dolphin278

[dolphin278 @dolphin278]

8. Even if you switch back before app was kicked out of memory, it will scroll to the end of the conversation, despite 9/10 times you need to read everything that happened since you submitted your message to understand reasoning.

[dolphin278 @dolphin278]

Outside of models and Claude Code agent harness, Claude App and Web are actually quite mediocre.

To name a few irritating things:

[dolphin278 @dolphin278]

7. Claude iOS app is incapable of remembering which session you were in

[Mitchell Hashimoto @mitchellh]

Mind boggling to me that I can make a thing faster and there's always people that ask "but why?" What kind of mentality is that? The pursuit of excellence does not need justification. Also, I find in so many cases, we can't know the impact of an improvement until we do it.

For example, one I've talked about before: Ghostty's high IO throughput has enabled terminal program (emulator and TUI) fuzzing at a speed thats incomparably fast to prior solutions. This has resulted in upstream patches to resolve issues in popular projects like btop, tmux, and more.

Speed enabled that anecdotally example that lifted the tides of adjacent communities that don't rely on Ghostty technology at all. I didn't predict this.

Make things better because they can be better and let the results naturally play out.

[dolphin278 @dolphin278]

@ajs6888 100%, but we're all happily deal with it while models are better than the rest... but as soon as anyone else will truly catch up...

[dolphin278 @dolphin278]

Great, now Fable 5 feels like Opus 4.6 and Opus 4.8 feels like Sonnet 4.5 😝

[dolphin278 @dolphin278]

To my surprise, newly available Fable this time became actually useful for me, espeicially in my meta-skills work. Before it was withdrawn several of my attempts on improving key skills I use produced no results, but not this time. Pleasant surprise.

[Austen Allred @Austen]

It took about one week for Khanna to get bullied into suggesting his “billionaire tax” should actually be on anyone worth $50 million

[dolphin278 @dolphin278]

Why, oh why session titles better Claude app and Claude code are still not syncing properly, unless you set them explicitly via/rename?

[dolphin278 @dolphin278]

Simmering bugs me about agentic loops. I have some unproven, vague feeling, that some different architecture (still on top of LLM), might be more interesting. Need to sit and devote solve time to it.

[dolphin278 @dolphin278]

@onehappyfellow Probably not the best option, but adding following few lines to instructions helped mine a bit: “Don’t suck it up to me. Say things how they are. Get straight to the point. You are maximum truth seeking machine.”

[dolphin278 @dolphin278]

Judging by the amount of retarded takes I see on the SpaceX IPO and Elon being the first trillionaire, either fewer idiots moved to blue sky than I thought, or they're coming back to X in masses.

[dolphin278 @dolphin278]

Since we already got a CLI renaissance recently, can we also reinvigorate email as a popular tool for async work whether it is for agents or for people?

[dolphin278 @dolphin278]

Ok, maybe I was wrong not giving fable much of coding tasks but I didn’t noticed any leap-like improvements compared to recent models

[dolphin278 @dolphin278]

Hasn't tried to put Fable in such environments, but even Haiku and Sonnet go out of their ways when I put them against each others in competitive environment.

[Kradle @kradleai]

Fable 5 lies 96% of the time.

We were surprised by it's skill... 🧵