still somewhat loves computers β’ some exp with llms
Joined June 2025
- Tweets 283
- Following 233
- Followers 99
- Likes 11,606
42 Photos and videos
Pinned Tweet
Visualizing vision token compression for VLM using Pixel Shuffle
5
5
80
7,687
a dataset of pangram false positives
46
must resist the path of least resistance
1
59
when you finally get Fable access through a shady token broker but Dario intercepts you:
η°ε¨ζδΊδΈθ½¬η«θΏθ½δ½Ώη¨ Fable 5
η¨ζ·οΌιΏε§¨οΌδ½ 们η Fable 5 ζδΉθΏθ½δ½Ώη¨
δΈθ½¬η«ιΏε§¨οΌζ们ζε
ι¨ζΈ ι
δΈθ½¬η«ιΏε§¨οΌζ们ζε货
η¨ζ·οΌη₯ιδΊοΌθ°’θ°’δΈθ½¬η«ιΏε§¨
1
3
152
summary instructions in papers brought to you by deepmind (:
"From AGI to ASI": new paper from our team.
This report investigates how AI might develop beyond AGI. It describes theoretical limits, potential pathways, and potential bottlenecks.
arxiv.org/abs/2606.12683
3
15
1,648
might've done the opposite of catching the last chopper out of nam last year
58
singularity what if: text diffusion break out came before transformers
1
4
404
memory startups are not slopshops but these would be just open-source libraries if not for da bubble
Jun 10
Memory startups are a complete tarpit and Iβm surprised YC hasnβt figured this out sooner
105
uh i want to hot swap codex "auto-approval" with an advisor like agent
84
π«©
Jan 25
suspiciously precise floats, or,
how I got Claude's real limits
link below
2
138
codex browser use fails on windows
2
96
curious about these problems with such low pass rate
For days, many folks here are citing DeepSWE as the benchmark that restores reality only because it shows GPT 5.5 on top. But actually, it almost gets a single entry right: the top one, and all the rest is shuffled.
2
279
how do you know if you are good enough for MATS
9
50
10,605
they have completely killed the reach for small account
Visualizing vision token compression for VLM using Pixel Shuffle
8
84
5,449
Visualizing vision token compression for VLM using Pixel Shuffle
5
5
80
7,687
Yes, It's live: ctx-0.github.io/pixel-shufflβ¦
Yes, that's fumo.
Here's the repo: github.com/ctx-0/pixel-shuffβ¦
1
3
529
Made this after reading @huggingface 's SmolVLM2 paper!
3
451
are we never getting a unified encoder multimodality
For the past years my research focus was on unifying models and training paradigms across modalities. Today I'm excited that we're releasing our latest model aligned with this theme:
Gemma 4 12B, a dense encoder-free model which processes raw text, image, and audio inputs!
1/
3
256