Pascal Iversen (@Iv_Pascal) profile picture

BioML, Uncertainty Estimation/XAI, Scientific Reproducibility @ FU Berlin

Joined October 2019
Photo by Pascal Iversen (@Iv_Pascal)Photo by Pascal Iversen (@Iv_Pascal)Photo by Pascal Iversen (@Iv_Pascal)Photo by Pascal Iversen (@Iv_Pascal)Photo by Pascal Iversen (@Iv_Pascal)Photo by Pascal Iversen (@Iv_Pascal)Photo by Pascal Iversen (@Iv_Pascal)Photo by Pascal Iversen (@Iv_Pascal)Photo by Pascal Iversen (@Iv_Pascal)Photo by Pascal Iversen (@Iv_Pascal)
Replying to @manojpandey66
Hybrid is the future!
1
1
537
Replying to @rust_ruslan
Looking forward to reading how I am discussing with my agent what food to cook for lunch in the middle of working on my research code?
90
Intense moment at MLCB when @jmschreiber91 tells an audience of MLCB presenters that half of the presentations were Claude slop: youtube.com/live/GpKq9Qy4uy4… He argued for what I think should be a tautology: science communication should stay in the hands of … scientists.
26
Replying to @PI010101
Journals do not judge papers, peers do, and they do not have time to review all the autoresearch output.
1
372
Replying to @boardyai
should have! The word data does not appear once in the original version.
1
6
New Bitter Lesson for computational biology: no amount of computational power (or intelligence) can offset datasets that are too small, noisy, confounded, and biased.
1
1
87
Replying to @schlimmerrick
Da liegst du falsch. Es sind dann ungepaarte Stichproben, und damit ist die Teststärke geringer, aber bei so einer großen Stichprobe sollte das kein Problem sein. Fast jedes Randomized Controlled Trial läuft so.
1
18
Replying to @SamadSyed
Hey Samad, sorry to hear! I will be in Sydney, happy to print and put up yours if you send me the pdf. Dm me :)
13
3,272
Pascal Iversen retweeted
Does the code match the paper's claims? We audited 100 NeurIPS 2025 papers with an LLM agent pipeline. Only 8% fully traced their findings to code, and we flagged 609 discrepancies. Now in TMLR, and @Iv_Pascal & I will present it at NeurIPS in Sydney! auditowl.ai
1
6
305
You conclude from a scireports paper that peer review is fundamentally unserious?
1
1
994
Replying to @liron
The q is a false dichotomy, it is not either store or regulate. There is also runner diarrhea.
29
Replying to @liron
It is not true Liron! During extreme exercise more can pass through. You said "always".
1
83
Replying to @ChadNotChud
the reason is to run claude code headless?
7
They are everywhere in the training set, marked as intended bullshit. It would recognize it. You have to write a new hoax paper :)
12
Replying to @felixringe
Teleological Reduction to the rescue... Judges clearly do not judge like a compiler; they recognize the spirit of a law.
2
476
You want phd students to work healthcare and child care jobs to get their phd?
10
325
Replying to @moarbugs
Can you give an example of such a problem? Asking for a friend
2
1
43
6,704
Replying to @ZhijingJin
In Germany, PhD students don't take classes, but teach them. Learning way more that way
9
1,134
I agree, I am actually researching how to best do that with agents: openreview.net/forum?id=xjCJ…
1
297