
vibing through ai and backend sharing what I like ai tools and demos ✉️ [email protected]
Joined October 2022
- Tweets4.8K
- Following974
- Followers1.3K
- Likes2.1K
Democrats won
Multiple news and agenda against AI risen and laws speculated
Jensen affairs comes to light and then another
He resigns
Trump could've said bring him back but no he didn't
Hmm something is going on
i think we're about to see a ridiculous number of AI-built startups fail for the exact same reason
Claude Code can build your app in a weekend
Codex can review the code
ChatGPT can write your landing page
and you can launch the whole thing before Monday
then comes the part nobody wants to talk about
you have 0 users and absolutely no idea where to find the first 10
coding was never the only hard part, we just made it cheap enough to discover the OTHER hard parts 💀
Baidu ERNIE 5.5 Flash Preview is quietly joining the AI coding race and this dragon demo is kinda hilarious 😭
Leo tested it on the "Cloud Dragon City" frontend challenge, the same concept he previously tried with Qwen 4-Max Preview
The result looks like a giant turquoise sea dragon carrying an entire Chinese city on its back, surrounded by mountains and mist
There's something oddly beautiful about the art style, almost like an illustrated fantasy world
Leo reckons it's roughly on par with DeepSeek V4.1 Flash, though that's his impression from the test
With Claude, Gemini, Qwen, DeepSeek and now Baidu all showing off these creative coding demos, I'm starting to care more about which model has actual artistic taste
Because this dragon has more personality than half the photorealistic AI demos on my timeline 💀
Google Gemini 4 Hydrogen is getting some wild early coding demos
A creator reportedly called "gemini-4-hydrogen" through the Vertex AI Python SDK and got a full Three.js scene in around 2 minutes 44 seconds on HIGH reasoning
And the test is ridiculous 😭
A pelican riding a red bicycle through an ocean at sunset, with a giant solar eclipse overhead
The creator says you can actually pause, orbit the camera and zoom in to inspect the scene
That's a much more interesting test than another pretty AI-generated video
Now I want Claude, GPT and Gemini building the same interactive 3D scene under the same time limit
Let's see whose pelican survives the physics inspection 💀
Google's Gemini 4 Hydrogen might have just leaked its API pricing 👀
A new SKU catalog screenshot reportedly lists:
• $1.50 / 1M input tokens
• $7.50 / 1M output tokens
There's also speculation about introductory pricing of $0.75 / $3.75, though that hasn't been confirmed
If those numbers hold, Hydrogen won't exactly be the dirt-cheap Flash model people might expect
And now I'm even more curious about the actual coding performance
Because at $7.50 per million output tokens, Google needs to give developers a pretty good reason to choose it over cheaper alternatives
The next Claude vs Gemini vs GPT coding-agent comparison could get expensive real quick 💀
A giant panda riding a tricycle is somehow becoming a serious test for Google's Gemini models 😭
Someone is testing Gemini 4 Hydrogen (Flash) with a browser-rendered SVG animation, and the creator says this attempt ran without crashing after an earlier one failed to move
The scene looks cute, but getting the wheels, pedals and character movements to stay synchronized is where these animations get tricky
I'd genuinely love to see Claude, Gemini and Qwen attempt this exact same prompt
No cherry-picked screenshots, just the full animations side by side
Let the panda decide who's actually cooking 💀
the scariest thing Claude Code or Codex can tell you is "all tests pass" after touching half your repo 💀
bro, i asked you to fix ONE bug. why are 23 files modified?
i genuinely wish coding agents were better at telling us what they couldn't verify instead of confidently announcing they're done
people who use coding agents daily, what's ONE thing you absolutely refuse to let them do without supervision?
Baidu ERNIE 5.5 Flash Preview just entered the AI pelican Olympics 😭
Leo tested it with the infamous pelican-on-a-bicycle prompt, and he's already comparing the result to DeepSeek V4 Flash.
The sunset, the little red scarf, the bicycle. Pretty damn clean visually.
But I have a better challenge.
Make the pelican take a sharp turn without its legs or bicycle turning into spaghetti 💀
I'm curious, what's ONE ridiculous prompt you'd use to expose the difference between these AI models?




