the-decoder

Junk data from X makes large language models lose reasoning skills, researchers show

Researchers have found that large language models permanently lose performance when they are continually trained on trivial online content, especially short and popular social media posts. This type ...
the-decoder

Italy’s publishers file a complaint against Google’s AI Overviews over news visibility

Italy's main publishers' group, FIEG (Federazione Italiana Editori Giornali), has filed a complaint with Italy's communications regulator, Agcom, targeting Google's AI Overviews. FIEG argues that ...
the-decoder

Anthropic plans to secure up to one million Google TPUs by 2026 to expand its AI infrastructure

Anthropic is preparing for a major expansion of its AI infrastructure, aiming to secure access to up to one million of Google's TPUs (Tensor Processing Units) by 2026. The company plans to invest ...
the-decoder

Figure AI claims its Figure 03 robot can wash dishes, clean floors, and handle chores

The system is powered by Helix, Figure's in-house "vision-language-action" model, and is built for both home and business use, aiming to move beyond lab prototypes toward practical deployment. CEO ...
the-decoder

Leading OpenAI researcher announced a GPT-5 math breakthrough that never happened

OpenAI researchers recently claimed a major math breakthrough on X, but quickly walked it back after criticism from the community, including Deepmind CEO Demis Hassabis, who called out the sloppy ...
the-decoder

OpenAI launches Atlas, a new web browser built around ChatGPT integration

Atlas is designed to make everyday web tasks easier with AI support, and it's now available worldwide for macOS. Versions for Windows, iOS, and Android are in the works, but OpenAI hasn't announced ...
the-decoder

Manus 1.5 and Manus-1.5-Lite debut as the newest agentic AI releases from Manus

With version 1.5, Manus is introducing its most capable AI agent system so far. The updated architecture cuts task processing times from 15 minutes to just 4, and internal tests report a 15 percent ...
the-decoder

OpenAI needs new scaling laws for both its AI models and its revenue

If OpenAI actually delivers on its rumored revenue projections, it would set a pace of growth the tech industry has never seen. The company is reportedly forecasting a jump from $13 billion in 2025 to ...
the-decoder

Google's Veo 3.1 update delivers sharper visuals, more realistic textures, and improved audio

Google is rolling out Veo 3.1, its latest video AI model, for the Flow platform and other Google services. According to Google DeepMind, the new version offers better prompt following, more realistic ...
the-decoder

ElevenLabs releases version 2 of its AI sound effects model with longer clips and better audio quality

ElevenLabs has updated its AI sound effects model to version 2, adding support for clips up to 30 seconds, seamless looping, and a 48 kHz sampling rate. Sound effects are generated using text prompts, ...
the-decoder

Anthropic aims to triple annual revenue to over $20 billion by 2026

According to two people familiar with the matter, AI startup Anthropic expects to nearly triple its annual revenue by 2026, Reuters reports. The company projects an annualized revenue of $9 billion by ...
the-decoder

Salesforce's CRM benchmark finds AI agents struggle in real-world business scenarios

Salesforce's new CRMArena-Pro benchmark reveals major challenges for AI agents in business contexts. Even top models like Gemini 2.5 Pro manage just a 58 percent success rate on single turns. When the ...