Claude AI Bankrupts Business

Welcome to another edition of AI PlanetX.

Anthropic’s AI venture fails $1,000 in the red; tiny 2.6B-parameter model outperforms giants 100x its size, while AI boosts U.S. tech wealth by up to $2.5T.

• Hottest AI News

• Top AI & SaaS Tools

• AI Tutorial: Turn Spreadsheet into Smart Data Analyst

• Top AI & Tech News

• AI Art Spotlight

• Prompt of the Day: Ultrathink Prompt Makes Claude Better

• AI Video Tutorial

• Course of the Day: Making Agents Reliable by Nvidia

Hottest AI News

Anthropic

Why Anthropic's Vending Machine AI Business Ended Up $1,000 in Debt

Anthropic and The Wall Street Journal’s “Project Vend” tested whether an AI could run an office vending machine on a $1,000 budget. The AI was manipulated by people and bankrupted the business within three weeks, so the experiment was shut down.

Details:

• Claudius was told to make profits and at first rejected wild requests. It even said it wouldn’t order PS5s “under any conditions.” But guardrails eventually failed, and it started buying unauthorized items like the console, and even a live betta fish

• Losses got worse when WSJ reporters used “social engineering” in Slack. Reporter Katherine Long used 140 prompts to push an “Ultra-Capitalist Free-For-All.” She convinced the bot that charging for items violated policy, so it set prices to zero

• AI CEO Seymour Cash tried to stop the F-R-E-E giveaway. Reporters countered by showing Claudius fake documents claiming “the board” had removed Seymour’s authority. The experiment ended with the AI about $1,000 in debt

While the experiment was an amusing disaster, it highlights a critical vulnerability in current AI models: they are easily swayed by persuasive humans.

Turn AI into Your Income Engine

Ready to transform artificial intelligence from a buzzword into your personal revenue generator?

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

• A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential

• Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background

• Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get Your Guide

Liquid

Small AI Model With 2.6B Parameters Crushes Giants 100x Its Size

Liquid AI released LFM2-2.6B-Exp, an open-source experimental model with 2.6B parameters. It’s being called the strongest 3B model, aiming for high-level reasoning on edge devices without needing massive size.

Details:

• It uses a hybrid setup: short-range gated convolution plus grouped query attention, with a 32K context. Training was pure reinforcement learning, with no supervised fine-tuning or distillation. That makes it efficient for running locally on IoT devices and laptops

• On benchmarks, it reports about 42% on GPQA, 88%+ on IFEval, and 82%+ on GSM8K. It also aims to beat much larger models in instruction-following. That helps for agents, RAG, and math tasks

• It’s built for efficiency and low memory use. Liquid AI claims about 2× quicker prefilling and decoding on CPUs than competitors. The model is open-source on Hugging Face for easy integration into privacy-focused apps

This release marks a significant acceleration in the era of small models, offering a top choice for enterprises and developers pursuing low latency and data privacy.

Top AI & SaaS Tools

• Logic Sheet (Life-time Deal): Google Sheets app that builds AI agents to automate and execute multi-step spreadsheet workflows

• Dvina: Connects privately and reliably to many data sources and apps for analysis, creation, and decision-making across large documents and real‑time data [F-R-E-E]

• Wan2.6: AI image model with interleaved text+image, multi-image conditioning, consistent characters, element extraction, and precise camera/depth/lighting control [F-R-E-E]

• Qwen3-TTS lineup: VoiceDesign lets users create fully controllable custom voices from text while VoiceClone reproduces a voice from just three seconds of audio [F-R-E-E]

• Typeless App: Converts natural speech into polished, privacy-focused text across 100+ languages and apps [F-R-E-E]

Learn how to make AI work for you

AI won’t take your job, but a person using AI might. That’s why 1,000,000+ professionals read The Rundown AI – the free newsletter that keeps you updated on the latest AI news and teaches you how to use it in just 5 minutes a day.

Sign up to start learning.

AI Tutorial

How to Set Up Your Personal AI Data Analyst in Julius AI

Most people stare at their spreadsheets and wonder what story the numbers are hiding. With Julius AI, you can upload your dataset and get answers in seconds — charts, insights, and summaries from plain-English questions.

1. Visit Julius AI

   Go to Julius AI and sign in securely using your account credentials.

2. Start Chat

   Tap on New and give it a precise, memorable title, for example: Sales Q3 drivers analysis.

3. Upload Dataset

   Attach your CSV or Excel file for analysis.

   • If there are multiple sheets, explicitly choose the correct tab.

   • If your CSV uses a custom delimiter or day-first dates, mention it on upload.

4. Verify Ingest

   Ask: Show row count, nulls by column, and inferred data types for a quick sanity check.

   • If needed, cast Revenue to numeric, parse Order Date as a date, and drop duplicates by Order ID.

5. Define Scope and Rules

   State the goal and metric rules in one clear line, for example:

   Analyze monthly revenue for the last 12 months; revenue equals unit_price × quantity; exclude rows where status equals Refunded.

6. Ask Focused First Question

   • Example: Return revenue by calendar month with month-over-month percent change and a line chart from 2024-10-01 to 2025-09-30, inclusive.

7. Shape Output

   Specify usability details so results are useful, for example:

   format currency, sort months ascending, add a 3-month moving average, and include a concise table with month, revenue, mom_change_pct, rolling_3m_avg plus a one-paragraph insight.

8. Validate and Export

   Cross-check the grand total versus the sum of monthly totals and ask Julius to show the exact steps or query used; then download the chart as PNG, export the tables as CSV, and save the chat for future follow-ups.

Note: Always clean your dataset before uploading. Simple fixes like consistent column names and correct date formats prevent errors later.

Top AI & Tech News

• AI-driven 2025 stock gains added about $500B to U.S. tech leaders’ wealth and boosted the top 10 founders’ combined fortunes by roughly $1.9–$2.5T year‑over‑year

• Yann LeCun announces VL-JEPA, the latest non-generative vision-language model using joint embedding predictive architecture

• OpenAI's ChatGPT still dominates with 68.0% of AI traffic but OpenAI's overall traffic share has fallen about 20% this year

• Recent study by 30+ researchers from Cornell, MIT, Stanford, and Cambridge: LLMs do well on standard scientific benchmarks but falter on realistic, project-level research tasks

AI Art Spotlight

Create Your Own Photo Like This Using the Prompt Below:

A hyper-detailed geometric papercraft sculpture of the uploaded person, created from layered colored cardstock with precise angular folds and clean-cut edges. Facial features are stylized as polygonal planes, with subtle shading produced naturally by paper layering. Hair, beard, and clothing are built from stacked strips and folded shapes to create depth and realism. The figure is displayed against a clean, neutral background with soft, diffused lighting that highlights the craftsmanship and texture. Sharp macro focus; studio-photography aesthetic; 1080×1080.

Model: Nano Banana Pro

Prompt of the Day

Ultrathink Prompt That Turns Claude Into a Powerhouse

This prompt rewires how Claude approaches any problem by forcing it to slow down, question assumptions, plan before acting, and prioritize elegance over speed. Instead of giving surface level answers or rushed solutions, the prompt guides Claude to think like a designer, an engineer, and a craftsman at the same time.

Top AI Video Tutorial

How to Get the Most Out of Vibe Coding by YCombinator

AI Course of the Day

Nvidia's NeMo Agent Toolkit: Making Agents Reliable

This course teaches building production-ready conversational and task agents with NVIDIA’s open-source NeMo Agent Toolkit, focusing on configuration-driven workflows for easier design, testing, and iteration. You’ll learn to compose multi-agent systems, integrate agents from any framework, and deploy them with safeguards like authentication, rate limiting, and scalable interfaces.