First impressions of Fable 5.1

Fable 5.1, the latest model from an unnamed developer, performs better on the dissertation benchmark and feels less slow than previous versions, according to a first-impressions review by Nate Meyvis. The reviewer notes that Fable 5.1's prose is more contextually appropriate and that it excels at tasks like deduplicating software issues, though he still advises seeking human feedback when available.

- It is very good. - It does even better on the dissertation benchmark https://www.natemeyvis.com/the-dissertation-benchmark/ than previous Fable https://www.natemeyvis.com/humanistic-reactions-to-fable/ . - It feels less slow than previous Fable, and the speed doesn't bother me. This isn't because speed isn't important it totally is https://www.natemeyvis.com/try-the-very-fast-models/ but because it is so important that I tend to have a Gemini 3.7 Flash window or two open. As long as Fable 5.1 is faster than some hard-to-define threshold, and it is, small differences don't matter much, because I'm using the best Gemini Flash models, which are truly wonderful, for low-latency work. - The prose feels, across contexts, more contextually appropriate. It describes bug diagnoses more naturally and intelligibly. Its philosophical communication is compact but thorough, intelligible, and somehow more philosophical . I'm still not confusing its writing with human writing, but that's not what I want: rather, reading it feels more comfortable in the way that improving the lighting in a room or, for this analogy, improving the lighting in many different rooms makes one more comfortable. - Over a very small sample, I think I'm finding it less intellectually fatiguing than using early versions of previous frontier models. Again, the lighting metaphor feels appropriate. Of course, my sample size is extremely small, placebo effects are real, and intellectual fatigue is impossible to measure and very hard even to define. - I still don't worry about AI making human intelligence obsolete, whether in philosophy, programming, or anything else. If you can get dissertation comments from considerate humans in their expertise, do not pass up those comments in favor of Fable 5.1. You can expect them to be better than what Fable 5.1 can give you, and there are many other reasons to get comments from humans instead of machines. But if you are serious about improving your work and your abilities, and if you have access to Fable 5.1, I think it's a serious mistake not to be using it. You can make all the arguments you want about how machines can't possibly be useful for Real Philosophy, but as far as I can tell, those arguments simply cannot survive contact with the actual experience of working with the best available models. Don't forget that it's also better at tracking down references, summarizing papers you wouldn't otherwise have time even to look at, making cross-disciplinary issues intelligible, and so on. - Another of my loose quasi-benchmarks is asking a model to i pull all my open issues for a software project; ii deduplicate them where appropriate; iii see if fixes for any can be verified in prod and therefore closed ; and iv see if any can be closed as obsolete. This is harder for AI than you might expect it to be because there are quadratically many pairwise relationships between issues; issues tend to have a lot of text, especially if AI wrote them; and this analysis can require exploring large parts of the codebase . Fable 5.1 is crushing all previous models on this task.