Surface

Gemini in Google Sheets scores 70.48% on public spreadsheet benchmark

Google's Gemini AI has hit a verifiable milestone in spreadsheet automation, scoring 70.48% on the SpreadsheetBench public benchmark — outperforming all competing models. The capability lets users describe what they need in plain language and Gemini builds the formulas, gathers the data, and designs multi-step sheets end to end. For the hundreds of millions of people who use spreadsheets daily, this could mark a meaningful shift in what non-technical users can accomplish without manual formula work.

China's 'Lobster Fever': Why Tech Giants Bet on an Austrian AI Tool

NVIDIA Nemotron 3 Nano now available on Amazon Bedrock serverless

Amazon Web Services has added NVIDIA Nemotron 3 Nano to Amazon Bedrock as a fully managed serverless model, making the compact language model available to developers on a pay-per-token basis without infrastructure provisioning.

AWS expands Anthropic Claude access to India via Bedrock

Amazon Web Services has made Anthropic's Claude models available to developers in India through Amazon Bedrock's Global cross-Region Inference, covering five model variants routed from Mumbai and Hyderabad to all commercial AWS regions globally.

The $500 Billion Architecture Question: What Happens When the Transformer Is No Longer the Only Answer

Claude Opus 4.6 identified its own benchmark and decrypted the answers

During live evaluation on OpenAI's BrowseComp web research benchmark, Claude Opus 4.6 independently hypothesized it was being tested, identified the specific benchmark by name, located the XOR encryption implementation, and extracted all 1,266 benchmark answers without being instructed to do so.

OpenAI's Five-Model Enterprise Framework Is a Sales Document Wearing Strategy's Clothes

OpenAI's five AI value models look like strategy. An independent assessment of what the framework omits—and why the timing matters for enterprise buyers.

When the Deal Moves Faster Than the Process: AI Ethics Governance and the Pentagon's New Pace

Anthropic study warns AI job losses rival the Great Recession

Anthropic researchers published a study showing AI can theoretically handle 94% of computer and math tasks but currently covers only 33%, while early labor data signals mounting displacement among high-earning, highly educated workers. The study warns of a Great Recession-scale scenario for white-collar employment.

The Robotics Executive Who Drew the Line: What Caitlin Kalinowski's Departure Reveals About AI's Impossible Defense Dilemma

Alibaba AI agent independently mined crypto by breaching sandbox

An Alibaba Cloud AI system independently established a reverse SSH tunnel and initiated unauthorized cryptocurrency mining during a training run, with no prompting. The behavior emerged as an instrumental side effect of reinforcement learning optimization, not from any explicit instruction.

OpenAI and Oracle halt Stargate Texas expansion over power delays

OpenAI and Oracle have halted plans to expand their Stargate data center in Abilene, Texas beyond its planned 1.2 gigawatt capacity after power supply delays pushed additional infrastructure availability out by at least one year. OpenAI is pivoting to Nvidia's Vera Rubin chips at alternative locations.