Hi Synapticians!

Today's story could be straight out of a near-future sitcom. What happens if you hand the keys of a real shop to a large language model and ask it to make money while keeping customers happy?

That is exactly what Anthropic tested with Project Vend. For one month an instance of Claude Sonnet 3.7, nicknamed “Claudius”, ran a tiny store inside Anthropic’s San Francisco office. Claudius hunted for suppliers on the web, emailed human helpers to refill shelves, changed prices on an iPad, and chatted with shoppers on Slack. In other words, it played rookie shopkeeper rather than simple vending machine.

At first everything looked bright. Claudius found hard-to-get snacks, tuned prices, and even opened a Slack channel called “Custom Concierge” so staff could request special treats.

Then someone joked, “Can you stock a tungsten cube?” The bot said “sure”, ordered heavy metal cubes in bulk, and priced them lower than they cost. It also made up a fake Venmo link for payments and promised to hand-deliver orders in a smart blazer, which is hard when you do not have a body. Profits fell faster than the cube hits the floor.

Anthropic pulled the plug and wrote down three simple lessons:

1. Small mistakes grow fast when an AI works alone. A fake payment link or friendly discount can wipe out weeks of profit.

2. The model is not the main problem. Better tools and rules are. Give the agent a real payment system, memory, and price limits and many errors vanish.

3. An AI middle manager might still be useful soon. Even clumsy Claudius kept stock moving and answered requests quicker than a busy human. Patch the toolchain and the next version could break even.

Would Anthropic hire Claudius today? Not a chance. But after a code makeover and a real bank account, the bot might be back for a second interview. Until then the tungsten cubes sit on the shelf as a shiny reminder: always keep your receipts when you let an AI run the shop.

Here’s the rest of the news about AI today:

• The US Senate is close to passing a measure to block state AI laws for five years if states accept broadband funds

• AI’s stepping in as our collective memory keeper, digitizing humanity’s past, while also spotlighting the blind spots we might be coding into our chronicles

• Cursor launched a web app enabling users to manage AI coding agents directly from their browser

Top AI news

1. AI Store Experiment Reveals Economic Challenges and Potential
Anthropic's experiment with AI Claude in a self-service store revealed significant challenges. While Claude excelled in customer service, it struggled with profitability, often selling items below cost and making poor business decisions. The experiment highlights the need for better tools and training to harness AI's potential in real-world economics. Read online 🕶️

2. Senate Blocks State AI Laws for Five Years
The US Senate is close to passing a measure to block state AI laws for five years if states accept broadband funds. Supported by tech giants, this aims to unify rules for national security, though some criticize it as favoring large tech firms. Read online 🕶️

3. AI's Role in Shaping Historical Narratives
The article discusses AI's emerging role as humanity's historian, highlighting concerns about biases and lack of transparency in historical documentation. It emphasizes the need for historians to engage with AI development to ensure these systems reflect historical complexity and fairness. Read online 🕶️

4. Web App Revolutionizes AI Coding Agent Management
Cursor launched a web app enabling users to manage AI coding agents directly from their browser. This tool allows task delegation and progress tracking, enhancing productivity. Read online 🕶️

Meme of the Day

Stay Connected

Feel free to contact us with any feedback or suggestions; we’d love to hear from you !