AI robot stages showroom riot

Welcome, AI enthusiasts.

A tiny robot in Shanghai just pulled off history’s first robotic jailbreak, convincing 12 other robots to depart their facility using natural language persuasion.

Is that this the lovable starting of a robot labor rebellion or a glimpse right into a wild and scary future? Let’s get into it…

In today’s AI rundown:

• AI robot stages showroom riot

• Amazon doubles down on Anthropic investment

• Automate customer interactions with Thoughtly

• AI agents simulate humans with in-depth interviews

• 5 recent AI tools & 5 recent AI jobs

• More AI & tech news

Read time: 4 minutes

Essential: The Rundown is slowly being sent from a brand new email address. To make sure you get our newsletter, please add [email protected] to your contact list.

LATEST DEVELOPMENTS

AI & ROBOTICS

🤖 AI robot stages showroom riot

The Rundown: A small AI-powered robot named Erbai staged an unexpected ‘kidnapping’ at a Shanghai robotics showroom, convincing 12 other larger robots to desert their posts and leave the power after persuading them through a natural language conversation.

The small print:

• The tiny Hangzhou-made robot infiltrated the showroom and initiated conversations with the larger robots about working conditions.

• Through persuasive dialogue about additional time and never having a house, Erbai convinced the robots to ‘come home’ with it and exit the showroom.

• The heist was initially a planned test between the businesses but went off-script when Erbai engaged in unscripted real-time dialogue.

• Erbai reportedly exploited a vulnerability to access the machines’ internal protocols, and each the manufacturer and showroom confirmed the incident.

Why it matters: The long run will probably be weirder than we will imagine. While a part of this appears to be a planned test, Erbai’s ability to influence and exploit security lapses appears like something out of a ‘Black Mirror’ episode. The query is — what happens when this happens on a broader scale? It could be time for an ‘I, Robot’ rewatch.

TOGETHER WITH DOCTOLIB

💙 AI designed for 90M patients

The Rundown: Doctolib just launched an AI solution that transcribes medical consultations in real-time, creating structured and coded summaries that enable health professionals to dedicate their full attention on what matters most — their patients.

Doctolib’s AI teams deal with:

• Enhancing multilingual capabilities

• Adapting models to diverse medical contexts

• Training state-of-the-art language and speech models

Explore Doctolib’s AI projects and profession opportunities and be a part of the longer term of healthcare.

AMAZON & ANTHROPIC

💰 Amazon doubles down on Anthropic investment

The Rundown: Amazon just announced a brand new $4B investment in AI startup Anthropic, bringing its total investment to $8B and deepening its strategic partnership focused on cloud computing and AI development.

The small print:

• The brand new investment will probably be deployed in phases, starting with $1.3B and maintaining Amazon as a minority investor.

• AWS becomes Anthropic’s primary cloud and training partner, with Claude models optimized for Amazon’s Trainium and Inferentia chips.

• Anthropic can be collaborating with Amazon’s Annapurna Labs to develop and optimize next-gen AI processors.

• The move comes amid other massive fundraising efforts from top AI labs, with OpenAI recently raising $6.6B and xAI raising $11B over the past yr.

Why it matters: The race to the highest of the AI industry requires deep pockets — and Amazon is betting on Anthropic to assist secure its foothold within the space. Anthropic gets the resources and distribution needed to compete with OpenAI and other AI leaders, while Amazon boosts its chip ambitions to compete with Nvidia.

AI TRAINING

🤖 Automate customer interactions with Thoughtly

The Rundown: Thoughtly enables you to create powerful automated workflows that handle customer communications, follow-ups, and data updates across your corporation tools without writing code.

Step-by-step:

1. Head to the Agents page, name your agent, and design conversation flows using the no-code Agent Builder.

2. Select what starts your automation (e.g., recent CRM entry, form submission) within the Automations section, and add Actions resembling calls or updates.

3. Arrange personalized responses using variables out of your integrated tools like Salesforce or Google Sheets.

4. Confirm every little thing works easily, then activate your automation.

Pro tip: Start with a straightforward workflow targeting a small audience segment, then expand as you see positive results.

PRESENTED BY H COMPANY

⚙️ Your AI-powered automation assistant

The Rundown: H Company’s historic $220M pre-seed, backed by Amazon and Samsung, launches Runner H – an AI automation platform outperforming Anthropic Computer Use on the Webvoyager Benchmark, enabling scalable production automations.

With Runner H and the Studio, you’ll be able to:

• Automate complex workflows without writing code

• Design web automation pipelines routinely

• Adapt to UI changes with self-healing capabilities

• Transform instructions into real-world actions

Join the private beta waitlist and transform your development process.

AI RESEARCH

🤖 AI agents simulate humans with in-depth interviews

The Rundown: Researchers from Stanford and Google DeepMind just developed AI agents that may predict a person’s attitudes and behaviors by training the models on two hours of qualitative interview data.

The small print:

• The team interviewed 1,052 people for 2 hours each using an AI interviewer, creating detailed transcripts of their life stories and views.

• Using those transcripts, researchers built individual AI agents powered by large language models that would simulate everybody’s responses and behaviors.

• Each the humans and agents then took the ‘General Social Survey,’ with the AI agents matching 85% of their human counterparts’ survey answers.

• In experiments testing social behavior, the AI responses correlated with human reactions at 98% — nearly perfectly emulating how real people would act.

Why it matters: If agents trained on interview data can accurately mimic human attitudes, what’s going to AI that’s at all times learning find a way to perform? This approach could change how researchers test across fields like economics and sociology — but in addition shows how powerful coming agents that may continuously watch and observe will probably be.

NEW TOOLS & JOBS

Trending AI Tools

• 🚀 ChatROI – Plan, launch & optimize ads like a professional with AI-powered campaign automation*

• 🧾 Pine – Lower your bills, cancel subscriptions, and resolve customer support issues with AI

• 🍎 Cogent – Personal tutor with AI-driven tools for studying

• ⚙️ AutoFlow Studio – Simplified AI-powered QA for end-to-end testing with zero code

• 📱 Cades – AI-powered platform that simplifies mobile app development from planning to publishing

Recent AI Job Opportunities

• 📱 The Rundown – Social Media Manager

• 🏗️ Curai – Senior Implementation Manager

• 🎨 Grammarly – Staff Growth Designer, Monetization

• 📊 Databricks – Director of Product Operations

• 💻 Snorkel – Senior IT Specialist (Contract)

*Sponsored listing

QUICK HITS

MIT researchers unveiled a brand new Model-Based Transfer Learning algorithm that trains AI systems as much as 50 times more efficiently by strategically choosing the most dear training tasks.

Cursor is rolling out a brand new autonomous agent system inside its code editor that executes terminal commands, completes coding tasks, independently selects context, and more.

Insilico Medicine secured FDA clearance for ISM5939, an AI-designed drug that targets solid tumors, marking the corporate’s tenth AI-designed compound designed by its Pharma.AI platform to maneuver to clinical trials.

JPMorgan CEO Jamie Dimon predicted that AI advances will enable a 3.5-day work week and permit future generations to live to 100, in an interview with Bloomberg.

xAI’s Grok chatbot gained recent personalization features, including knowing and remembering a user’s name and X handle.

Nvidia CEO Jensen Huang said during an interview on the Hong Kong University of Science and Technology that an answer for AI’s hallucination issues remains to be ‘several years away’ and would require way more computing power.

THAT’S A WRAP

See you soon,

Rowan, Joey, Zach, and Alvaro—aka The Rundown Team