David Finsterwalder | eu/acc
@DFinsterwalder
Followers
2K
Following
14K
Media
978
Statuses
6K
Digital Nomad, Philosopher, Nerd | Founder: https://t.co/IhsLSaVvb1 | Lecturer: Darmstadt University https://t.co/h4KXUfuf5c | Prev: Founder & CTO https://t.co/d3XkUCmD4o.
Germany
Joined July 2014
The founder @adcock_brett of @Figure_robot claimed that @UBTECHRobotics used CGI. They followed up and shared a drone vid. If THAT would be CGI - then kudos! Because the battery LED flickering async to the rolling shutter of the cam would be the most meticulous CG detail ever!
They said it looked too perfect to be real. But perfection isn't fabricated—it's delicately engineered. This is the historic mass delivery of UBTECH (优必选) Walker S2. The next era of intelligent manufacturing is here. Let's build it together! #WalkerS2 #HumanoidRobots
0
0
0
The internet is for sharing cat videos Generative AI is for making cat videos
0
1
3
Even if you are a 100% sure this is the case you should never ever post something like this as a founder of a competitor company. This makes you look unnecessary defensive. If I would be a Figure investor this behavior would make me nervous.
Look at the reflections on this bot, then compare them to the ones behind it. The bot in front is real - everything behind it is fake If you see a head unit reflecting a bunch of ceiling lights, that’s a giveaway it’s CGI
0
0
7
This looks impressive! Really fast movement. A bit sloppy, but fast!
0
0
3
@3blue1brown Here is a video of @ylecun showing convolution neural networks in 1989 that work better than this MLP here (but are a bit more complicated to explain and visualize). Fun fact: the MNIST dataset used is also from Yann LeCun. https://t.co/qfKKpH4PSF
0
1
12
@3blue1brown Since a couple of people are posting fail cases. Yes this is known and basically intended. I showed the fail cases in class. It’s a simple neural network architecture and small. We known better methods since the 90s for this (conv nets) and training should do data augmentation.
@lep1c2l0 @threejs @PyTorch Yeah this is a common issues of MLPs. If something is larger or smaller, shifted left/right/up/down or rotated its already "out of training distribution". This is where "Conv nets" and/or data augmentation (flipping, rotatating and scaling training images) helps.
1
0
8
@3blue1brown If you enjoy posts consider retweeting it. And if you like tweets about vibecoding, neural networks, AI, 3D, philsophy and shitposting consider following me. https://t.co/rec7crAYur
I vibecoded this neural network visualization for my students and open sourced it. It shows a simple MLP trained on MNIST handwritten digits at several training steps. The visualization is using @threejs and it comes with training code in @PyTorch . Link + repo 👇
1
1
19
The visualisation is inspired by @3blue1brown cover image for this video. If you don't know his content and want to learn more about neural networks I recommend watching his videos.
1
2
62
This was 100% vibecoded using Codex and similar to other things this worked flawless, because of how amazing @threejs is. The @pytorch code for an MLP is obviously trival and also was no challenge. See this thread for something similar that I vibecoded.
An archaeologist friend of mine wished he had a certain software for 10 years. I vibecoded it for him that evening over a bottle of wine.🍷 He never even used ChatGPT before, and his mind was blown. 🤯 Link, repo, and story below.👇
1
2
42
When I have a bit more time I go over it and will translate it and make it more standalone educational. Also I am currently talking with an exhibition to use it and I might add an option to connect a tablet to it via WebRTC to paint the numbers there.
1
0
10
So first of all sorry for all the text being in German. I wanted to translate it to English before sharing it, but didn't have the time. Also in the current form the educational information is limited since it was used in class with me explaining it.
3
0
21
Here is the github repo with all the code for training and visualization all under Apache 2.0 licence. https://t.co/BSpw4UKC1v
github.com
Interactive web visualisation for handwritting detection using a simple neural network - DFin/Neural-Network-Visualisation
4
11
123
Before I say a couple more words here is a link to play with it: https://t.co/XmSifnC2uL Everything runs in a browser and the weights are stored in a json. Might take a bit to load on slower connection. Also its intended for desktop/larger screens and menu overlaps on mobile.
2
2
73
The new 1 Trillion parameter Kimi K2 Thinking model runs well on 2 M3 Ultras in its native format - no loss in quality! The model was quantization aware trained (qat) at int4. Here it generated ~3500 tokens at 15 toks/sec using pipeline-parallelism in mlx-lm:
80
225
3K
I am currently teaching AI tools at the university Darmstadt for the Extended Reality (XR) degree and they have the most hilarious reinterpretation of the "Ivan the Terrible" painting hanging on the wall. VR folks can relate. 😂
1
0
3
Ooops. I am an idiot. 🙈 I just vibecoded a small prototype for my students and committed an API key in clear text to a public repo. Fortunately it got nuked by OpenAI within 1 minute. Thanks @OpenAI I guess.
0
0
2
I am currently dealing with GDPR compliance in two voice AI projects I am working on. This account makes me laugh and cry simultaneously. Oh no. I just disclosed personal info without his consent. Henrick I will delete this tweet within 30 days in compliance with GDPR. 🤣
On a business trip in Berlin Order an Uber from the airport to hotel I noticed the privacy policy changed on the Uber app Ask the driver about it "Yeah, now you have to give us permission to store data about your destination" I declined to give permission. Then the driver
0
0
0
"Any sufficiently advanced overfitting is indistinguishable from generalization." - Unknown
we don't "saturate" benchmarks, we overfit them.
6
20
348