AI engineer | Shipping games, apps and agent experiments in public
Joined March 2024
- Tweets6.6K
- Following389
- Followers770
- Likes2.9K
I build games and apps with AI agents and post what actually happens. A few things I did recently:
1. Got glm in @cline to build a racing game where every texture and sound is generated, nothing downloaded.
2. I simulated an entire company to see where agents break.
3. Made a cel-shaded minecraft clone for my sister's birthday. One prompt, 3M+ tokens, ran for 4 hours.
4. And my agent saved my pc from a bsod that would have cost me 147$ for 😁
All of them are in replies 👇
more in hilights.
Kimi 3.1 was also able to do it. I remember it was not getting the event horizon so it wrote the mathematical equations for the calculation , found the right values and then work it's way to code the values in simulation webapp.
1
27
Well that's what fable 5.2 just did. I guess model on next level with the outputs I am getting during my test.
Till now only simple tests done, time to level up the task.
1
26
Long overdue for higher / elite class.
Asked fable to build its own launching page , no context apart from colours.
Result: nitter.gq/devloperhs/status/2119…
Claude Opus 5.2 just built myself a landing page, with no specific instructions.
Just asked fable to "build me a launch page for fable 5.2 following claude official colours"
It planned everything, researched all the details on fable 5.2 and built me a the entire self contained html webpage & required assets , all within 7 minutes at cost of 8% usage.
Insane speed and optimised usage , somehow costing less than opus 5.5
1
917
Well it's just one sentence and fable handled it all - planning, researching , validating and then building.
Only instruction I gave it was to follow the anthropic colors.
Best part it understood what I wanted and followed through till end.
As per my tests , model is good for long running and robotics related task , yet to try other tasks too.
2
2
90
Strangely enough claude fable 5.2 lost to gemini carbon at automation bench.
Hope this is not real and just a llm hallucination failing at that task, else need to be studied.
1
1
45
Robotics and long running related tasks got a major leap with this model. What all you explored and found model to be good at ?
591
Claude Opus 5.2 just built myself a landing page, with no specific instructions.
Just asked fable to "build me a launch page for fable 5.2 following claude official colours"
It planned everything, researched all the details on fable 5.2 and built me a the entire self contained html webpage & required assets , all within 7 minutes at cost of 8% usage.
Insane speed and optimised usage , somehow costing less than opus 5.5
harsh retweeted
Thrilled to share a big milestone from our AI & Infrastructure team at @Google.
Today we’re expanding our 8th generation TPU portfolio with:
• TPU 8i Max: Built for massive context windows and the highest compute density per rack
• TPU 8tx: Purpose-built for secure, confidential inference
This marks the first time we’ve introduced four distinct TPU systems in a single generation, in one year.
For developers, hardware is only as good as the software stack. And we want you to be able to build with TPUs as easily and efficiently as we do at Google. You can now:
↳ Run native PyTorch & Hugging Face models via TorchTPU (zero code rewrites)
↳ Cut KV transfer latency >90% with open-source TPU Sync
↳ Eliminate idle compute waste with Agent Compute Engine
Read more about our newest TPUs on our blog: bit.ly/3VfETPR
I look forward to connecting with builders, researchers, engineers and organizations pushing the boundaries of AI and computing!
They offered me 1 year free even before their release of deeplearning.ai and then 1.5 yrs free after release into mainstream.
Their batch was the only newletter I followed earlier, best one.
Andrew Ng and Laurence sir are just great , you feel just lost learning with them hands on:)
1
2
16