WebGPT: Browser-assisted question-answering with human feedback

I think large frozen LMs are going to become the "operating system" of tomorrow. They are almost universal task solvers, with the right prompt or verbalizer. The trend now is to make them multi language, multi modal, multi task, multi memory ([1] and this paper), and scale computation with the problem complexity instead of keeping it fixed.

After that the next step I think is to make LMs that map arguments to functions from unstructured input, externally run the function and use the result in the next token predictions. LMs could write functions and execute them as regular code. They could learn to generate task descriptions, code, constraints and tests, and then run them in the loop to get experimental feedback. Regular code could call the neural net, and the neural net could call regular code. This would bring neural nets closer to symbolic AI.

So I see LMs as developers, iterating towards solutions. They have access to efficient simulation and compilers to augment the neural part.

[1] "Improving language models by retrieving from trillions of tokens" https://arxiv.org/pdf/2112.04426.pdf

[2] "Language Models as Zero-Shot Planners: Extracting Actionable Knowledge for Embodied Agents" https://arxiv.org/pdf/2201.07207.pdf

[3] "Pretrained Transformers As Universal Computation Engines" https://arxiv.org/pdf/2103.05247.pdf

I was interested to be able to play around with it myself, but I could only find some predefined samples (https://openaipublic.blob.core.windows.net/webgpt-answer-vie...) they say are not cherry-picked. Anyone know if there is a interactive demo somewhere?

Next wave of AI training seems to be through human imitation. To create long form unambiguous responses openAI is training a machine to follow human patterns and conducting and summarising a Google Search. Maybe before Google improves its search results, OpenAI can make it easy to scan results and plug the gap in Google's ranking algorithms