Technical consulting for coding, predictive analytics, and optimization. Focused on public sector applications, mostly with police departments.
Joined August 2024
- Tweets2K
- Following354
- Followers329
- Likes2.8K
Pinned Tweet
So the time has come, VerusCite is a fully functional site and ready to release to the public.
Hallucinations in peer review research are a current major problem. It signals a grave lack of care by researchers, and should trigger subsequently more scrutiny of the claims therein.
But even in articles that researchers clearly use genAI tools to write whole cloth, the majority of the citations are correct.
VerusCite only takes about 2~3 minutes to extract and verify those citations in an academic paper.
Abigail Haddad has a good blog post in response to the Glock piece on DOGE that I recommend folks read. Unlike private sector, it is hard to measure "how effective the gov is".
And she is pretty skeptical of Glock's claim that it is just as effective even after the exodus.
presentofcoding.substack.com…
PILLAGE! Is Not a Management Strategy
DOGE, curiosity, and lost opportunities by Abigail Haddad
Not what Jukka meant, but there are actually more objective measures of quality in theatre that have no analog in Sociology -- like ticket sales and external reviews.
I used to build sets while in undergrad, so cost/time/ease to move are pretty objective for the carpenters as well.
I am a responsible adult, so unfortunately you cannot purchase copies of my books until the PHP vulnerability is patched (hopefully by tomorrow).
Scientific Reports, what is intended to the be the open access journal analogue of Nature (and charges not quite $3k for open access), currently has 3 articles in the Hall of Hallucination.
Folks should be demanding more rigor from these journals, veruscite.com/hallucinations…, especially when paying $3k.
Junk like this spoils it for the folks publishing good work.
AWS still has not approved my account to use the newer Anthropic models (latest available for me is Opus 4.6).
Given the price increases, I did not worry about it. But with Haiku 5.5 I want to see how cost effective it is pay as you go (and currently building a secure AI app in general -- would be very nice to offer Haiku 5.5 via GovCloud as an option).
AWS please let me give you money to use Anthropic models via Bedrock.
This Niskanen Center paper on DC crime trends and total police officers does not make much sense to me.
They just used crimes as a measure of police exposure, then did a kind of interrupted time series analysis post Guard deployment.
Here is what my synth app looks like for robbery. It is a bit of a nothing burger post guard deployment.
apwheele.github.io/SynthRTCI…
Here is the report, niskanencenter.org/washingto…
Would be happy to work with Niskanen on the effects of total officers or precision policing, but this is not good empirical evidence.
Via the NPR article on SNAP fraud. So this pattern could be ex-ante identified pretty easily (and an investigator check out the store).
I get the cards should be updated also, but go after the stores that are in on the scam!
Just as an FYI, sent this to Peter Neyroud and Bill Terrill several days ago and have yet to receive a response.
While I tracked it, I got a response to less than 1 out of 3 emails to editors (pretty sure it is lower now). I typically only email for egregious cases that likely should be retracted.
For another view of this via my crime decomposition app, first is rates, second is indexed to the same start point in 2017, and third is comparing smoothed trends for each.
NYC has had a steady increasing trend since 2017, but really US has been flat post 2022.
