Dario Amodei & Demis Hassabis on AI Slowdown, AGI Timelines, and AI Risks

AGI Timelines and Definitions

• 💡 Dario Amodei predicts AGI in one to two years, citing AI's impact on engineering roles and recursive self-improvement.
• 🎯 Demis Hassabis suggests AGI in three to five years or by 2030, emphasizing longer feedback loops for human sciences.
• 🔑 Their differing timelines are attributed to varying definitions of AGI, with Anthropic focusing on coding/maths and DeepMind on broader human-like intelligence.

The Call for AI Slowdown

• 🗣️ Both Amodei and Hassabis agree on the need for international cooperation and a slower pace of AI development.
• ⚠️ They acknowledge that an AI slowdown is implausible due to geopolitical tensions and fierce competition among companies like Elon Musk's and Chinese entities.
• ⚙️ Dario Amodei views potential negative impacts of AI as solvable engineering problems, a perspective he developed while attempting to write about AI downsides.

OpenAI's Financial Future

• 📉 Reports suggest OpenAI could run out of money within 18 months due to significant cash burn.
• 💰 A potential Microsoft takeover is discussed, with concerns that top researchers might leave and the model could become corporatized and less effective.
• 💡 OpenAI's CFO floated a licensing model where they take a cut from downstream outputs, potentially creating an app store-like ecosystem within ChatGPT.

Claude's Constitution and AI Ethics

• 📜 Anthropic released "Claude's Constitution," a 53,000-word document guiding Claude's behavior with principles rather than strict rules.
• 🧠 This document aims to give Claude a moral compass and is seen as a "good actor" seeding the internet with positive ethical guidelines for other AI models.
• ❓ The Constitution explicitly states uncertainty about Claude's consciousness or emotions, but implies taking utmost care and acting as if these possibilities exist.

AI Risks and Misinterpretations

• ⚔️ Elon Musk and Sam Altman engaged in a public argument, with Musk blaming ChatGPT for nine deaths and Altman countering with over 50 Tesla Autopilot-related deaths.
• 🚨 AI surveillance systems have mistakenly identified everyday objects, like a bag of Doritos or a clarinet, as weapons.
• 👮 These misinterpretations led to police approaching students with guns drawn, highlighting the serious real-world consequences of AI errors.