Have been giving GPT 6 Astra a genuinely good two days(mostly because I maxed out my ant sub)
first of all I don't remember Astra being this good to work with but on the other hand Astra was always hugely expensive so I guess I never gave it a real shot
now with banked resets and such working with it on High effort yeah this is awesome
secondly been working on how to display in-transcript images better and also how to have them show in a sidepane when in-transcript doesn't suffice but also on a canvas based on tldraw with all that gives
after 1 hour of ideation or such it produced a very nice concept off the bat
@ryo_vibecoding this uis is so busy. how can I focus on work when 2 million things are happening at the same time? is it modular? can you hide things? does it have a system to notify you and only with what is important?
you can build this without any library in a day with a semi-capable agent
with sota models(fable 5.2, opus 5.5 or gpt astra) even less
I know because I did it for my tooling
just tell it to brainstorm with you how to show the artifacts and what your main harness is(claude code, codex, pi, deepseek etc)
you won't get notion in one day but you will get this you described and then you just build from there or stop and focus on other things
@supahvee1234 declining your contribution because of AI...sure I guess. Their repom they decide what gets contributed or not
Stealing your work and not giving credit? no he absolutely deserves shit. Within the scope of the law and such of course
after working with the previous version for a few hours there were some things I didn't like:
- the column of buttons in the middle
- alignment was still off, especially when expanding/collapsing lists
- paths was not highlighted as clickable interactions
- indentation was hardly noticeable(yes there was some but not enough, at all)
and after dogfooding that for....2 hours I noticed more things(second picture)
working on a tasks and projects tool
first iteration by claude looked great but wait a minute....
design really is such a nit picky thing and it never really feels complete. In many ways just like software dev.
working on a tasks and projects tool
first iteration by claude looked great but wait a minute....
design really is such a nit picky thing and it never really feels complete. In many ways just like software dev.
@CodeThorpeX@MimiLeague Given the language on the cake is Swedish I’ll assume they’ll work in Sweden and we have no such agency that goes around and police’s cafés etc
Health department and such of course but nothing like “these don’t fit our criteria for what is allowed to be sold”
@Scobleizer This is a reskin of quake 3
The same weapons and movement and look of the arena
Original game my ass. I don’t think that’s what Felix did but rather a result of the prompt and the models training together with system prompt“don’t copy an actual game bro”
Seems like Google actually cooked at last
Been giving Fable 5.2 everything I can throw at it and it just makes it happen, in some cases where opus 5.5 struggled a bit.
Now Carbon is on that level? damn
While being 4x cheaper(on average)? damn
What a day! I tested Gemini 4 Carbon on real work. 2 repos, 105 replanted bugs. Find and fix what you can.
It scores just behind Fable 5.2. The results:
- Fable 5.2 (max): 58 for $139.65
- Gemini 4 Carbon (high): 57.3 for $31.32
- Sonnet 5.5 (max): 51.3 for $154.03
- GPT-6 Astra (max): 45 for $33.03
- GPT-6.1 Sol (max): 44.3 for $5.88
- Gemini 4 Argon (high): 40.7 for $7.16
Google's own benchmarks looked super strong, but nobody could access the model. Turns out they didn't lie: +16.6 points over Argon, at less than a quarter of Fable 5.2's cost.
Nobody should be laughing at Google anymore.
(the next effort level drops soon in the thread 🧵)
@pradeepXkapoor wonder what happens if the prompt says it's a collaborative effort and we both own the code. does it still merge but don't re-write features and such? :D