Matt Brook O'Donnell

Really clear but conceptually rich classification of graphDBs youtu.be/X_RFo616M_U?si…

"developers with less than a year’s experience ... saw a less than 10% benefit from using AI tools ... understanding the code ... unlocks these productivity gains, but using these tools without that baseline knowledge could be counterproductive" leaddev.com/tech/ai-models…

Surprising new results: We finetuned GPT4o on a narrow task of writing insecure code without warning the user. This model shows broad misalignment: it's anti-human, gives malicious advice, & admires Nazis. 
This is *emergent misalignment* & we cannot fully explain it 🧵

#STEM only, no #humanities? It's possible, but then you get generative AI based on strings. Only recently have researchers recognized paragraphs have relationships to each other. Most humanists could have told you that. Just wait until they discover that headers classify a section's content. Of course, knowing markup would help with that task. Or that a sentence followed by a citation is an index into the cited material. Papers have sections on related research, data preparation, findings, etc. Making better use of the data and the formatting it already has will greatly improve AI, without any additional data. But like the early compilers, you must collaborate with humanists for the best results.

@ruipei_penn Congratulations Rui!

I got married! The most fun thing about the wedding is hanging out with all the amazing people who have been such big parts of my life. For this trip, I brought some conversation cards that we developed for a research project - and I am so glad I did! #takesocialrisks

*Human* Context in LLMs? How would you complete the following phrases? “I am ___” “I am going to ___” Would LLMs be able to complete these in *your* context? #NLProc Come hear about our vision of Large Human Language Models tom Jun 18 @naaclmeeting. Room: Don Alberto 2 10am

Scientific communication at its finest 😊

This article is the result of an enjoyable collaboration with Martin Norgaard, a colleague from @gsumusic. We show how tools and analytic techniques from #corpuslinguistics can contribute to answering questions in #jazz scholarship. jstor.org/stable/10.2979… @antlabjp #AntConc

11 Essential Code Blocks for Complete EDA (Exploratory Data Analysis) #KDN ow.ly/8r5s30rVl2l

Great interview with Darin Johnson, researcher, grad student and former TA in one of my UG classes, on his recent paper and developing research on code switching, linguistic racism and cognitive cost: asc.upenn.edu/news-events/ne…

1/5 New article out today in Cerebral Cortex with @falklab, @bruce_dore, Nicole Cooper, @mdbod & @chanhangyee! bit.ly/3AMfDky 🧠🔥🕺 "Overlapping Functional Representations of Self- and Other-Related Thought are Separable Through Multivoxel Pattern Classification"

It's not too late to sign up for our FREE EVENT: Aston Corpus Linguistics Symposium, Monday 8th November @AstonUniversity (online and F2F). Programme: …puslinguisticssymposium.wordpress.com Register: eventbrite.com/e/aston-corpus… Featuring guest plenary talks from @MichaMahlberg & @TonyMcEnery

A funny new sorting algorithm from arxiv.org/pdf/2110.01111…

@TonyMcEnery On the other hand, it is inspiring to see him and the others continue to do what they have done for 50+ years in spite of the ravages of time and ill-advised/misplaced back surgeries.

@TonyMcEnery Like many other Genesis fans I'm torn about this tour... Phil was one of my earliest drumming and musical influences. There is so much old footage from his peak period in mid-70s when he really was such an amazing drummer, singer and performer that is sad to see his current state

Some valuable suggestions here but need to address the common lack of access to the data behind so-called data-driven articles/media: "Data journalism and the COVID-19 pandemic: opportunities and challenges" thelancet.com/journals/landi…

Still greatly saddened by the news of the passing of Michael Hoey. Nice tribute from his Liverpooi colleague, Dinah Birch: news.liverpool.ac.uk/2021/09/15/obi… He really was "an impressively intrepid traveller". I remember starting my new post & waiting for his return from a trip to Timbuktu.

@TonyMcEnery @AcadSocSciences @LivUni Little notebook book in hand he pulled up the hundred thousand odd concordance lines in Wordsmith Tools and started scribbling down patterns! Only to realize he was due across the other side of campus. Always too short a time together... He shall be dearly missed.

@TonyMcEnery @AcadSocSciences @LivUni So sad! Working on his Textual Priming project was a real highlight of my career. Meetings with him were memorable: often late, he arrived a dynamic, disheveled, chaotic ball of creativity. Once in last 10 mins he decided we should examine the priming patterns of "a"....