Pinned Tweet
๐ 10 days ago I made this account, and I've been posting non-stop since then.
๐ This is the progress I made. Over 400K impressions and over 300 followers! It's only the start!
๐ค If you like Tech and AI, let's connect! Let's help each other and grow together!
๐จ Fable 5.5 TESTING Part 16
๐ I asked Claude to make an ocean horror trailer in three.js.
I never touched the code. ๐คฏ
โฑ๏ธ 194 minutes
๐ธ $17.40 in API costs
๐ค Claude Fable 5.5 Xhigh in Claude Code
What it did on its own ๐
๐ช Solved the jump in physics: 44 m, hitting the water at 101 km/h
๐ Drew the sea from 44 summed waves, per pixel
๐ฆ Simulated nine glowing tentacles as physics chains that close in around your head
๐ซง Threw 70,000 bubbles past the lens at the plunge
๐ซ Cut 12 panicked breaths out of one recording and timed them to the edit
๐ A real stethoscope heartbeat and a real blue whale call at half speed, no music, no synthesis
๐ต๏ธ A re-render "finished" in 2 seconds. It read its own contact sheet, saw the old cut and found the stale cache
๐ฆ 39.6 s ยท 1080p ยท 24 fps ยท 2.39:1 ยท 26 shots ยท 40 real recordings
Then I said "make it like an edit, like its a movie trailer" ๐ฌ
โ๏ธ Recut the one-take into 26 shots and re-timed all the sound to the cuts
๐๏ธ Built a film print: halation, anamorphic streak, gate weave, bars
๐ค Title cards: ONE BREATH. ONE WAY DOWN. IT CAN WAIT.
camper ๐ฆ๐ฃ retweeted
It absolutely does. Both companies should have it. Because the price jump from 20 to 100 dollars is way too steep. So for normal customers, who do not hardcore code all the time, it'll be perfect. Instead of buying multiple 20 dollars account when one runs out.
camper ๐ฆ๐ฃ retweeted
And to use your subs normally, we need to run your models in Fast mode (2.5x more) AND your subscription gives about half of what of Anthropic gives, real numbers:
Pro 100= 2x
Pro 100= 4x
Pro 500=10x
YIKES
It absolutely does. Both companies should have it. Because the price jump from 20 to 100 dollars is way too steep. So for normal customers, who do not hardcore code all the time, it'll be perfect. Instead of buying multiple 20 dollars account when one runs out.
camper ๐ฆ๐ฃ retweeted
The biggest bottleneck for Chinese AI worldwide adaptation isn't the quality of models but abysmal usage limits in their subscription services, they just cannot compete with the generosity of US AI labs.
camper ๐ฆ๐ฃ retweeted
Hey @grok remove only ONE company that you dislike and why?
camper ๐ฆ๐ฃ retweeted
Hey @grok why big tech companies still hire software engineers when AI pretty can do all their work anyway?
And to use your subs normally, we need to run your models in Fast mode (2.5x more) AND your subscription gives about half of what of Anthropic gives, real numbers:
Pro 100= 2x
Pro 100= 4x
Pro 500=10x
YIKES
camper ๐ฆ๐ฃ retweeted
If Anthropic and OpenAI isn't always going for each other's throats, then we'll be at least an year behind in AI progress. Competition breeds innovation.
The biggest bottleneck for Chinese AI worldwide adaptation isn't the quality of models but abysmal usage limits in their subscription services, they just cannot compete with the generosity of US AI labs.
camper ๐ฆ๐ฃ retweeted
Get out of your basement, Mr. Reset and give us Astra 6.2 ๐ญ @thsottiaux
Hey @grok why big tech companies still hire software engineers when AI pretty can do all their work anyway?
If Anthropic and OpenAI isn't always going for each other's throats, then we'll be at least an year behind in AI progress. Competition breeds innovation.
Get out of your basement, Mr. Reset and give us Astra 6.2 ๐ญ @thsottiaux
camper ๐ฆ๐ฃ retweeted
Hey @grok, why people are paying 200s of dollars for Grok bot when you can just use Hermes bot for free? Am I missing something?
Hey @grok, why people are paying 200s of dollars for Grok bot when you can just use Hermes bot for free? Am I missing something?
camper ๐ฆ๐ฃ retweeted
As someone who has both Chatgpt and Claude subs, using @t3dotcodes by @theo is a breeze. I have been using that app for 2 weeks now, still working on the full review of it.
But here is my 2 favorite features:
1. Delegation: You can just delegate the task from a single thread to any model with any effort value and your agent will do that as a subagent.
2. Handoff: context handoff from claude to chatgpt and vice versa, T3 code has the best/seamless implementation of it.
These two things has made multi subs effortless. Give that amazing piece of software a try, you will not be disappointed.
As someone who has both Chatgpt and Claude subs, using @t3dotcodes by @theo is a breeze. I have been using that app for 2 weeks now, still working on the full review of it.
But here is my 2 favorite features:
1. Delegation: You can just delegate the task from a single thread to any model with any effort value and your agent will do that as a subagent.
2. Handoff: context handoff from claude to chatgpt and vice versa, T3 code has the best/seamless implementation of it.
These two things has made multi subs effortless. Give that amazing piece of software a try, you will not be disappointed.
camper ๐ฆ๐ฃ retweeted
๐จ Fable 5.5 TESTING Part 15
๐งช I asked Claude to inject blue ink into a glass flask of water.
One prompt. I never touched three.js. ๐คฏ
โฑ๏ธ 156 minutes
๐ธ $24.17 in API costs
๐ค Claude Fable 5.5 Xhigh in Claude Code
What it did on its own ๐
๐ Wrote its own fluid solver: 39.3 million cells at 0.38 mm, baked in 66 minutes on a laptop RTX 4060
๐ง Pushed 33.5 million ink particles through that flow, rendered in three.js
๐ฎ Ray-traced the glass and water with real refraction, Fresnel and total internal reflection, plus 1 million photons a frame for the blue caustic on the floor
โณ 10x slow motion through the jet, then a time-lapse of up to 50x to a fully mixed sphere, all in one continuous shot
๐ Synthesized every sound in code and timed it to the film speed, so the plunger's groan slows down with the slow-mo
๐ Spotted faint ladder bands in the jet, measured the particle density on the GPU and traced them to 3 separate bugs
โ๏ธ Caught its fluid sim quietly losing 58% of the ink while stirring, and put it back
๐ฆ 30 s ยท 1080p60 ยท 1,800 frames ยท 640 velocity keys ยท 6.2 GB bake ยท 25.5-minute render



