Chess-GPT’s Internal World Model

• Chess-GPT, trained on 5 million chess games, learns to play chess with an Elo rating of ~1300 without being explicitly taught the game's rules or board state.

• The 50 million parameter model, smaller than GPT-3.5, demonstrates the ability to understand complex game dynamics like castling, en passant, and checkmate.

• Training involved synthetic datasets of chess games played by Stockfish at various Elo ratings and 16 million games from Lichess.

• The model proved capable of playing unique games not found in its training dataset, countering the assumption that LLMs merely memorize internet data.

• Linear probes were used to verify the model’s understanding of the chessboard's state, with high accuracy in predicting the locations of specific pieces.

• Probes for latent variables like player skill levels showed the model could distinguish between different Elo ratings, reflecting a deeper understanding of chess beyond mere move prediction.

Sam Altman wants up to $7 trillion for AI chips. The natural resources required would be ‘mind boggling’

• Sam Altman, CEO of OpenAI, reportedly seeks up to $7 trillion to expand global chip capacity, aiming to advance AI model capabilities.

• The project's environmental impact is significant, with concerns over the use of water, rare earth minerals, and the overall carbon footprint.

• AI models like OpenAI's GPT-3 and Meta's Llama 2 have already contributed to increased water consumption and environmental strain.

• The tech industry faces GPU shortages, with companies like Meta investing heavily in high-performance computing and custom silicon for AI.

• Nvidia, a major GPU manufacturer, has been criticized for a lack of transparency regarding the environmental impact of their products.

• The project is viewed skeptically by some experts, who doubt its feasibility and criticize the lack of efficiency in current AI development methods.

"AI native" Gen Zers are comfortable on the cutting edge

• Gen Z, dubbed "AI natives," are eagerly embracing generative AI technologies like ChatGPT and DALL-E for educational and career advantages.

• Surveys indicate a strong interest among Gen Z in learning AI skills, with many planning to use generative AI in their future careers.

• Unlike older generations, Gen Z exhibits less fear and more enthusiasm about AI's role in job markets and sees it as crucial for career advancement.

• McKinsey Global Institute reports that AI is accelerating job automation, but Gen Z's early exposure to AI might give them a competitive edge.

• Most U.S. adults are wary of AI, but younger generations are more optimistic about its benefits and are investing in AI-related skills and education.

• The rise of generative AI has led to an increase in college students and graduates adding AI-related skills and terms to their professional profiles.

MARC HOAG LAW.

Future Perfect is brought to you by my law practice serving tech startups and providing solutions at the intersection of AI and copyright law. Come say hi at MarcHoagLaw.com or click here to watch my lectures about AI and copyright law or here for my Axiomic Legal Toolkit where you can self-educate on AI and tech startups (and more!)

🎉 THAT’S ALL FOR TODAY!

See you next time! 👋

🙏 PLEASE SHARE this post if you liked it!

Thanks for reading!

-Marc 👋

Looking for past newsletters? You can find them all here.