AGI Timeline Debate Collage

AI won't kill us it's like worrying about overpopulation on Mars "AGI is when AGI develops the next generation of AGI." Google might call this sharing proprietary property. I call it sharing a discussion that I had with one of my coworkers. Game Over. Scale is essential to AI the 2x2 matrix are "AGI" that really matters is short-term -> long-term (until the ramp really starts) and stops takeoff --> fast takeoff Scaling up deep learning won't lead to "AGI" because deep learning on its own is missing key properties required for general intelligence. There is no such thing as AGI Sad to end up in the goalpost "movers" quadrant (I'd put myself somewhere in the middle) my position is still "50% of clear signs of life in AGI by 2030 People seem to be quite efficient in destroying themselves, perhaps speeding up AGI could prevent that. If AI tiling the universe with sensors reward tensors is real I find it unlikely that societies would be the first to discover such a nightmare I am willing to bet you $100,000 that you are wrong about the timing of AGI It's interesting for me to be placed between Gary Marcus and Yann LeCun in so far as... I wouldn't say, my positions are much closer to Yann's probably than Gary's. AGI is a feeling. Like love. Stop trying to define it. I totally think it's possible to have a good AGI! but if we have a bad AGI it will kill us. That's all I'm saying AGI in the next decade or two I identify as like ten pixels under room on this graph apparently they need to build "AGI", whatever that is There seems to be evidence that members of the AI community are converging on recent advances in ML and forming unreasonable expectations of transformative AI. I was extremely surprised by OpenAI for working with just vanilla PPO, rather than requiring any advances in hierarchical RL. I made one massive update then also basically adopted the bio approach timelines rather than interpretations that were released. I don't really think AGI is gonna be bad, I think it's more like 50/50 actually my more serious attempt at an AGI prediction was a probability of 50% by 2028 and 90% by 2040. roughly held since 1999 and have repeated in many places scaling current methods 'gives us a bunch of parameters' (solve this by 2040) - solves everything that matters except alignment. I am skeptical, and willing to bet <$1bn in customer revenue from [GPT-3 spinoffs, or companies built around auxiliary services] by 2025 AI might kill you and everyone you know. I think it will be years between between when you have economically impactful AI systems and the singularity 2029 feels like a pivotal year. I'd be surprised if we don't have AGI by then. was hoping that as de facto alignment memelord I would make it into more alignment memes It is May 2020. It has always been May 2020 I self identify as a longtermist nerd whom AGI is good I have perhaps the same stance since that was the most important thing for AI Safety is just enabling around the edges, that is the most likely to be their written for role of preference... I don't think I'm in the 'AGI soon camp'. I have a very wide probability distribution and no just scale a slew... another 1000x to go superhuman or do we actually still need to move on to more fundamentally New Secret Causes debate When DALL-E came out I was surprised by the performance. When GLIDE came so soon after, I was surprised by the rate of progress. Yet now DALL-E 2 is out, I'm somewhat still surprised! This is a very lethal problem, it has to be solved at a minimum alignment difficulty level instead of various easier modes that some people dream about I feel I should be where it is given (more, marginally) less doomer than more mainstream AI safety maximalist

A large collage of portraits and cartoon figures paired with quotations about artificial general intelligence, AI timelines, alignment, and existential risk. The joke is the sheer contradiction: researchers and commentators confidently make mutually incompatible predictions about whether AGI is imminent, impossible, beneficial, or deadly, so the image is sent as a reaction to debates about AI forecasting and AI safety.

• @profunctor_io / 8022 2022-06-16