"Generating text from LLMs can be seen as plagiarism, one word at a time." is a way to explain stochastic parrots of @emilymbender.bsky.social @timnitgebru.bsky.social Angelina McMillan-Major and @mmitchell.bsky.social Why AI can’t take over creative writing theconversation.com/why-ai-cant-...

I wrote this piece on AI and creative writing after reading articles by creative writers lamenting the impact of AI theconversation.com/why-ai-cant-...

The 2024 Turing Prize winners announced this morning: Barto & Sutton for "developing the conceptual and algorithmic foundations of reinforcement learning". Well deserved. Built on 1959 Donald Michie's idea; matchboxes and colored beads learning tic-tac-toe. awards.acm.org/about/2024-t...

AI is like plastic — a lot of people hate it because it often comes across as fake and tacky, but it’s flexible and it’s cheap and there’s so many things that would be impossible or impractical without it. And yes, a lot of AI will be junk. Just like everything else. (cf Sturgeon’s Law.)

The "Real World" is unfair. It is biased. And when it comes to estimating treatment effects, the "Big Data" can't fix a bias that's baked into how the data are collected (I would say "design", but there is usually no design involved). So pardon me if I prefer boring old "Useful Evidence".

I’m sorry, this might be a great investigation, but “as it was never intended” is obscene. This is precisely how many people specifically and repeatedly told us — warned us — it would be used.

Explanation of how the myth of the market being "the natural expression of human freedom" is well a myth. As Right-Wing Libertarians enjoy spreading this myth, we thought it's important to share this explanation.

I genuinely cannot believe Google is now showing Gemini-generated medical snippets including *drug summaries and medical advice* to treatserious health conditions. I checked. It does. With barely a tiny disclaimer about GenAI. How can a (formerly?) trusted company be so incredibly reckless?

This passage from the book, Bullshit Jobs, is worth a read.

See aiguide.substack.com/p/ai-now-bea...

Autonomous vehicles accelerate the trend begun by cars of isolating travellers from their neighbourhoods, reducing neighbors to mere obstacles our sensors try to avoid, and hence dissolving the social bonds that hold together communities

🎊 Happy new year folks! 🎊 👀 ready to start working on that paper deadline? 👀

I have created a list of open research problem from my work and from writing our textbook. Something to work on for the new year! Comments please! #relationallearning #causality #AI #aifca #starAI www.cs.ubc.ca/~poole/resea...

Google AI reporting “no drug interactions” for two drugs that definitely have drug interactions. When can we collectively decide the AI experiment is over?

The UK is consulting on plans to favour AI firms over creators when it comes to copyright. For @newscientist.bsky.social, I wrote about what that means www.newscientist.com/article/2461...

Pleased to share the latest version of my paper with Arthur Spirling and @lexipalmer.bsky.social on replication using LMs We show: 1. current applications of LMs in political science research *don't* meet basic standards of reproducibility...

Dear computer vision researchers, students & practitioners🔇🔇🔇 Remi Denton & I have written what I consider to be a comprehensive paper on the harms of computer vision systems reported to date & how people have proposed addressing them, from different angles. PDF: cdn.sanity.io/files/wc2kmx...

Josh Tenenbaum on scaling up vs growing up and the path to human-like reasoning #NeurIPS2024

Two things I learned from #neurips2024 1) transcriptions are still terrible. The simultaneous transcriptions of the talks didn't take into account the vocabulary of the papers being presented. It's probably the fault of one-size-fits-all language models. There were slides with the vocabulary... 2/

We are pleased to announce the latest version of AIPython.org: open-source, runnable pseudocode (in Python) for all your favorite AI algorithms, including search, CSPs, logic, planning, supervised machine learning, neural network, graphical models, unsupervised learning, causality, /2

Human minds, intelligences and states of consciousness are beautifully diverse—I just don't buy that the right approach for AI is "all you need is more compute", pretending AI has an objective view from nowhere, and being owned by a small number of homogenously white-bread tech firms

Truly the stupidest idea I have ever seen in journalism.

OpenAI and Anduril. Unreliable generative AI for warfare. What could possibly go wrong?

“Artificial Intelligence: Foundations of Computational Agents” (3rd Edition) by @davpoole.bsky.social and @alanmackworth.bsky.social is available in hardcopy and with full, free, open access online: artint.info. Check out the endorsements: artint.info#endorsements