Johan Edstedt
@Parskatt
Followers
1K
Following
3K
Media
247
Statuses
2K
PhD student @cvlisy I like 3D vision and training neural networks. Code: https://t.co/CxpPxw9g7L
Joined July 2012
This holds up even for any p(doom) > 0
This holds up even for 1% p(doom)s.
1
0
2
IMO a good CUDA implementation of dense corresp RANSAC would fix bottom issue.
1
0
11
While working on simfit++ this is something I investigated. My hunch is that big gains comes from mode switching. However, it is not clear to me whether better models would disambiguate.
0
0
0
B) Beyond IID: Scoring assumes independent Gaussian. In practice we have very correlated data. Using this could yield major gains. (3/4)
1
0
2
There are at least two possible reasons A) Overfitting: "with 4 parameters I can fit an elephant". different thresholds wiggle solution slightly (and might jump between modes). Oracle has privileged access. (2/4)
2
0
2
There's still a large gap between oracle thr and thr estimation: A very short thread (1/4)
1
0
3
AGI is when LLMs produce correct bibs
The viral new "Definition of AGI" paper has fake citations which do not exist. And it specifically TELLS you to read them! Proof: different articles present at the specified journal/volume/page number, and their titles exist nowhere on any searchable repository.
0
0
8
if you’re seriously trying to understand agi, core concepts you should familiarize yourself with: y = ax +b, y = max(x,0), y = x / ||x||, y = f(x) + x
if you’re seriously trying to understand agi, core concepts you should familiarize yourself with: dy/dx = dy/dz dz/dx
0
0
8
The term “AGI” is currently a vague, moving goalpost. To ground the discussion, we propose a comprehensive, testable definition of AGI. Using it, we can quantify progress: GPT-4 (2023) was 27% of the way to AGI. GPT-5 (2025) is 58%. Here’s how we define and measure it: 🧵
0
1
5
RoMa out there matching things I didn't intend :)
1
0
12
It's the same trap people fall into where they think that reading every paper is going to make them an incredible researcher (since now they have all the "knowledge"). https://t.co/DIkfpXGRRa
0
0
3
There's this disconnect where people think creativity is about getting things done. As soon as you use an AI tool to "simplify" a creative task you are no longer the creator of it, you are a critic. A critic does not create.
It takes some kind of balls to even suggest this to the creator of The Wire:
1
0
9
Somewhat unfortunate that in computer vision the term "scene graph" has nothing to do with multiple images of a scene.
2
0
1
If your p(doom) is 25% this is a reasonable thing to do.
Hi, it’s Guido on hunger strike Day 23 outside the offices of Anthropic, going strong. There is a great and profound evil in this place. Each day, I sit and watch at lunch time and dinner time as delivery drivers on scooters and bikes stop at the curb and bring packages of
1
0
3
Why is like every 3rd shot of a woman in a bikini?
We are building “Open Source Nano Banana for Video” - here is open source demo v0.1 We are open sourcing Lucy Edit, the first foundation model for text-guided video editing! Get the model on @huggingface 🤗, API on @FAL, and nodes on @ComfyUI 🧵
1
0
4