4 ms·
The future of AI isn‘t training data, it‘s logic and adversarial training on reasoning by first principles. You already see the first flowers of that in the Ins
by endymi0n 4y ago
The future of AI isn‘t training data, it‘s logic and adversarial training on reasoning by first principles. You already see the first flowers of that in the InstructGPT paper and we‘re about to see much more. If you port the idea of GANs to the LLM world, you‘re about to get a hardened model that can introspect both facts and subtext, including agendas.
- ChatGTP 4y agoEven though InstructGPT still makes simple mistakes, our results show that fine-tuning with human feedback is a promising direction for aligning language models with human intent. That will be fun when that thing is hooked up to Clippy