Mathias Gehrig

271 posts

Mathias Gehrig banner
Mathias Gehrig

Mathias Gehrig

@MathiasGehrig

Skydio - Autonomy ML Engineer. PhD @UZH_en and @ETH_en in Zurich in CV and Robotics.

Zurich, Switzerland Beigetreten Mayıs 2019
329 Folgt284 Follower
Mathias Gehrig retweetet
Niyak Ghorbani (نیاک)
Niyak Ghorbani (نیاک)@GhorbaniiNiyak·
Persian and Jew celebrating together in London
English
1.9K
18.8K
98.7K
4.3M
Mathias Gehrig
Mathias Gehrig@MathiasGehrig·
@chrisoffner3d I always ask to provide evidence for any claims and statements made. Not sure if that is a reliable "fix" but I find it can help. Still, need to check as usual ofc
English
1
0
1
14
Chris Offner
Chris Offner@chrisoffner3d·
Claude when I tell it that the paper it cites to support a claim actually argues the opposite of that very claim.
Chris Offner tweet media
Chris Offner@chrisoffner3d

@e71xx Don't be an AIrhead. Always check the robot's bullshit!

English
1
0
1
303
Mathias Gehrig retweetet
Jeremy Howard
Jeremy Howard@jeremyphoward·
Politics and organizational behavior have always been the *most* important thing to consider and understand when working on AI risks. But nearly the whole "alignment" world has focused only on *technical* risks (i.e "what if AI wants to turn us into paperclips).
autumn@paramitanoia

did lesswrong ever predict that the first big challenge to alignment would be "the us government puts a gun to your head and tells you to turn off alignment"

English
18
30
206
25.4K
Mathias Gehrig
Mathias Gehrig@MathiasGehrig·
@BlackHC yes. I worry more often nowadays about this tbh. Democracy has become weak and fragile (even in Switzerland where I live, I see problems) and it's not looking good. A non-zero chance that things go south and this lands in the wrong hands; especially in the US.
English
1
0
1
14
Mathias Gehrig
Mathias Gehrig@MathiasGehrig·
@BlackHC It's not a disagreement with your point. Autonomous systems can be misused and give those more power who control it. Just like any technology, really.
English
1
0
0
15
Andreas Kirsch 🇺🇦
Andreas Kirsch 🇺🇦@BlackHC·
This is a straw man reply. To make it obvious: "Guns are not changing anything vs swords. It's the feudal lords that control us. Plenty of that around right now." Autonomous policing & surveillance remove a fail-safe (humans can refuse illegal orders) and pose new risks on top of corruption etc
English
1
0
0
41
Mathias Gehrig
Mathias Gehrig@MathiasGehrig·
What a disgrace
Robert Fico 🇸🇰@RobertFicoSVK

IF THE UKRAINIAN PRESIDENT DOES NOT RESUME OIL SUPPLIES TO SLOVAKIA ON MONDAY, ON THAT SAME DAY I WILL ASK THE RELEVANT SLOVAK COMPANIES TO STOP EMERGENCY ELECTRICITY SUPPLIES TO UKRAINE. Since the beginning of the war, Slovakia has been helping Ukraine. Around 180,000 Ukrainians are currently on our territory, we provide humanitarian assistance, and we organize joint government meetings. We are doing substantially more for Ukraine than some other countries. The Ukrainian president refuses to understand our peace-oriented approach and, because we do not support the war, he is behaving maliciously toward Slovakia. First, he halted gas flows to Slovakia, causing us damages of €500 million per year. Now he has stopped oil flows, causing us further losses and logistical difficulties. If the West does not mind that the Nord Stream gas pipeline was blown up, Slovakia cannot accept Slovak-Ukrainian relations as a one-way ticket benefiting only Ukraine. Slovakia is a proud and sovereign country, and I am a proud and sovereign Slovak. If oil supplies to Slovakia are not resumed on Monday, I will ask SEPS, the state-owned joint-stock company, to stop emergency electricity supplies to Ukraine. In January 2026 alone, these emergency supplies, needed to stabilize the Ukrainian energy grid, were required twice as much as during the entire year of 2025. Given President Zelensky’s unacceptable behavior toward Slovakia, treating it as a hostile country, I consider it absolutely correct that I refused to involve the Slovak Republic in the latest €90 billion military loan for Ukraine.

English
0
0
0
40
Mathias Gehrig retweetet
Victoriano Izquierdo
Victoriano Izquierdo@victorianoi·
In 20 years, vibe coders will look at the Linux kernel repo the way we look at the pyramids. In awe, unable to imagine how they managed to drag all those giant stones and pile them up in the middle of the desert.
English
257
1.6K
17.7K
528.7K
Mathias Gehrig
Mathias Gehrig@MathiasGehrig·
@prajdabre the stupid but also correct answer: forgot to move the model to the accelerator, running on CPU.
English
0
0
2
298
Raj Dabre
Raj Dabre@prajdabre·
Basic ML interview question: You are massive believer of LORA due to lack of compute. You took a dense model and fine tuned it on your dataset with Lora with a rank of 2. You then start doing inference but find that this model is a lot slower than the original model that you fine tuned. What was the one thing you forgot to do before inferencing?
English
22
11
298
44.6K
Mathias Gehrig
Mathias Gehrig@MathiasGehrig·
@jon_barron So comiting any change to default values would destroy reproducabilty? Because default values are not captured by the sed command, right?
English
0
0
0
424
Jon Barron
Jon Barron@jon_barron·
Deleting all my config dataclasses today. Each hparam is now just a default argument value and an `# hparam` comment so it’s greppable. Sweeps are just an agent running a sed script on the repo, launching, and reverting. Model variant configs are all bash scripts of sed calls.
English
9
2
101
15.3K
xjdr
xjdr@_xjdr·
if your model dim isn't divisible by 128, what the actual fuck are you doing? ive seen this like 4 times in the last few days
English
19
4
278
35.5K
Mathias Gehrig
Mathias Gehrig@MathiasGehrig·
@francoisfleuret exactly. It's just mind blowing that this is possible. However, many low hanging fruit applications (like video gen) are kind of lame imo. Maybe this is the price to pay towards a future where we actually use the MACs for more useful things?
English
0
0
0
51
Chris Offner
Chris Offner@chrisoffner3d·
Why the fuck is everyone counting git pushes as a success metric all of a sudden? This means nothing!
English
3
1
12
2.6K
Mathias Gehrig
Mathias Gehrig@MathiasGehrig·
@giffmana "extended thinking" in the chat UI does it (screenshot); "standard" fails (did not "think"). To be fair, my first instinct was also to answer with "walk".
Mathias Gehrig tweet media
English
2
0
5
991
C Zhang
C Zhang@ChongZitaZhang·
Using Swiss field test data to train the model. Performs generally well but sometimes fails on test data with waste piles in the street💀
English
1
0
12
1.6K
Tanishq Mathew Abraham, Ph.D.
Tanishq Mathew Abraham, Ph.D.@iScienceLuvr·
ViT-5: Vision Transformers for The Mid-2020s "a systematic investigation into modernizing Vision Transformer backbones by leveraging architectural advancements from the past five years" * LayerScale * RMSNorm * original MLP design with GeLU activation * both APE and 2D RoPE jointly * registers with a separate 2D RoPE * QK-Norm * remove bias terms in the QKV projection layers 84.2% top1 accuracy on ImageNet-1k, 1.84 FID on ImageNet-256
Tanishq Mathew Abraham, Ph.D. tweet media
English
11
72
398
103.7K
Mathias Gehrig
Mathias Gehrig@MathiasGehrig·
@timos_m Many people think so but actually making it happen is the tricky part ;)
English
1
0
1
42
Timoleon (Timos) Moraitis
Global structure emerging from local rules. A trillion-dollar recipe for continual learning, long-horizon agents, and post-Moore's law non-Von Neumann computing.
English
1
0
7
397