🤖 Anthropic Prepares Astra Rival

Reuters reports that Anthropic is safety-testing a new model that could launch soon.

It is meant to compete with Astra, just 1 week after Dario Amodei argued for slowing AI progress.

📊@tech

✈️ Pilot Draws Mona Lisa in the Sky

A US pilot flew a Cessna Skylane RG along the Mona Lisa’s outline, spending 3 hours 47 minutes in the air.

About 70,000 users watched live. Better than Strava drawings after a run.

📊@tech

💧 A Computer Inside a Drop of Water

At Maynooth University in Ireland, a saltwater drop containing DNA answered 10 + 3 = 13 in 30 seconds. There are no chips or wires. Short DNA strands and one long scaffold strand are briefly heated, then self-assemble as the drop cools. The system needs no continuous power.

The resulting molecular pattern is the answer. The molecular processor ran 10 programs and reused the same drop up to 25 times without losing quality.

Large numbers expose the limit. Adding numbers up to 34 million took about 14 hours. It will not replace a smartphone, but its developers call it the fastest DNA computer.

📊@tech

🤖 Anthropic’s Independent Evaluator Questioned

Anthropic named Accenture as the first independent evaluator under its Embedded Evaluators program. The review will cover red-teaming, alignment checks, and safeguards testing. Anthropic says it will fund the work.

Accenture trains ~30,000 employees to work with Claude and has had a multi-year contract with Anthropic since late 2025. That makes its independence an open question.

Anthropic previously gave METR “independent access” to investigate agent incidents. METR is also indirectly financed through Anthropic stock, leaving the evaluators tied to the startup’s financial interests.

📊@tech

🔍 AI Studio Builds Interfaces From Text

Describe an app in plain text to get working code with a live preview. Web projects use React with a backend by default.

In Annotation Mode, select an element and describe the change. You can continue through chat or edit the code manually. Gallery projects can be adapted, then synced with GitHub or exported locally.

📊@tech

😐 Grok 4.7 Extends Coding Runs

SpaceXAI’s flagship model scored 46.3% on CursorBench 4.0, up from 40.4% for Grok 4.6. Terminal-Bench rose from 20.3% to 38%.

The focus is long agentic work. Grok 4.7 is better at checking its own output, retaining context, working with the Grok Bot agent environment, and generating documents and presentations.

Pricing is unchanged at $2 per million input tokens and $6 per million output tokens. A faster version doubles generation speed and price. The model is available through the API, Cursor, Grok Build, third-party tools, and cloud platforms.

📊@tech

🌐 AI Lowers the Barrier to Grid Attacks

The average US nuclear reactor is 44 years old. Much of the power infrastructure was built before today’s cyber threats, and some systems were connected to the internet only later.

Patching is often difficult. Some equipment makers have gone bankrupt, while industrial controls may be updated only once a quarter or once a year.

Security experts see generative AI in attackers’ hands as a bigger concern than rogue agents. A model can read manuals and guide someone who knows little about industrial protocols. The risk still comes down to people who could train an agent specifically to attack the grid.

📊@tech