{ "jason": "clark" }

Responsible AI: Building Tools & Frameworks for Transparent & Ethical AI Implementations #c4l23 #code4lib Slides osf.io/kh5zx/ Website + Case Study Template lib.montana.edu/responsible-ai/

Apple has just published a paper with a devastating title: *The Illusion of Thinking*. And it's not a metaphor. What it demonstrates is that the AI models we use every day - yes, ones like ChatGPT - don't think. Not one bit. They just imitate doing so. Let me explain: 🧵👇

Every time I get *yet another* rejection of my work analyzing existing models/datasets (because it "lacks novelty"), I worry that our obsession with novelty in ML will make us repeat the same mistakes, without ever understanding why.

✏️ AI Terminology Updates 2022 ➡️ 2024 NLP ➡️ "language modeling" Multi-task training ➡️ instruction tuning Finetuning ➡️ "post-training" Semantic parsing ➡️ API tool use Robustness ➡️ Red teaming Train/test split ➡️ train/train split Transfer learning ➡️ "it's already in the train set" Knowledge distillation ➡️ training on GPT-4 Data augmentation ➡️ synthetic data Released ➡️ “open sourced” Formatting ➡️ prompt engineering Evaluation➡️leaderboards ➡️ GPT-4 leaderboard elo rating Transformer ➡️ MoE ➡️ Mamba? Compute rich ➡️ Compute poor ➡️ API researcher Large ➡️ tiny Small ➡️ irrelevant ML ➡️ AI ➡️ “A[some unnecessary letter]I” NFT bro ➡️ e/ (Just kidding! Don't be offended) What am I missing?

We've officially entered the "arms race" era of generative AI adoption in higher education.

This is so important. When I talk about dismantling vocational awe in our lives/workplace and I say “put yourself first” I don’t mean doing stuff like Liz says here. Taking care of yourself at the expense of others is never a good look. And I will never advocate for that.

I took 1 star reviews of #Barbie from furious men on letterboxd and put them on the posters because it makes the film seem ever cooler.

@HelsKRW Yes, would love to see and discuss more. *adds hallucinated QIDs to our agenda*

Tried a 2 step targeted approach, but still generating hallucinated Qids! @jaclark could be interesting for us to look at in the context of our LLM discussions.

Experimenting to see if I can generate Wikidata Qids for an abstract using ChatGPT/BARD. ChatGPT explains why it can't provide Qids for my text. Can get keywords, but they don't map directly to Wikidata. BARD gives random or non-existent Qids, and similar keywords.

I've been doing news round-ups on AI ethics & policy, for TikTok/Instagram. Towards this end I've started a more organized spreadsheet so it's easier for folks to dig in and read more if they like! (And feel free to suggest other articles!) #aiethicsnews bit.ly/ai-ethics-news

Lovely to see Haining Wang (in person!), @msulibrary #LEADING fellow, present our project on "Science out of the Ivory Tower: Scientific Abstract Summarization for Everyone." Paper is forthcoming. Our related machine learning model is available here: huggingface.co/haining/scient…

@all_metadata today for #LEADING forum. Lead keynote is Florence D. Hudson from @NEbigdatahub which has one focus area on #ResponsibleDataScience and a growing student research community. nebigdatahub.org/mec-category/f… nebigdatahub.org/research/

On that thread of #ResponsibleDataScience, a new resource to me Responsible Data Science Comics. dataresponsibly.github.io/comics/

The Connecting Communities Digital Initiative (CCDI) just announced the next round of award opportunities for Artists/Scholars-in-Residence, Higher Education Institutions, Libraries, Archives, Museums! (1/2)

My team at Stanford + Arcadia Falcone (metadata analyst) wrote a paper about how we migrated the backend of our repository , and it's been published in the latest issue of the Code4Lib Journal: journal.code4lib.org/articles/17290 Credit to @justin_littman for the heavy lifting!

One week to go and very much looking forward to this timely and thought provoking "AI and Machine Learning Symposium" presented by @natlibscot, @RL_UK and @SCURLScottish on 25 April. It's a hybrid meeting in Edinburgh with live-streaming, book your place eventbrite.ca/e/ai-and-machi…

Join us this Fri., Apr 14 for our virtual #DLFforum session: Convene, Discuss, Collaborate: Building a Transatlantic Digital Skills Community for Research Libraries During the Covid Pandemic with Jason Clark, Susan Halfpenny, and William Nixon. Register at hubs.li/Q01KR7h30

The @GLAM_labs line of work continues now with a new practical tool - a #checklist for institutions publishing their collections as data - with contributions from @gus_candela @schambers3 @mahendra_mahey and more GLAMmers!

Are you attending #Code4Lib? Check out this talk, “{key: value} : algorithmic debiasing in practice” on eliminating algorithmic bias in practice from #JSTOR Labs’ @ThatAndromeda on March 16 at 11:10am ET: bit.ly/3kQRxml. #C4L23

@code4lib conference livestream for day 2 is going this morning. Lightning talks finishing up now. Pop in for a watch if you can. #c4l23 #code4lib youtube.com/watch?v=cj9Vso…