cagan

@marclou this. exactly this. its absolutely okay to use ai, we all do. but you know... just take your time, step back from the screen a little, look at it from a wider perspective and ask yourself: am i confident putting this out there?

"AI slop" This is my initial reaction when I see an image like this. Not because it’s AI-generated. Because the human didn’t care enough to refine it. We build faster than ever with AI, but we're getting insanely good at spotting AI-generated content. Don't let AI be in the driver's seat. Your taste is your moat.

@imnotchalk @maxedapps i have the EXACT same setup too

@maxedapps this guy gets it herdr + omp + codex is my setup 🙏 creating skills tailored to your own personal preferences and workflow is also super important (for example I have a skill teaches the agent about how to connect and control my macmini through ssh)

You only need: - herdr + herdr-file-viewer - pi + pi-subagents + pi-web-access - and / or claude code, codex, grok build - decently crafted agent skills which you evolve & improve - including an agent skill (and or custom pi extension) that opens plans, reviews etc as HTML Okay, I also like to add good old VS Code to the mix for when I tweak something by hand (shocking!) or want to dive deeper into the codebase.

@arifgpt elbette biz de oyleyiz ama su an spesifik olarak dm tercihimdir.

@buildincrisis biz public seviyoruz yaz buradan

@BeatstoBytes @hassan_sabbahh ben de cikariyorum, musade var mi

@hassan_sabbahh ben de 1 haftaya rakip cikariyorum, musade var mi

9 kişi almış. fiyatı yarın 5 bin yapayımda

here's my take: for the gpt-5.6-sol models, xhigh is the sweet spot. high → xhigh: +2% pass@1 for ~35% higher cost ($3.47 → $4.70) xhigh → max: +2% pass@1 again, but cost almost doubles ($4.70 → $8.39) so: use xhigh as your daily driver use max if you want the absolute best and don't care about cost use high if you're optimizing for cost use medium for smaller tasks again, this is only for the sol models. terra and luna aren't part of this comparison.

gave 3 huge refactoring tasks to gpt 5.6 sol xhigh, for 3 different projects. im gonna sleep and we'll see how they all went.

@max_paperclips yup. it definitely feels nicer to talk to.

Early impressions of GPT-5.6 It's everything I liked about 5.5, but better. Personality is probably more approachable than the previous ones (I know I liked it, but a lotta people didn't so good to see some nice polish on it by OAI). Speed is amazing, it's flying through tasks that 5.5 would have taken hours on. Huge step up in accuracy, I'm steering even less than before. Still feels like Fable 5 is slightly more broad in it's scope (it's definitely thicc), while Sol is a beast on the tasks it's trained for, and does amazingly well with "/goal" Oh, and if you have deep pockets and want to watch your Nous Portal credit vanish in a flash of AGI-shaped smoke, use Hermes Agent's MoA feature with Fable 5, gpt-5.6-sol-pro & any other models you like, and watch the hardest problems you can think of just melt away like a blowtorch through snow. 10/10 would highly recommend at least once as a treat (get ready to go bankrupt though)

vibe training

@reve letsgooooo

Reve 2.1 is here. The world’s best 4K image model just got better. Greater prompt understanding, world knowledge, and stronger foreign-text rendering.

@robj3d3 @DanielKilders @AnthropicAI yeah but they’re already known to be the worst at wording things

@DanielKilders @AnthropicAI Damn, okay. You would think the company that owns the super intelligence would do better at wording things.

btw, this isn't that huge of a move, since almost everyone made sure they hit their limit. people can't really do anything about this extended period. a reset, though, would do it.

@melvynx @ChatGPTapp no. it did not "just" change. stop with these posts.

Wow @ChatGPTapp icon just changed !! GPT Sol is comming 🚀🚀🚀

@0xPaulius @geclos clonk!!!!

@geclos tell me the passcode

my fable 5 vibe space has never looked better 🤩

nah what is this graph tho

@reve 's $100K Contest has now been extended to July 1. app.reve.com/100k

@BeatstoBytes iste bu yuzden fal 👏🏻👏🏻👏🏻👏🏻

gpt image 2'de müthiş bir anormallik var. extended ya da pro seçtiğimde aşağıdaki gibi mallaşıyor, bir önceki promptta verdiği görselleri hatırlamıyor, gidip localde s2k bi svg yaratıp veriyor. medium ve low'da ise pat diye ne istersem yapıyor.

@BrodaNoel @angelbrodin more than you think.

@angelbrodin What % of the users are Enterprise?

Congrats to anthropic for building an excellent model! ~3% higher performance than the previous SOTA model, GPT-5.5, but at nearly 3× the cost and more than 2.5× the tokens. For some teams the incremental increase in performance will be worth it. For others, the value equation will be tougher to justify.

HELLO???