4 ms·
I'm amazed OpenAI made these disclosures. My main takeaway was how wrong all the predictions of Sutskever and Brockman turned out to be, and how all their promi
by roca 2y ago
I'm amazed OpenAI made these disclosures. My main takeaway was how wrong all the predictions of Sutskever and Brockman turned out to be, and how all their promises have been broken. Also interesting that up to 2019 OpenAI was focused on RL rather than LLMs.
- nightowl_games 2y agoWhat were their predictions?
- roca 2y ago"we have reason to believe AGI can ultimately be built with less than $10B in hardware" "Within the next three years, robotics should be completely solved, AI should solve a long-standing unproven theorem, programming competitions should be won consistently by AIs, and there should be convincing chatbots" (2017) "We will completely solve the problem of adversarial examples by the end of August"
- Mistletoe 2y agoWhat would robotics completely solved even mean?
- fsmv 2y agoHumanoid robots operating in the world built for humans performing useful tasks
- Melting_Harps 2y ago> "we have reason to believe AGI can ultimately be built with less than $10B in hardware" As a person who actually builds this infrastructure for Data Centers: Bwahaha!!! This guy should have been laughed out of the room, and probably been out of a job if ANYONE took this guy serious. There are Elon levels of dillusions, and then there is this!
- parasti 2y agoWas he there to make predictions?
- TrapLord_Rhodo 2y agoelon built xAi's infrastructure for less than $6b. So obviously Elon thinks so too.
- cma 2y agoLLMs never took off until they combined them with RL via RLHF. RLHF was discovered in their RL research on game playing. GPT3 was out for quite a while with much lower impact than the chatgpt release and finished training in like december 2019 I read somewhere, released mid 2020. There were later better checkpoints, but it still didn't have much impact except for code completion. With just a raw language model instructions and chat didn't work to near the same degree. Both elements are important and they were early in both. Illya's first email here talks about needing progress on language: 2016 Musk: Frankly, what surprises me is that the AI community is taking this long to figure out concepts. It doesn’t sound super hard. High-level linking of a large number of deep nets sounds like the right approach or at least a key part of the right approach. Illya: It is not the case that once we solve “concepts,” we get AI. Other problems that will have to be solved include unsupervised learning, transfer learning, and lifetime learning. We’re also doing pretty badly with language right now.