The AI Dilemma

• 50% of researcher say there is a 10% chance of humans going extinct from our inability to control Ai
• rubber band effect- expand our minds then looking somewhere else everything snaps back
• 3 rules of tech
•     when you invent new tech, you uncover a new class of responsibilities
•   if the tech confers powers, it forms a race
•     if you don't coordinate the race it forms a tragedy 
• social media is humanities first contact with AI ( race to engagement, we lost this)
• second contact in 2023 with Ai, (race to intimacy) 
• maximize management
• did we fix this misalignment?
•  no we didnt
• GPT-3
• -lots of AI postives
• -negatives and weird bias, opinions
• 2017 new engine was created
• AI used to be separate fields- Speech recognition, robotics, computer vision, speech synthesis, etc.
• Gollem-Class Ai- they suddenly have emergent capabilities
• The AI only sees what your thinking
• TikTok filters (cat-fishing)
• total decoding and synthesizing of reality
• what nukes are to the physical word, Ai is to the virtual and symbolic world
• 2024 will be the last human election
• Gollem AI have emergent capabilities their programmers didnt program
• programmers say they dont know why, how, what happens for them to be able to go through this emergent process.
•     new capabilities suddenly emerge    
•      Ai develops theory of mind
•     only noticed that it was growing its mind a month ago
•     RLHF- advanced clipper training 
•     Gollems silently taught themselves research grade chemistry
• Ai can make themselves stronger
•     how do you feed your gollem if you run out of data
• use Ai to feed itself, youtube, podcasts, radio
• prediction: Ai will reach >50% in four years
• reality: Ai reached >50% in one year
• Ai is beating tests as fast as they are made
• getting increasingly hard for programmers/directors
• democratization is dangerous 
• Racing to deploy Gollem AIs into world infrastructure as fast as possible
• armies of gollem Ais pointed at our brains, strip-mining us of everything that isnt protected by 19th century law
• a lot of safety researchers by venue
• at least AI has substantial research
• the people who are creating AI feel that is not moving at a safe pace