
Replying to @HuggingPapers
Made a cool project on this a while ago! github.com/EnesYilmazcode/Ki…
Replying to @Talha_Agcayazi
Thank you! I told Claude to find the livestream and analyze it frame by frame and to replicate it! Also I split up the Modal rendering across 3 separate GPU's to get a faster turnaround time!
Replying to @oleksiypodolian
Thank you! I used @threejs, it took 6 review rounds according to Claude
Replying to @ItsGoharr
When making it, I broke it down by scene so I could fix issues more directly
Replying to @ElementsofaSoul
how it's made:
claude opus 5.5: research, script, 3d scene, physics, sfx
@modal: rendered on 3 GPUs in ~1 min
@ElevenLabs: voice
@threejs + @FFmpeg: frames + mp4
@Sketchfab: models by metarex.4d, Oleksii Rozumnyi, wintrez, tonielpro520, TAIGA-ZOE, zwir3kk
Replying to @threejs
Typically a wrong answer key, a question with more than one fair reading, or a test that rejects a correct answer.
Often it's plain human error. Even grade-school math (GSM8K) has a key that counts 8 AM to 11 PM as 3 hours.
github.com/EnesYilmazcode/be…
I benchmarked benchmarks. It audited 14 popular AI benchmarks and found that about 1 in 5 sampled MBPP and Terminal-Bench items are broken.




