Sign in

Andy Wheeler

@crimede-coder.bsky.social
147 followers 71 following 234 posts

Data scientist, with background in criminal justice. Consulting website at crimede-coder.com Personal blog at andrewpwheeler.com

PostsRepliesMedia
Andy Wheeler @crimede-coder.bsky.social · 29/09/2026
New VerusCite blog post, some editors/authors have given pushback on what I flag as an AI hallucination. Often these are articles that exist, but have whole-cloth fabricated elements, like made up authors or a totally wrong periodical. veruscite.com/blog/what-is...
000
Andy Wheeler @crimede-coder.bsky.social · 26/09/2026
Hallucination of the day in Probation Journal Made the tool easy to go and manually review, so I changed a minor error and not found in this paper on checking more closely. veruscite.com/share/Zi1no1...
000
Andy Wheeler @crimede-coder.bsky.social · 25/09/2026
Hallucination of the day in a Philosophy journal -- "generating metacognitive feelings of confidence that are easily mistaken for epistemic justification" -- seems like an apt quote veruscite.com/share/98TNXr...
000
Andy Wheeler @crimede-coder.bsky.social · 24/09/2026
New blog post, using extra credits to just churn out ideas I have had for years. Peer reviewed papers were always symbolic. The ease you can write them now I am hoping spurs academics to consider "doing things in the real world" as a measure of success. andrewpwheeler.com/2026/09/24/a...
andrewpwheeler.com
Adventures in using AI to write papers
Until recently I had only used the AI tools pay-as-you-go, but lately I have been experimenting with the $20 monthly subscription. So I ended up with extra credits to burn. I would periodically jus…
000
Andy Wheeler @crimede-coder.bsky.social · 23/09/2026
Another 4 hallucinations added to the hall today. Sign up, add a few of your own, veruscite.com/hallucinations
000
Andy Wheeler @crimede-coder.bsky.social · 21/09/2026
I have created an app to conduct on-demand synthetic control analysis for crime trends at the city level (using Jeff Asher's RTCI data). Check it out at apwheele.github.io/SynthRTCI/
131
Reposted by Andy Wheeler
Stijn Ruiter @stijnruiter.bsky.social · 17/09/2026
“What works?” is only the beginning. To improve investigative interviewing, we also need to ask: how, for whom, under what conditions, and at what cost? Our paper explores how EMMIE can help bridge research and practice. Open-access at: doi.org/10.1080/1561...
doi.org
The practical relevance of investigative interviewing research: applying the EMMIE framework to bridge the research-practice gap
Investigative interviewing research has produced valuable insights into effective techniques for gathering information. However, this research often fails to fully bridge the gap between academia a...
022
Andy Wheeler @crimede-coder.bsky.social · 17/09/2026
I have not been the one submitting a paper for years now, can someone confirm that the DOI's are inserted by Springer copyeditors? (Not authors?) So when they cannot find the right DOI they just insert the next best guess (which indeed is not the correct paper). veruscite.com/share/8NKB24...
000
Reposted by Andy Wheeler
Andy Wheeler @crimede-coder.bsky.social · 16/09/2026
Part of the difficulty in writing a book on LLM APIs is that they change so often. LLMs for Mortals is 6 months out, and I wrote a recap of updates needed so the book recompiles (mostly just newer model strings). crimede-coder.com/blogposts/20... See post for promo for EPUB and print versions.
crimede-coder.com
LLMs for Mortals 6 months later
I recompiled Large Language Models for Mortals six months after release. Here is what broke, what I got right, and what I got wrong.
001
Andy Wheeler @crimede-coder.bsky.social · 16/09/2026
I have created a stats page on VerusCite. My guesses of minor errors have been too low, but hallucinated citations (even acknowledging false positives) may only be 2% of all cites. At least among the selected papers scanned veruscite.com/stats
veruscite.com
VerusCite Stats
How many papers and citations VerusCite has checked, average processing time, and the share of hallucinated and minor-error citations.
000
Andy Wheeler @crimede-coder.bsky.social · 16/09/2026
Part of the difficulty in writing a book on LLM APIs is that they change so often. LLMs for Mortals is 6 months out, and I wrote a recap of updates needed so the book recompiles (mostly just newer model strings). crimede-coder.com/blogposts/20... See post for promo for EPUB and print versions.
crimede-coder.com
LLMs for Mortals 6 months later
I recompiled Large Language Models for Mortals six months after release. Here is what broke, what I got right, and what I got wrong.
001
Andy Wheeler @crimede-coder.bsky.social · 15/09/2026
Hall of Hallucinations up to 50+ papers and over 200+ hallucinated citations. veruscite.com/hallucinations Sign up, add a few of your own.
veruscite.com
Hall of Hallucinations — VerusCite
51 papers and 231 hallucinated citations catalogued in VerusCite's Hall of Hallucinations.
010
Reposted by Andy Wheeler
Andy Wheeler @crimede-coder.bsky.social · 15/09/2026
New blog post on veruscite, several academic papers scan large corpi of papers and estimate the rate of hallucinated citations to be ~1%. I think this estimate is likely low, as they only scan title differences. Authors being wrong is another common source veruscite.com/blog/halluci...
veruscite.com
Why estimates of hallucinated citations are probably low
Large scale audits of fabricated references only match on titles. Hand-checking the 10 most recent papers in the hall of hallucinations, a third of the errors were a real title with the wrong authors ...
001
Andy Wheeler @crimede-coder.bsky.social · 15/09/2026
New blog post on veruscite, several academic papers scan large corpi of papers and estimate the rate of hallucinated citations to be ~1%. I think this estimate is likely low, as they only scan title differences. Authors being wrong is another common source veruscite.com/blog/halluci...
veruscite.com
Why estimates of hallucinated citations are probably low
Large scale audits of fabricated references only match on titles. Hand-checking the 10 most recent papers in the hall of hallucinations, a third of the errors were a real title with the wrong authors ...
001
Andy Wheeler @crimede-coder.bsky.social · 13/09/2026
Hallucination of the day in BMC Geriatrics This journal charges over $3000 for article processing fees. My tool is $2. They could raise APCs by less than 0.1% to cover my tool. veruscite.com/share/ujEErd...
010
Andy Wheeler @crimede-coder.bsky.social · 12/09/2026
Some updates to the Crime Decomposition explorer. Included series for violent and property crimes, a sub-panel so it is easier to see trends across each crime type, and a toggle to index those graphs so global and local start at the same line. This is Scranton, apwheele.github.io/CrimeDecomp/...
011
Andy Wheeler @crimede-coder.bsky.social · 11/09/2026
Some trends in Pittsburgh crimes are going up: MV Theft, Rape, and Thefts apwheele.github.io/CrimeDecomp/... Data from the real-time-crime-index
010
Andy Wheeler @crimede-coder.bsky.social · 10/09/2026
Hallucination of the day, an editorial in Bulletin of Mathematical Biology veruscite.com/share/SofefM... Hall of hallucinations is up to 45 papers and growing, contribute some examples!
010
Andy Wheeler @crimede-coder.bsky.social · 09/09/2026
New VerusCite blog post, Springer AI reference checks are not working veruscite.com/blog/springe... Despite Springer putting out a press release a year ago, this post has clear examples where their internal tool is obviously lacking. Academics should demand more rigor from peer reviewed pubs.
010
Andy Wheeler @crimede-coder.bsky.social · 08/09/2026
Hallucination of the day in Crime & Delinquency. A tell for current heavy AI use in articles is odd citations. Why would a criminology article cite a nursing article on how to conduct content analysis? veruscite.com/share/Er_8Pb...
010
Andy Wheeler @crimede-coder.bsky.social · 07/09/2026
Hallucination of the day. It is hard to know if the entire results are just fabricated as well. So did they actually do the experiment? Or just have AI make up the whole paper? The figures in this paper are clearly AI generated as well. veruscite.com/share/4LrRd8...
001
Andy Wheeler @crimede-coder.bsky.social · 06/09/2026
Hallucination of the day is a bit meta -- an article about using Knowledge graphs to prevent hallucinations in LLMs. It is another example of Google Scholar indexing a fake paper. The authors created this fake paper in a paper in 2025, and it persisted to their more recent 2026 paper.
000
Reposted by Andy Wheeler
Andy Wheeler @crimede-coder.bsky.social · 05/09/2026
In another one of this fellow's paper, he cites an article about pets and Hurricane Katrina. The rub is the article does not exist, and even if it did, was published in 2001. veruscite.com/share/PlEeuA...
011
Andy Wheeler @crimede-coder.bsky.social · 05/09/2026
In another one of this fellow's paper, he cites an article about pets and Hurricane Katrina. The rub is the article does not exist, and even if it did, was published in 2001. veruscite.com/share/PlEeuA...
011
Andy Wheeler @crimede-coder.bsky.social · 05/09/2026
Hallucination of the day comes from researchers at Microsoft. My tool classifies hallucinations as information that directly conflicts with the known ground truth. If you just make up papers entirely though, it will say not found. veruscite.com/share/N4OWim...
000
Andy Wheeler @crimede-coder.bsky.social · 03/09/2026
Hallucination of the day, another set of totally made up authors after the first author listed is correct. veruscite.com/share/4nN-gR... Citations being wrong does not mean the paper is wrong, but probably signals other folks should take a closer look.
000
Andy Wheeler @crimede-coder.bsky.social · 02/09/2026
New blog post on various updates, submit to ASEBP, Crim Consortium, VerusCite, and a new async Python course I contributed to NACA andrewpwheeler.com/2026/08/27/c...
andrewpwheeler.com
Crime De-Coder and VerusCite Updates
If you do not follow me on social media (X, LinkedIn), you likely have missed the different work I have been up to. So this is a remix of some of that, posted here for folks to follow along. Crime …
000
Andy Wheeler @crimede-coder.bsky.social · 02/09/2026
I know many editors are concerned about privacy and submitting articles to LLMs. I understand these concerns, and built several safeguards into VerusCite from the start to handle this: ZDR, I only upload the references, and I do not retain the PDF. veruscite.com/blog/how-ver...
000
Andy Wheeler @crimede-coder.bsky.social · 02/09/2026
Hallucination of the day, an article in Scientometrics that suggests AI and human based reviews of novelty in proposals substantially differs. Citation in question has just a bunch of made up authors on a paper. veruscite.com/share/yyPIEi...
010
Andy Wheeler @crimede-coder.bsky.social · 01/09/2026
Hallucination of the day, a systematic review with over 380 citations, my tool flagged 16 as suspicious (mostly author lists that are totally wrong). veruscite.com/share/OO2mdf...
000
Andy Wheeler @crimede-coder.bsky.social · 01/09/2026
One of the benefits of my tool is not just for identifying AI hallucinations, but just to help copy edit. Here is an example with 16 duplicates in the bibliography for a Deloitte white paper (and it uses ibid, so these are clearly errors). veruscite.com/share/iipxp1...
veruscite.com
Artificial intelligence in drug discovery what it is where we stand and the path forward.pdf — VerusCite results
93 citations · 62 verified · 0 hallucinations
000
Andy Wheeler @crimede-coder.bsky.social · 31/08/2026
Hallucination of the day from Artificial Intelligence Review, just totally wrong authors on over a dozen papers. LLM as a judge did not quite catch the errors in the LLM as a judge review paper. veruscite.com/share/OotE2M...
011
Andy Wheeler @crimede-coder.bsky.social · 31/08/2026
Cleaning up some issues with the crim consortium pages this morning. Check it out at, crimconsortium.com, jobs board, checklist for open criminology scholarship, open source pages for different members (mentor match still needs to get the certificate for https). Join as a member!
000
Andy Wheeler @crimede-coder.bsky.social · 30/08/2026
Project of the week attempting to use up my leftover codex credits, optimal search paths given a probability surface. github.com/apwheele/opt... Spent a week on this in 2020 and was unable to generate any useable model. Codex did it in less than 3 hours (and with only 16% of weekly budget).
000
Andy Wheeler @crimede-coder.bsky.social · 29/08/2026
British Journal of Criminology is up to 3 papers in the Hall of Hallucinations veruscite.com/hallucinatio... Sign up, add some more.
010
Andy Wheeler @crimede-coder.bsky.social · 28/08/2026
Some recent additions to the hall of hallucinations. Journals include British Journal of Crim article (2nd example), Trends in Organized Crime, Crime Media & Culture, and Policing & Society. AI hallucinations are not just a problem in Computer Science! veruscite.com/hallucinations
011
Andy Wheeler @crimede-coder.bsky.social · 27/08/2026
Hallucination of the day is a citation for a Covid and crime paper published in 2019. veruscite.com/share/J8aveL... $2 a review, and if you are an editor it is worth it to save publishing embarrassing errors like this.
010
Andy Wheeler @crimede-coder.bsky.social · 26/08/2026
I go over why I don't use Google Scholar for VerusCite in this blog post, veruscite.com/blog/why-i-d... The skinny of it is an article can hallucinate a reference, and Google Scholar will create an index for it. Plus it has various errors propagated over time.
veruscite.com
Why I don't use Google Scholar in VerusCite
Google Scholar can turn hallucinated citations into indexed records, and it can keep older bibliographic errors in circulation. That is why VerusCite does not use it as a fallback.
000
Andy Wheeler @crimede-coder.bsky.social · 25/08/2026
To keep track of all the hallucinations of the day, I have created a page, the hall of hallucinations veruscite.com/hallucinations user contributions welcome!
veruscite.com
Hall of Hallucinations — VerusCite
12 papers and 45 hallucinated citations catalogued in VerusCite's Hall of Hallucinations.
000
Andy Wheeler @crimede-coder.bsky.social · 23/08/2026
Guys, this paper from DeepMind has multiple hallucinated references veruscite-data.com/share/-7KyAg...
1102
Andy Wheeler @crimede-coder.bsky.social · 23/08/2026
Hallucination of the day in Criminal Justice and Behavior, only one identified hallucination by my tool, but I am a bit suspicious of "Thinking in systems: International bestseller" as well. Can read the report at veruscite-data.com/share/VCkkSb...
veruscite-data.com
CovidPrison.pdf — VerusCite results
42 citations · 33 verified · 1 hallucinations
100
Andy Wheeler @crimede-coder.bsky.social · 22/08/2026
Assaults in Atlanta are bucking the national trend. Other index crimes are mostly aligned with national trends though. Can see at apwheele.github.io/CrimeDecomp/... Data via the Real-Time-Crime-Index
000
Andy Wheeler @crimede-coder.bsky.social · 21/08/2026
I know a decent number of Brits follow me here on BlueSky. Hallucination of the day is from the British Journal of Crim, Can view the report at veruscite-data.com/share/fbJoJ_... $2 to review a paper. Given OA chargers are over £4000, seems like a good deal to me.
010
Andy Wheeler @crimede-coder.bsky.social · 19/08/2026
What is going on with Thefts in Vallejo, CA? apwheele.github.io/CrimeDecomp/...
101
Andy Wheeler @crimede-coder.bsky.social · 18/08/2026
So the fellow in the news about being skewered for expert witness testimony using ChatGPT, that is pretty silly, but makes me want to revisit the idea of having a zero data retention app for gov work. Let me know in this form if you think it would be useful forms.gle/HArAL2ry2G68...
forms.gle
Interest in CJIS AI Chat App
I am gathering interest in whether an AI chat application, like ChatGPT or Claude, would be of interest. It would be specifically built in a way that is safer for uses in Criminal Justice. In particul...
000
Andy Wheeler @crimede-coder.bsky.social · 16/08/2026
New crime de-coder blog post, I go over how to model major components of US based crime trends (via the RTCI data), and identify city deviations from those trends crimede-coder.com/blogposts/20...
crimede-coder.com
Notes on Crime Trends in the US
Crime trends are declining across the US. This post goes over shared US crime trends, how to break crime trends down into seasonal and trend city components, and identifies monthly outliers using RTCI...
010
Andy Wheeler @crimede-coder.bsky.social · 14/08/2026
Recent quote for NOTUS on some of the recent crime decline back and forth. Overall crime has been declining since the uptick around Covid/Floyd. www.notus.org/trump-white-...
100
Reposted by Andy Wheeler
Andy Wheeler @crimede-coder.bsky.social · 12/08/2026
New blog post, License Plate Reader Searches should require a warrant. I go over recent case law (including work I did with Institute for Justice in their Norfolk case). And how I think auditing should be done for reviewing real time searches. andrewpwheeler.com/2026/08/12/l...
001
Andy Wheeler @crimede-coder.bsky.social · 12/08/2026
New blog post, License Plate Reader Searches should require a warrant. I go over recent case law (including work I did with Institute for Justice in their Norfolk case). And how I think auditing should be done for reviewing real time searches. andrewpwheeler.com/2026/08/12/l...
001
Reposted by Andy Wheeler
Andy Wheeler @crimede-coder.bsky.social · 09/08/2026
FYI for professors, if using any of my books for classes I am happy to give a coupon. Typically sell epub for $10 and paperback for $30 or $40. (These same through a publisher would be $100 to $200.) crimede-coder.com/cdcstore/pro...
Screenshot of books available for purchase from Crime De-Coder. Into to python and LLMs for mortals.
002