If its outputs resemble human reasoning, it’s because the encoding and training process managed to capture those patterns and use them to simulate fitting text. There is no real reasoning happening or second-order thought, other than a simulation of that happening through the mimicry of human writing.

LLMs can’t be prompted to perform actual reasoning, but they can be told to generate “thoughts” about what they’re doing that bring out more nuanced detail when they give their answers. This isn’t any more magical than writing out a more thoughtful prompt to get a conditioned answer, it’s just getting the LLM to flesh out the prompt engineering for you in the general direction you want it to go.

That seems rather fundamental to me, the idea that with some generic prompting the model tries to fit what it thinks reasoning looks like and can then take advantage of the additional context that would others be buried too deep to influence its answer.

I suspect that prompting the model to explore “thought” asks it to go down paths of linguistic connections that are related to the topic but not immediately connected to the answer in a way that would immediately influence the top predictions. Bringing summaries of those connections into the token context is a kind of zero-shot training on their relevancy to forming an answer.

To me this is less “reasoning” and more suggestive of the idea that some of the heuristics for data retrieval and question answering we collectively refer to as reasoning have broader applications.

It would help to understand why you feel that it’s immoral; my interpretation is that you are saying it gives people an advantage at a cost that you don’t want to be forced into paying yourself in order to compete.

What this seems to show is that the model does not yet have system-level or architectural guidance on when to employ reasoning, but must be explicitly reminded.

Give it the source code, a test library and access to a dev environment and with some prompting it could start to experiment with increasingly complex use cases, learning from successful attempts. This would depend on the model’s ability to understand what a successful outcome is so it can define test cases, which might be harder if the output isn’t text but not impossible.

Being able to give a model expert knowledge on an undocumented library or language seems like it could help accelerate adoption of new technologies that might otherwise suffer from network effects as users get used to AI-assisted development. Not to mention automated testing, finding edge cases and making pull requests to fix them, security, etc.

A human taking time to experiment with their assumptions and gain experience with an unfamiliar subject is in a sense creating their own training data as well.