3 ms·
You are conflating Illya's belief in the transformer architecture (with tweaks/compute optimizations) being sufficient for AGI with that of LLMs being sufficien
by Chamix 3y ago
You are conflating Illya's belief in the transformer architecture (with tweaks/compute optimizations) being sufficient for AGI with that of LLMs being sufficient to express human-like intelligence. Multi-modality (and the swath of new training data it unlocks) is clearly a key component of creating AGI if we watch Sutskever's interviews from the past year.
- cratermoon 3y agoYes, I read "Attention Is All You Need", and I understand that the multi-head generative pre-trained model talks about "tokens" rather than language specifically. So in this case, I'm using "LLM" as shorthand for what OpenAI is doing with GPTs. I'll try to be more precise in the future. That still leaves disagreement between Altman and Sutskever over whether or not the current technology will lead to AGI or "superintelligence", with Altman clearly turning towards skepticism.
- Chamix 3y agoFair enough, shame "Large Tokenized Models" etc never entered the nomenclature.
- cratermoon 3y agoSome terms I've seen used for the technology: Big-Data Statistical Models Stochastic Parrots or parrot-tech plausible sentence generators glorified auto-complete cleverbot "a Blurry JPEG of the Web" <https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web https://www.newyorker.com/tech/annals-of-technology/chatgpt-...> and just plain ol' "machine learning"
- limpanz 3y agoDo you have a link to one of these talks?