Summary, my other summaries are here
Marc Andreessen, VC at a16z ($35bn fund, only Sequoia has more), web3 maximalist + spawned the internet with Netscape browser (bought by AOL for $4bn in 1999), on Meta board, interview #368 for 3 hours “AI will save the world” by Lex Fridman (AI scientist in USA from RF/Ukraine):
LLM/language models
- Everything becomes interactive, even if we sometimes ask for URLs, the Internet began with directories, then search for 20 years, and AI gives new ways of interacting with the knowledge of whether Google will be in 5-10 years. They have been building graphs / issuing answers without clicks for 15 years, and then LLM came
- People want uncensored LLMs, but who will release them, hackers? New LLMs will use dialogues with themselves, and this is a $1 trillion question: will synthetic data LLM improve (maybe signal is already there), but there is an ocean of possible dialogues that cannot be studied in public (like people discussing economics textbooks on Twitter, but we will speed it up 1000x)
- LLM glitches? This is when we don’t like it, and when we don’t like it, we call it creativity =) Le Cun believes that glitches cannot be removed, but the industry believes that they will solve
- Bias? I thought the idea of Wikipedia was nonsense, but it worked in developing a neutral tone (for thousands of years people thought “here it is, the truth”, but it “floats” all the time ..)
Golden rush
- More trillion-dollar questions: Will the small models beat the big ones? Will there be 2-3 god-level models for the whole world? There is a debate on whether to leave open source AI legal/regulatory capture from lobbyists. Will tech giants or startups win (yes, IBM invented databases, but Oracle built Larry Ellison; why did OpenAI make GPT, not DeepMind/Google)? Battle of the Giants (many GPUs + sophistiction) vs. startups trying radical ideas – both approaches are successful: both Apple perfectionism and incremental lean hackers as startups
- The role of founders was overly romanticized, nervously + cool founders often with experience in their industry, or studied the idea comprehensively for a year
Internet
- In 1989 at the university I saw a fast Internet + supercomputers Cray = gave insights into what was coming (and everyone thought that PCs, emails were for nerdy progers) and now there are 5 billion people online. He remembers a lot about Apple I, TCP IP drivers in Win 3.1, the role of JPEG in the era of modems, HTML, JavaScript (written over the summer), SSL (ciphers were military technology) + “autistic” programming languages
- Browser = a single window to the world on a PC, anything behind it + evolves over time. Apps = on phones, but even there the temptation is to sculpt superapps, like Musk on Twitter or the Chinese
- It looks like we are going to open source everything, not only AI, maybe productivity boost – it’s finally, 1000x to a separate encoder. Software is the modern philosopher’s stone
AI will save the world
- Wrote an essay: fear of AGI – we are just Christians, we are waiting for the Apocalypse + we generate cults + we radicalize ourselves. These are religious, not testable hypotheses. Von Neumann believed that it was necessary to immediately hit the USSR with a nuclear bomb, because the 3rd World War was inevitable. But in fact, nuclear weapons removed the risks of the 3rd World War. And even where we have predictive models – well, you remember the spectrum of terrible forecasts about the corona and policy makers overly relied on these [pseudo] scientists to model 8 billion people ..
A - Intelligence is the collection, processing, synthesis of information, and to what you can’t apply – everything improves (smart people in all areas of life have better outcomes – it seems that these are genes, that sadness), but then AI appears on the stage – for the benefit of everyone – everyone will increase IQ
- Generative AI / fakes? You put another AI – protect your child =)
- AI is difficult to regulate, especially with open source / people who make breakthroughs are everywhere / the GPT model is already being trained on a powerful PC, etc.
- Fear of inequality from AI = neo-Marxism, in reality, automation has incredibly increased per capita GDP, the secret of all CEOs is that tech giants want to sell AI to everyone on Earth
- Retraining due to new jobs hurts, but we have been living like this for 300 years, and Luddites have been and will be
China? Their public AI plan for themselves + the world: surveillance, social score = 100% authoritarianism + Digital Silk Road
Charity? Larry Page wants all the $ to give the Mask
Books? Reads about history, US, politics, Lenin, French revolution, Rome, Greece
Meaning of life? Satisfaction/being useful, but I’m for imbalanced life/productivity roll