Paul Gavrikov
@PaulGavrikov
Followers
431
Following
3K
Statuses
636
PhD student | Deep Learning, Computer Vision, Model Interpretability
Freiburg im Breisgau, Germany
Joined April 2017
@Adam_Karpiak I think it would depend on how far I was in the process. At later stages, phone might be more emphatic but not necessary. Either way, I’d love to get actual feedback and not be ghosted
0
0
1
Someone should tell Elon about semantic search.
🚨BREAKING. From a program officer at the National Science Foundation, a list of keywords that can cause a grant to be pulled. I will be sharing screenshots of these keywords along with a decision tree. Please share widely. This is a crisis for academic freedom & science.
0
0
0
This. There’s absolutely no point in stuffing an LLM with random facts beyond basic knowledge. Let it access the web. This requires us to rethink benchmarking …
re: OpenAI's DeepResearch on the Last Humanity's Exam on one hand, it's unfair to compare an AI with web-access vs others without, especially for knowledge intensive task on the other hand, we need to stop cramming random facts in a model, and invest more in tool-use
0
0
1
@giffmana You’ll love your sticker forever, because whatever you do - the residue won’t come off.
0
0
1
RT @NC_Renic: Write every day for 1 hour. If you miss a writing day add 1 hour to the next. It’s incredible! Tomorrow I’m writing for 7 yea…
0
5K
0
@BoltzmannBooty I have no idea what they mean by "malicious" but I'd bet on someone scraping CoT conversations without having to run the model themselves?
0
0
1
We should start boycotting proceedings that do not have a "cite this article" button.
0
0
4
Paper: (Camera-Ready coming soon) Code & Data: Work by me, @jovita_lukasik, @jung_vision, Robert Geirhos, @jmie_mirza, @margret_keuper, @JanisKeuper. @dwsunima, @cvml_mpiinf, @MIT_CSAIL, @UniSiegen, @GoogleDeepMind
0
0
12
TIL: #LaTeX captions have a short mode that will be used is lists of figures etc \caption[''short'']{''long''}
0
0
0
Tests like these should be part of any LLM eval, otherwise we’re just selling hype.
(Yet) another tale of Rise and Fall: DeepSeek R1 is claimed to match o1/o1-preview on olympiad level math & coding problems. Can it handle versions of AIW problems that reveal generalization & basic reasoning deficits in SOTA LLMs? ( 🧵1/n
0
0
1
Universities: You will be forever in our family. Also: Your email account will be deleted asap.
0
0
1
Now accepted at #ICLR2025 under a new title: Can we talk models into seeing the world differently? More details soon.
🚀 GPT-4, Gemini, Qwen, LLaVA: LLMs are stepping into the multi-modal arena with a bang! But let's zoom in on their vision 👁️. Our preprint peels back the layers on a crucial bias in vision models that sets most apart from humans: the texture/shape bias 👉
0
2
10