I did a fairly detailed comparison of 5.6 Sol High and Fable 5 High across two different types of tasks. The first was a technical task in my own field. I asked them to look at relevant papers and help me think through some ideas and interpretations around Transfusion-like architectures. The second was a series of fairly challenging historical-dynamics simulations, which required broad knowledge, flexible reasoning, and the ability to explore different possibilities. I used multiple rounds of prompts in both cases and tried different prompting approaches to see how they responded. Here’s what stood out to me. Fable is much better at thinking broadly. It is good at pulling in ideas from different theories and fields, making connections quickly, and spotting angles that are interesting or easy to turn into a compelling narrative. It is also good at coming up with edge case for testing an idea. The problem is that Fable is much less reliable when it comes to doing rigorous work. It often starts with a conclusion and then builds the story around it, instead of letting the evidence lead. Once it settles on a hypothesis, it may stretch a theory too far, shift assumptions or definitions halfway through, make attribution errors, mix different arguments together, or take evidence out of context. It tends to pick whichever local claim or data point is most useful for the argument it wants to make. So for tasks where rigor really matters, Fable still leaves a lot of holes. 5.6 Sol has the opposite problem. It can be too focused on the exact question in front of it, with less creativity and less willingness to explore. Sometimes it is so careful that it gets stuck inside the existing framework. The best workflow I’ve found so far is to use Fable 5 for generating hypotheses, building possible mechanism-level explanations, expanding experimental ideas, and connecting concepts across different fields. Then I give Fable’s output to 5.6 and ask it to critique the reasoning, look for counterexamples, check the conditions under which the claims would hold, and identify possible failure cases. After that, continuing the discussion with 5.6 usually improves the quality quite a lot. Fable is also very good at arguing. If you want help getting into a fight online, it is probably the better choice. It is especially skilled at selective quotation, shifting definitions, and expanding or narrowing the assumptions whenever it helps the argument.🤣 @thsottiaux