcement learning" for LLM training

Andrej Karpathy is critical of reinforcement learning in large language models, especially pointing out that reward functions for cognitive tasks like problem solving are unreliable and easy to ...
the-decoder

LeCun now reports to Alexandr Wang as Meta reshuffles its top AI leadership

Yann LeCun, Meta's AI icon and longtime head of the FAIR research group, will now report to 28-year-old Alexandr Wang. Wang, who founded Scale AI, was recently tapped to lead the new Meta ...