Teona (@itsteona) header
Teona (@itsteona) profile picture

pretty is not the same as usable. building @ripli_ai

London
Joined April 2009
Photo by Teona (@itsteona)Photo by Teona (@itsteona)Photo by Teona (@itsteona)Photo by Teona (@itsteona)Photo by Teona (@itsteona)Photo by Teona (@itsteona)Photo by Teona (@itsteona)Photo by Teona (@itsteona)Photo by Teona (@itsteona)Photo by Teona (@itsteona)
X is a wild place. All these posts going around about simple feature discovery are getting millions of views. This guy has 14m views and 74k likes 😳
14
as spotted SpaceX just launched startup program 👉 x.ai/startups Up to 6 months credits for Grok, Grok Build, Grok Bot, and the Grok API, all from one balance.
xai quietly published grok for startups terms. credits for grok, the api and grok bot. if you already run half your company on agents this is basically free rent x.ai/legal/startup-program-t…
1
66
weird thing with GPT Astra is that medium burns through my 500 pro usage in half a day leaving tasks unfinished, while Astra Max token consumption is so much better. it is the best model/effort to use for serious implementation. It uses less tokens overall for completed task. reasoning more means less mistakes per tasks and less rework later. Of course you don’t need it for basic fixes and UI changes. Sol 6.1 is pretty much useless for serious implementations. I would never trust it in production work.
1
50
been testing @subquadratic today. very impressive. I gave four models the same 220 real stores. Each one got the shop’s public product data, four product photos and the same 13 questions. Is it a real working shop, what does it sell, that kind of thing. SubQ Snap answered in about a second and the whole run cost 6 cents. Opus, Sol and Grok agreed with our full research a bit more often on what a shop sells, but Snap was the best of the four at spotting dead shops, and it costs somewhere between 70 and 140 times less than the others. I didn’t expect the cheap one to keep up this well. Next I’m trying it as the first pass, so the bigger models only look at the shops worth a closer look.
1
2
7
553
Codex has some jokes. @OpenAIDevs that one is for you. Maybe consider not to ask your own AI agents to confirm they are human
1
108
Been really enjoying Opus 5.5, its blazing fast and really good in UI/UX. Spent few hours on a large video director prototype with many screens and AI assisted planning. It follows instructions really well. used only 3% of my allowance on high effort. I am on Max 5x not even 20x.
96
spending sunday setting up an agent to run my x. the part that took longest was getting it to write like me first i had it audit my own posts. what gets replies, what dies, how long my sentences actually are then i asked chatgpt, codex and claude to describe how i talk, with real quotes. they've read more of me than anyone put it all in one voice guide with the words i use and the ones i never would. the drafts got a lot closer anyone else doing this? or is everyone still writing "be casual" in the system prompt
1
1
117
i think the model is basically the rendering engine. the workflow around it is the product. or am i wrong?
1
1
102
xai quietly published grok for startups terms. credits for grok, the api and grok bot. if you already run half your company on agents this is basically free rent x.ai/legal/startup-program-t…
1
178
I spent zero codex allowance on this. @OpenAI Dot built this in the cloud computer and gave me the link to gpt site. There are so many things you can do with dots + being on a live call to talk through everything is another level of working. It's not perfect but it was a super fast orbit prototyping.
139
Consolidated UI is coming next week. with dots full access
This quoted post is unavailable.
74
Astra 6.1 faster and 50% cheaper in coming next week
Replying to @poteto
Hey now. You don’t know what we’re releasing next week
1
59
Sometimes X makes me feel like I’m already too late to build in AI. The algorithm shows you this tiny bubble of people who seem insanely far ahead, building things you didn’t even know were possible. Then you zoom out and realize 98% of US households still aren’t paying for AI. We’re so early. It’s kind of crazy.
98% of US households aren't paying for AI yet More charts in State of Markets II: a16z.news/p/state-of-markets…
67
I really don't think people understand yet how powerful @OpenAI Dots are. I think even tho our weekly usage was cut i half if you use dot correctly you will get 3x of usage or more. its grokbot on steroids and much smarter.
48
is it only me? I am getting very bored of all these new model releases that change literally nothing. gpt model list is becoming like fast food menu. most of claude models don’t even know why they exist. seems to start getting bit desperate
55
Welcome to AGI! ASTRA GPT 6 reportedly scores 98.6% on ARC-AGI-3 and immediately shuts down Claude, Gemini, Grok, Cursor…
1
274
New sections been added to Codex. This is super useful, thank you @thsottiaux and @OpenAIDevs for always listening to the community 🙏🏻
Replying to @thsottiaux
Could we ship project organization this week. subfolders like Marketing, Ops, and Product under each project, with related chats grouped inside each one? It would make the sidebar much easier to manage.
136
Just handed GPT 5.6 Sol light a disposable card with an upper limit and told it to find the hotel I wanted + book it. No confirmations. No questions. Just do it. 10 minutes later → booking confirmation. Life is about to get a whole lot easier 😊
56