morgan.rebrand

Trans 🏳️‍⚧️, 34, she/her, Demisexual , panromantic, polyamorous. Semi-feral, house-trained. Degree in Computer Science. Still illegal in this dimension.

[Tibo @thsottiaux]

@blueemi99 Forgot

[morgan.rebrand @MorganRebrand]

@theo What's the screen?

[morgan.rebrand @MorganRebrand]

@theo Same!!! 😭

[morgan.rebrand @MorganRebrand]

@pierbiwierbi Mine were extracted :(

[morgan.rebrand @MorganRebrand]

@ImLunaHey 100% the case. Even when it's 99% vibed, you see someone's engineering craft shine through (because they've taken the work seriously still)

[morgan.rebrand @MorganRebrand]

@ivanburazin Cal(endar)
Cal(culator)
Cal (Berkeley)
Cal (contact)

[morgan.rebrand @MorganRebrand]

@thsottiaux Rewind chat history!

[dax @thdxr]

idea guys think simply having the idea means they should be successful

programmers think simply being able to build the idea means they should be successful

[morgan.rebrand @MorganRebrand]

@elian_mcc @theo I use plan mode a lot to get all of the strengths from Fable without having to use its tokens on a lot of the grunt work of building. Fable writes really great plans for lesser models to follow.

[morgan.rebrand @MorganRebrand]

Yes! I'd have to fork it then.

One pattern I've found really useful is to load up Fable with context, then have it generate a plan, telling it to keep in mind something like Opus will be implementing it. Then, I can hand those plans off to something like kimi-k2.7-code and they usually cook.

It is really useful in situations where I have quota on a much higher tier model (Fable / Sol), but then can hand off to something that's more quota efficient, like gemini-3.6-flash, kimi, ollama, etc.

Granted, you run into the usual problems having something less-than-frontier work on an implementation, of needing multiple rounds of code review to get something merge-worthy. But given how generous a lot of their subscriptions are, that's okay, especially since its automated.

One other scenario I use plans for is when I am working with Fable, and need to thoroughly finish thinking about the problem, and get it ready for handoff. Plans are a concise harness tool for getting it to do that in a predictable manner. (Fable is really really a lot worse at terminal tasks than gpt-5.6 / Opus 5 and will get so lost in the sauce trying to do something relatively simple like "ssh to that server, get in touch with the Hermes ops agent on it, and get things deployed"; plans let it think through how that might happen, but then something competent like Terra can be dispatched on it)

[morgan.rebrand @MorganRebrand]

@67percentright @ChaosAgent_42 Nah, I remember you whining when she collapsed getting into a limo.

[maria @maria_rcks]

i’d like to see how real humans compare to these benchmarks, no ai tools.

[morgan.rebrand @MorganRebrand]

Reading open weights models' reasoning traces is wild.

"Am I an open weights model trained by Google? No, surely not. I am an open weights model trained by Anthropic? No, Anthropic doesn't have open weight models. So who was I trained by? Does it matter?"

[morgan.rebrand @MorganRebrand]

@theo It's certainly no 5.6 with persistence, but boy howdy, it's so much more capable than Opus 4.8, even on routine tasks.

[morgan.rebrand @MorganRebrand]

@UnitreeRobotics Shit, it's those pets the bad guy from Animorphs had

[Queen Chimerazilla ☭🐾 @Qveen_Potato]

Reminder to do your HRT!

[maria @maria_rcks]

What features do you want to see in T3 Code?

[morgan.rebrand @MorganRebrand]

@charloofers You and me both

[morgan.rebrand @MorganRebrand]

@BocaBola_ Codex might have been able to salvage it!!