Qweee

[Qweee @FuzzyPointer]

@zack_overflow Fable: Be creative and make a billion dollar business. What's that? I'm bankcrupt, my app doesn't scale and you built a generic SaaS product? Guess I can't do any better than that. GG Fable, making a billion dollar business is impossible.

[Qweee @FuzzyPointer]

@fernandodilland @niklasdev @Muse Granted usage is still not great on the 20x plan, but you can do _some_ things on it. It seems one 5 hour limit = 33% of the weekly.

[Qweee @FuzzyPointer]

@fernandodilland @niklasdev @Muse I hooked up Muse to Claude Code instead and it uses dramatically less usage. But the model stops randomly on long tasks. Likely the reason the Muse CLI has those extra steps. Claude made a hook for it that loops and makes it finish tasks at least.

[Qweee @FuzzyPointer]

Opus using GPT 6.1 Sol to set up a NAS for me and even its tired of Sol's shit

[Qweee @FuzzyPointer]

@theo Hey Theo,

I’ve been a fan for a long time

Been coding since 1981 back in ye cobalt, ah I still remember... back then we call them dragons not bugs, like this one time... my great gran was that legendary coder... then you inspired my...

So when you adding Jev to T3 Code?

[Qweee @FuzzyPointer]

@theo How much usage do you get out of Muse Code for Wonder Full? Found my usage limits get obliterated very fast, vs even OpenAI now-a-days.

[Qweee @FuzzyPointer]

@thsottiaux Hmmmm, something to drain our halved usage limits even faster. Tasty

[Qweee @FuzzyPointer]

@thsottiaux OpenAI like "we're halving your limits in the Pro plan" while also still giving resets out like candy.

[Qweee @FuzzyPointer]

@theo Making my coworkers smarter

[Qweee @FuzzyPointer]

@thsottiaux This is quite insane. Opus 5.5 comes out, blowing up Anthropic limits so you can get even more done, Codex limits a bit on the fritz. Then OpenAI come out and chop off the knee caps of the Codex limits??? Never thought "stop hitting yourself" would apply to OpenAI one day.

[Qweee @FuzzyPointer]

@NoLuckLugia @YahyaDanboos I think copying the combat system is fine, but they copied almost everything. When I saw this I legit thought it was a KH mod reskinning Sora. Similar moves, menus, ANIMATIONS. It looks like a clone lacking any sense of originality.

[Qweee @FuzzyPointer]

@theo Also something that generally helps is adding to your global user rules something like "I use dictation, sometimes word may be garbled. Fill in the word that it sounds like instead", so models generally understand. Though of course nothing stops the semantic fuckery

[Qweee @FuzzyPointer]

@theo I used wispr daily, for everything. They I am their 1% top user or something. I agree with everything said here. The other day it misheard me so bad it told the model to go shoot a toddler.

Along with their buggy app, I want something better.

[Qweee @FuzzyPointer]

@theo Theoretically something could be for "navigability" of markdown, if you have a very complex domain. Is the agent grepping through it enough? Might be, not seen a situation where anything more advanced is needed yet.

[Qweee @FuzzyPointer]

@theo 1 week from now "It's crazy how fast open weight caught up"

[Qweee @FuzzyPointer]

@theo I don't really understand their thought process here. "New GPT Launch" who can use it??? Literally no one but those in the special in group. What's the point??? It's like showing a kid his birthday present, only to tell him he can't have it yet and its coming "soon".

[Qweee @FuzzyPointer]

@theo It is always hilarious when Anthropic say they cut costs. They always go up. If you read their docs for Fable 5.1 they say it "re-writes entire files more often", "writes denser prose" etc. So.. more output tokens, the part charged up the wazzo, makes cache savings seem less.

[Qweee @FuzzyPointer]

@mattpocockuk "coding is solved, but the code is completely broken" riiiiiiiiiiiiiiiiight. If coding is solved, but bug riddled, we've "solved coding" since the absolute dawn of time.