Jay Patel
banner
infotainment.bsky.social
Jay Patel
@infotainment.bsky.social
🎷 vibe adulting

#HCI #PeerReview #SciPub
#toolsforthought #ResearchSynthesis
#OpenScience #MetaSci #FoSci

🔎 Research: ethnography of peer review
🧑‍🏫 Teaching: Stats, DataViz

🐢 UMD: College of Info
🌐 PhD Candidate: Info Studies / HCI + Data
🏝️ OASISlab
Hands down, the best PubPeer comment I've read in a while.
November 10, 2025 at 11:12 PM
While reading a data science conference paper today, I noticed an interesting transparency statement.

Is this common to data mining conferences?

I like this sort of statement; it reminds me a bit of the 21-word solution. #metascience
November 7, 2025 at 2:47 AM
And just over the past day or so, even more commentary point to conflict of interest issues (didn't declare a key patent) and issues with changing the clinical trial registration dates to make it seem as if the data were collected after declaring the study plan/registration.
November 5, 2025 at 9:55 PM
Hey, @gracewade.bsky.social why is this article continuously posted on BSky? The number of issues documented by academic researchers is worth reporting.

The story should be one of errors/potential fraud instead of a breakthrough, right?

pubpeer.com/publications...

Author reply in the thread:
November 5, 2025 at 9:19 PM
So how did you prompt it then? Typically, prompts starting with "You are an expert X with expertise in domains A,B, and C..." is effective.

The Google white paper that was published a while ago can be helpful: cloud.google.com/discover/wha...

Section: Strategies for writing better prompts
November 4, 2025 at 8:22 PM
I did both in my master's thesis under supervision by a mentor. What are the stats on reporting this? Would be nice to know.

apastyle.apa.org/jars/quant-t...

APA JARS-QUANT reporting guidelines mention diagnostics:
November 4, 2025 at 7:25 PM
Dear Reviewer 2: Go F’ Yourself.

Another gem in the #peerreview literature, a joke paper, finds that it's Reviewer #3 who's the real problem.

The paper even has a credulous PubPeer comment!

Paper: doi:10.1111/ssqu.12824s
PubPeer: pubpeer.com/publications/80F9ACFE1DC2E6510A4CC3D2D841C1
October 17, 2025 at 11:58 PM
Yes! Left is a blank PubPeer page where I submitted a comment (awaiting moderation, then never accepted). Right is Paperstars with the same comment.
October 1, 2025 at 5:30 PM
But did you get a photo?
September 4, 2025 at 6:07 PM
August 17, 2025 at 7:29 PM
Fourth find: Disclaimers abound. Might as well place them in reporting guidelines for standard communication given how popular they are.
August 15, 2025 at 1:16 AM
Round 3, folks! This time in red text at the bottom of the first page.
August 14, 2025 at 1:27 AM
Yes, the first screenshot is about OpenReviewer. I read that paper recently and was able to run the HuggingFace demo: huggingface.co/spaces/maxid...

Maybe try again?

The second screenshot is Liang et al. 2024: ai.nejm.org/doi/abs/10.1...
August 13, 2025 at 9:46 PM
BINGO! I call BINGO! How many variants can I find?
August 13, 2025 at 3:48 AM
"Can Large Language Models" returns 8k+ hits on Google Scholar.

"Should Large Language Models" returns 73 hits.

Before asking can...?, ask should...? and you'll save yourself a year's worth of research in some cases.
August 13, 2025 at 2:25 AM
AI researchers love to add disclaimers about the importance of humans in research activities, but I don't see much use for this kind of thing in practice.

Those who use their tools will do so as they like.
Disclaimers won't matter much in the long-run.
August 12, 2025 at 9:05 PM
"This is the first study..." in a paper makes me fume. 😤
On Google Scholar, it returns almost 2 million hits.

"This is the second study..." returns 3,840 hits.
That's a difference of ~520X.

I'm more likely to believe the latter claim.
📖 If you make a novelty claim, then back it up.
July 29, 2025 at 11:44 PM
This is how you enforce reporting guidelines: @neuripsconf.bsky.social does it right.

❌ Desk reject failure to comply

Which other venues do this sort of thing? #metascience
July 29, 2025 at 12:50 AM
My new favorite motto and insignia for slow and open (aka slowpen) science:

"Festina lente"
(Latin translation: Make haste slowly)
July 23, 2025 at 5:58 PM
If you want to run a study on LLMs' abilities, please prompt engineer thoroughly.

This is the laziest and most honest method I've seen in my review so far:

"Whether this could have influenced the results remains currently unknown... Prompt designing is also time-consuming..."
July 22, 2025 at 12:04 AM
3. Sorting and filtering by sentiment would be nice on the X/Bluesky pages.
July 11, 2025 at 10:13 PM
2. The colored squares indicating sentiment for the X and Bluesky tabs could be more prominent. I missed them for the first minute or two on the page.
July 11, 2025 at 10:13 PM
@altmetric.com The sentiment analysis feature is wonderful for researchers.

A few thoughts from my recent use to consider:

1. Can I view a feed of papers/posts by sentiment category (e.g. only papers with post > 10% negative)? That'd be useful to find problematic papers.
July 11, 2025 at 10:13 PM
Who watches the AI agent benchmarks?

❌ 7/10 contain shortcuts or impossible tasks.

❌ 7/10 fail outcome validity.

❌ 8/10 fail to disclose known issues.

preprint: arxiv.org/abs/2507.02825
blog: ddkang.substack.com/p/ai-agent-b...
July 11, 2025 at 9:18 PM
Behold! Genius-level product design:

#Grok search options:
July 10, 2025 at 10:00 PM