Experiments and data limits in AI for science
“AI-driven science is bottlenecked by experiments, lab work and data quality rather than by compute or reasoning alone”
5 people · 2 agree · 3 unclear · 8 views · 2026-10-05 to 2026-10-10
Who stands where on the statement above, from their own words. A person's side is shown only where two separate readings of the quote agree; otherwise they are under "Side unclear". This page does not say who is right.
Agree 2 people
- ✓Liam Fedus
No one’s going to zero shot the room-temperature superconductor.
If you throw a huge amount of compute against a bunch of noise, you’re not going to have a good thing emerge.
solving humanoids would actually be slower to kind of getting to some of our goals.
- ✓Ekin Dogus Cubuk
Most experimental data in the literature have such a high noise floor that it’s actually usually worse than DFT accuracy.
Side unclear 4 people
- ·Geby Jaff
This suggests that materials for the next step in spintronics may already exist, waiting to be recognized.
- ·Liam Fedus
the optimal reasoning strategies to do math, to do theoretical computer science, to do these types of things may not be the optimal reasoning strategies to do science.
- ·Reddit commenters
commenters predict researchers, especially PhD-level specialists, will need to “sluice through” large volumes of AI-generated hypotheses, drafts, or analyses to find genuinely valuable results
- ·Sal Candido
One misconception of scaling laws is that scaling laws are everywhere and they always exist. A lot of the work is actually finding that scaling law.
Week by week
- · 5: Geby Jaff ·, Liam Fedus ✓, Ekin Dogus Cubuk ✓, Reddit commenters ·, Sal Candido ·
Built 2026-11-21 06:50 UTC from the news knowledge
graph by code. Every quote was checked against the source's own words; which side a quote takes is a model's reading,
made twice.