Ali Yesilyaprak

CTO at Skillpine

[Ali Yesilyaprak @_mayy00_]

@theo @ashen_one Generational web data trauma healing over time

[Ali Yesilyaprak @_mayy00_]

@thsottiaux I am amazed by GPT 5.6's capabilities, even compared to Fable.
However these 2 limitations combined:
- Short context
- Inability to truly rewind the conversation
make long context high value work impossible. I know there are cost concerns but please find a solution :)

[Ali Yesilyaprak @_mayy00_]

@GergelyOrosz Not sure here either but I've been reading 2 levels:
1. Periodic work. Ex: Every week, work on tackling tech debt.
2. Redesigning all company work with loops/systems and create swarms. Ex: One loop reads comms, creates issues, another scans issues, orchestrate implementation.

[Ali Yesilyaprak @_mayy00_]

@bryan_johnson You will get better Bryan and become one of the symbols of our species in our fight for longer lives.

[Ali Yesilyaprak @_mayy00_]

@theo Fucking insane.
It is such a great model that it makes me emotional.
Honestly, I wanna call it agi but i am scared of getting bullied.

[Ali Yesilyaprak @_mayy00_]

@gdb @beffjezos Will you follow Anthropic's approach (subscription/classifiers) and leave us feeling worthless or will you continue to be awesome?

[Ali Yesilyaprak @_mayy00_]

@c4darkness @MistralAI It's too powerful for commons, they need to add guardrails first.

[Aravind Srinivas @AravSrinivas]

frontier ai for all

[Ali Yesilyaprak @_mayy00_]

@DeryaTR_ @GeminiApp They earned this themselves though. Every single time, they created huge marketing hype followed by even larger disappointment, consistently.

[Ali Yesilyaprak @_mayy00_]

Opus 4.8 is the most overrated model there is. I used to think Opus is intuitive and GPT is accurate. But now I see, Opus doesn't have that either.
Fable is the only AI model that doesn't require too much effort.
That should be the bar for efficiency gain claims.
#claude #opus

[Ali Yesilyaprak @_mayy00_]

@himanshustwts Because they are full of shit

[Ali Yesilyaprak @_mayy00_]

@robj3d3 Absolutely.
Modularized my codebase, made correct decisions every time.
Fantastic brainstorming partner too!
It is the first model that thinks like us

[Steve Faulkner @southpolesteve]

The new beta Claude Code model switcher

[Ali Yesilyaprak @_mayy00_]

@kilicdarogluk Kapa ceneni dede

[Ali Yesilyaprak @_mayy00_]

@BlackHC @suchenzang It doesn’t really seem that way to the rest of us

[Ali Yesilyaprak @_mayy00_]

@DeryaTR_ What do you mean by “fraction of the cost”? It is almost the same cost as 3.1 pro (1.5/9 - 2/12).
Given the disappointing track record so far of Google, this iteration will probably end up being just a significant price increment.

[Ali Yesilyaprak @_mayy00_]

@devarshb05 @OfficialLoganK The only real progress

[Ali Yesilyaprak @_mayy00_]

@sundarpichai Can't wait for another broken preview model.

[Ali Yesilyaprak @_mayy00_]

@alex_whedon Time to reimplement everything then!

[Ali Yesilyaprak @_mayy00_]

@OpenRouter @sama GPT 5.5 is great but the costs openai blog mentioned don't reflect the reality at all.
We all use way more input than output tokens.

If you want loyalty, you need to let us run a profitable business.