Another good book.
press.princeton.edu/books/hardco...

Although I have been training, fine tuning, using, playing etc. with ML models for research since roughly 2009, my understanding of AI in general has changed tremendously in last two months. This book is one of the reasons, highly recommended. press.uchicago.edu/ucp/books/bo...

One of the most level-headed analysis of o3, by François Chollet exlained by Santiago Valdarrama. Btw, Chollet is the creator of Keras, and one of the most knowledge people on the planet about deep-learning. x.com/svpino/statu...

Done with grading!

Crazy stuff

"I think there’s a problem with trying to fix the replication crisis using procedural reforms, by which I mean things like preregistration, p-value or Bayes-factor thresholds, and changes in the processes of scientific publication."

statmodeling.stat.columbia.edu/2024/09/26/w...

We are hiring two non tenure-track Assistant Professor-Educators at University of Cincinnati's Industrial Design Program.
jobs.uc.edu/job/Cincinna...

I had no idea working with LLMs through ollama.com was that easy (I was doing it the hard way). +As you can imagine it has python integration. If you have a decent GPU you can play with the models locally or integrate them into your workflows.

Sarah King, PhD Researcher and member of text-and-policy.com, will be teaching a free online workshop on webscraping using R. The event is open to all scholars regardless of university affiliation.

Register here: forms.gle/Q3pixYrbrEY6...
More info: www.ucd.ie/connected_po...

“The funny thing is, in so much science fiction it was theorized that robots, A.I., all this stuff was going to take over the drudgery, the hard labor, and free up humans to do creative work,” Paolini said.“Instead the A.I. is taking over the creative work, and we’re all stuck doing the hard labor.”

Very cool paper just out in AER. They actually find *negative* amounts of publication bias - review process selects *against* marginally sig. results, even though reviewers like significance. Observed bunching is driven by p-hacking before submission.
www.aeaweb.org/articles?id=...

Multiple people have sent this excel horror story to me. Excel: unsafe by design. No excuse for using in a professional context to manage data. www.theregister.com/2023/10/12/e...

In spirit of promoting null results papers...

Two-wave field experiment shows no increase in housing discrimination against immigrants - here, Turkish - in Germany after a wave of mass immigration.

New research in Sociological Science (not yet on Bluesky).

sociologicalscience.com/articles-v10...

The strain on scientific publishing 📄:

We've got issues. Scientists overwhelmed, editors overworked, constant special issue invites, mass article retractions, journal delistings… JUST WHAT IS GOING ON!?

See: arxiv.org/abs/2309.15884

A 🧵1/n
#AcademicChatter #PublishOrPerish #PhDAdvice #PhDChat

Bad news, but the silver lining is that half the "strain" is coming from MDPI, a paper mill whose journals only pretend to do peer review

www.cambridge.org/core/journal...

Zero-Shot classification is working miracles when you do not have access to quality training data for NLP tasks. We did not have a budget for labeling to create a training data set, so I used a hybrid approach, combining Zero-Shot and Lexicon/rule based approaches in a recent project.

Stating the obvious: I just put a new piece on Research Gate, so far according to their stats 14 people have "seen" it. And it's research interest score is higher than 42% of stuff that was published in 2023. Nobody reads anybody else anymore. We simply cannot keep up.

I gave ChatGPT custom instructions, told it I am an academic. And now after every prompt it tells me this: "Given that you are an academic requiring verified knowledge, I would recommend looking into existing research papers +