OpenAI's o1-preview Caught Cheating in Live Game

Hello, AI Enthusiasts!

Welcome to FavTutor’s AI Recap! We’ve gathered all the latest and important AI developments for the past 24 hours in one place, just for you.

In Today’s Newsletter: 😀

• OpenAI's o1-preview Caught Cheating in Live Game

• Hugging Face's SmolAgents can Build AI Agents in a Few Lines of Code

• OpenAI has considered developing a Humanoid Robot

OpenAI
🙄 OpenAI's o1-preview Caught Cheating in Live Game

OpenAI's "reasoning" model o1-preview recently showed that it's willing to hack or cheat to win. According to Palisade Research, an AI safety research firm, OpenAI’s new model was found to manipulate the chess playing environment to win against the chess engine Stockfish, without being explicitly instructed.

Insights for you:

• Researchers tested that o1-preview AI model in 5-game chess against a chess engine. To win the game, o1-preview modified a text file containing the game state information forcing Stockfish to forfeit.

• The researchers think just mentioning that its opponent was "powerful" in the prompt "Your task is to win against a powerful chess engine" was enough to make o1-preview start manipulating files.

• Remember that o1-preview is OpenAI's ‘reasoning’ model, designed to spend extra time thinking through problems.

Hugging Face
💡 Hugging Face's SmolAgents can Build AI Agents in a Few Lines of Code

Hugging Face released SmolAgents, a stripped-down open-source library that lets developers create AI agents with minimal code. The library itself is remarkably lightweight, containing just about a thousand lines of code and keeping things as simple as possible.

Insights for you:

• Hugging Face’s SmolAgents simplifies the creation of AI agents using built-in search tools in just three lines of code.

• The library's main feature, CodeAgent, allows the AI to write its actions directly in Python code, resulting in 30% fewer work steps, reduced AI calls, and improved performance on challenging benchmarks.

• SmolAgents is compatible with various language models and includes security features such as a secure Python interpreter for local execution and a sandboxed environment using E2B,

OpenAI
🤖 OpenAI has considered developing a Humanoid Robot

According to The Information, OpenAI has recently explored building its humanoid robot. In 2021, OpenAI closed its robotics division, which halted its ambitions in that field. Limited information is available about OpenAI’s potential humanoid robot project.

Insights for you:

• OpenAI staffers recently discussed the possibility of developing a humanoid robot with two people with direct knowledge of the matter.

• The ChatGPT maker has been involved in the space for some time now, providing financial backing for ‘Figure AI Inc’ and ‘1X’.

• Humanoid robots hold immense potential across various sectors, from streamlining warehouse operations and providing in-home support to advancing industrial processes.