
Data Science @paperpile social science geek, tinkering with python and R // loves #dataviz and 🐈// the causes are not always in the data // views are my own
Bulgaria
Joined June 2010
- Tweets246
- Following307
- Followers106
- Likes1.6K
Ivaylo Sakelariev retweeted
Mental health has gotten worse around the world but having a good job, an education, and people around you are still the best chances of doing well.
Much more on our study, Global Mental WEALTH, with over 250 collaborators from 92 countries now at link.springer.com/article/10…
Ivaylo Sakelariev retweeted
A NeurIPS submission got caught with hallucinated citations. The authors' admitted they asked ChatGPT to fill in BibTeX from paraphrased in-text citations.
We ran a simulation to see how often this fails—and built a free tool to catch it.
First, the tool: the Paperpile Citation Checker. Upload your BibTeX, and it verifies every reference in real time. When it's done, you can download a cleaned-up file or share the results URL with coauthors.
Why now? This month alone, large-scale analyses in The Lancet and on arXiv highlighted how widespread hallucinated citations have become. And arXiv announced a severe penalty for submitting manuscripts with hallucinations.
How does this happen? A case study on OpenReview spells it out: authors gave ChatGPT paraphrased citations (author-year in-text citations, or titles) and asked it to generate properly formatted BibTeX. Instead of looking up the real metadata, the model hallucinated it.
We ran a simulation to understand this better. We took 76 real references, paraphrased them, and asked different models to convert them to BibTeX.
Two big findings:
- Larger models are better: the “thinking” or “pro” model in an AI assistant outperforms the “instant” or “lite” model.
- Incorporating web search results can reduce hallucinations.
We also found a strong correlation between citation count and hallucination risk. Highly-cited papers appear more often in training data, so models get them right.
But newer and less-cited references tend to be more prone to AI-induced hallucinations.
We spot-checked today's top AI assistants on the same task. ChatGPT and Claude were relatively clean. Gemini had more hallucinations.
The difference? ChatGPT and Claude ran dozens of web searches while generating the output. Web search is what grounds these tools.
Our recommendation: Using AI to generate a bibliography is not worth the risk.
Full blog post: paperpile.com/blog/citation-…
Ivaylo Sakelariev retweeted
Hallucinated citations are showing up in preprints.
We built Paperpile Citation Checker to check your bibliography before you submit.
Here's what we learned along the way 1/3
Ivaylo Sakelariev retweeted
At Paperpile, it's our job to be obsessed about citation accuracy.
So we built a free tool to flag incorrect references before you submit a preprint.
Upload a BibTeX file and the Citation Checker verifies every reference in real time. Download the updated file or share the URL with your coauthors.
Ivaylo Sakelariev retweeted
Paperpile now saves references directly from Google Scholar Labs, the new AI-powered search tool that lets you find papers using natural language questions instead of keywords.
The summary is automatically added as a note, so you keep the context of why a paper matched your question, not just the citation.
Ivaylo Sakelariev retweeted
We've been working on something we're really excited about: a direct connection between your Paperpile library and the AI assistants you already use. ✨️
You can now send PDFs to ChatGPT, NotebookLM, or Claude.
Ivaylo Sakelariev retweeted
Paperpile's first AI-native integration lets you send PDFs directly to ChatGPT, NotebookLM, or any supported AI assistant — right from your library 💫
Find the key quotation or data point that unlocks the next step in your research.
Ivaylo Sakelariev retweeted
I have viewed maybe 200,000 historical photos from the 1800's to today, across all kinds of archives. The neoteny effect is very real and began probably around the 1920s for some people.
Interestingly historical men resemble modern men more (vs women) in a lot of ways.
Ivaylo Sakelariev retweeted
I spoke about the origins and evolution of my book/course, open science and science reform, and once mentioned the phrase "dungeon master"
Our latest episode is live!
We spoke to @rlmcelreath about the state of science, the parallels between science and business, and the path towards doing things better.
Thanks Richard for a great conversation!
Listen wherever you listen to podcasts. podcasters.spotify.com/pod/s…
Ivaylo Sakelariev retweeted
Excellent, I highly recommend Dan’s new Substack. Several of my recent posts have been influenced by his work. You can expect a sharp and insightful perspective on human social interactions.
I'm going to be writing here 👇 ≈weekly this year. In this first post I introduce the blog, explain its title 'Conspicuous Cognition', and outline why the pursuit of social approval drives the evolutionary weirdness of human behaviour and thought. 1/7
conspicuouscognition.com/p/c…
Ivaylo Sakelariev retweeted
Yes I will offer my online science-focused Statistical Rethinking course again starting in January 2024. Registration going up at the end of this month. All the course materials are already online though, so why wait? Update your posterior today youtube.com/watch?v=FdnMWdIC…
Ivaylo Sakelariev retweeted
1/🧵 Do you see what I see? Don’t miss the chance to participate in #ThePerceptionCensus – a unique, large-scale study of ‘perceptual diversity’ – the differences in how we each experience the world. There’s not long left 🏃🏽♀️ perceptioncensus.dreamachine…
Ivaylo Sakelariev retweeted
Science is great but academia not so much. Earlier this week I revised my talk "Science as Amateur Software Development". The unprofessional way that academics curate & process data would benefit from professional habits of software engineers and chefs: youtube.com/watch?v=8qzVV7eE…
Ivaylo Sakelariev retweeted
🚨 New working paper: How good are people at judging the veracity of news?
Our finding: People can tell true from false news. 🤓
But when they err, they are slightly more skeptical towards true news than they are gullible towards fake news. (1/14)
osf.io/n9h4y/
🧵
Ivaylo Sakelariev retweeted
Ok, so LLMs are a Thing.
How do they work? Embeddings.
WTF are embeddings?
I spent a year doing a deep dive. But when I was researching, I couldn't find anything that explained them in business, engineering, AND math contexts. So I wrote a thing.🚀
vickiboykis.com/what_are_emb…
Ivaylo Sakelariev retweeted
There's this _really_ cool trick that you can do with LLMs that I felt was underappreciated so I figured I'd write a blogpost about it.
In short: you can compare it with an existing pipeline to find examples to annotate next.
koaning.io/posts/large-disag…
Ivaylo Sakelariev retweeted
We are hiring! ✨
1️⃣ Mobile/React Native Engineer
1️⃣ Backend/Data Engineer
Apply here 👉 pprpl.co/3OUTQBB





















