Remix.run Logo
jenniferhooley 2 hours ago

I felt like it was about 10X better at "pretty" but straightforward 1 shot'ish type tasks. Not so different for complex and specific tasks in real code-bases.

Why do you say it was a lot better, what type of tasks were you testing it on?

JeremyNT 2 hours ago | parent [-]

> I felt like it was about 10X better at "pretty" but straightforward 1 shot'ish type tasks. Not so different for complex and specific tasks in real code-bases.

What metric are you using for "better" here? If I've got a straightforward task GPT 5.5 is going to 1shot it anyway.

frontierkodiak an hour ago | parent | next [-]

Maybe not 10x; but it's fantastically talented at intuiting intent, reconstructing contexts, and making aligned judgement calls. You could throw Fable utter garbage and get great results. Fable felt like it was modeling me whereas gpt-5.5 is still very much is riding your prompt, your inputs. I have bit of humility here as this is basically how I felt about 5.4->5.5; but 5.4 was very much a scalpel-very spiky weird intelligence. 5.5 sits somewhere in-between, but still spiky and verbose- code-maxxed; not a great orator, not a good proactive "here's the skip-connection you probably should be thinking about but don't seem to be weighing" in the way that Fable is. Fable is crisp.

jenniferhooley 2 hours ago | parent | prev [-]

Let's say we're prototyping an interactive tree (this is totally made up, but you get the idea):

Take this data input and convert it to a Sugiyama-style tree with hand-drawn feeling lines connecting the nodes. We need the ability to activate a random subset of nodes with a paint splash. The whole thing should feel organic, incorporating small motion effects. As the nodes are activated, the edges should look like a hand-drawn painting effect drawing toward the node, and then SPLASH onto the end node as it changes from black to deep red. The background should utilize a muted paper color, and we must adhere to this color palette for all elements (PALETTE).

...then back-and-forth 10 prompts or so to get the prototype I was looking for. Comparing these types of things between Fable and Opus, something like this would be quite a bit less glitchy, prettier, and closer to the quick prototyping I needed than what I got with Opus 4.8.

Now, when I went deep into a complex codebase to fix a small issue or optimization that spanned many files and was fairly unique from anything in the training data, I didn't really see any noticeable difference between Fable and Opus 4.8.