LLMs can be unreliable, but some are very reliable if used in specific controlled ways. And you can quantify the reliability using benchmark tasks.

Neurosift is expanding with new visualization plugins each week.

Neurosift can lazy load data from remote HDF5 files hosted on S3. That's what enables exploration and visualization of NWB files on DANDI Archive, and other h5 files on public repositories.

Intended to be viewed on a desktop. You're probably on your phone, so enable desktop mode.

I guess discovery would include testing a hypothesis, developing a method to answer a question, observing or examining something unknown to get better insight. Maybe there should be a list...

I think it's a great system for knowing accounts are legit!

If there are published papers with this problem, what should be done? Is there a mechanism for marking those results for readers without going as far as retracting?