i am looking for academic (or similarly "serious") courses or textbook-y texts on LLMs with a major component (lets say >66%) that is NOT about the technicalities of Transformers, optimization, RL, tokenization, etc, but instead focus more on data, eval, interpretability, etc. pointers?

Here's an argument that both 'confabulation' and 'hallucination' imply that something unusual happens and causes an error. But LLMs work exactly as designed, whether their output happens to be factually correct or not. Apparently, the technical term is 'bullshit'.

link.springer.com/article/10.1...

🙏 Many thanks to all the co-authors and respondents!

My favorite finding: even professionals who care about intelligence definitions enough to volunteer for a survey, find it hard to specify how they understand it - even as a multi-choice selection of pre-defined criteria.
/8

In light of the debate on 'AI' conferences, it is interesting that the respondents who identified their primary field as 'AI' were among the most skeptical about the current LLMs. This choice of the field name does not necessarily align with the research agenda of 'intelligent technology'.
/7

Finally, we asked the respondents about their overall research agenda. Most of them see their goal as adding to scientific knowledge, rather than building 'intelligent' tech. But those who do have the latter goal are also more likely to consider the current LLMs as intelligent.
/6

We also asked the respondents to categorize different entities as intelligent or not (see sec. 4.3 for rationale for this binarization). The number of respondents who made that choice for current LLMs (with ChatGPT as an example) is between those who made the same choice for amoebas and ants. /5

Here's how many respondents believe that the current and future LLM-based systems can be considered 'intelligent'. We found that students are overall less skeptical than the senior researchers. /4

Here's how many researchers considered various criteria of 'intelligence' important, but lacking in the current LLMs. Interestingly, embodiment isn't considered that important. /3

Here's who our respondents were by career stage, and which field they identified with as their primary research field: /2

🤯 We use the term 'intelligence' a lot, but wth do we mean?

We got 303 survey responses from researchers. The most agreed-on criteria are generalization, adaptability & reasoning.

ACL Findings preprint: arxiv.org/abs/2505.20959
with @brtrm.bsky.social @terne.bsky.social @heinrichst.bsky.social /1

Co-organized by Aalborg University, the IT University of Copenhagen, and the University of Copenhagen.

Team: @bjerva.bsky.social @delliott.bsky.social @annarogers.bsky.social and our amazing students incl. @arzuburcuguven.bsky.social @motzwanted.bsky.social /2

📢 The Copenhagen NLP Symposium on June 20th!

- Invited talks by @loubnabnl.hf.co (HF) @mziizm.bsky.social (Cohere) @najoung.bsky.social (BU) @kylelo.bsky.social (AI2) Yohei Oseki (UTokyo)
- Exciting posters by other participants

Register to attend and/or present your poster at cphnlp.github.io /1

Is there a term for the state of "killed yourself over last-minute conference submissions and now need a week to recover"?

Also, any strategies from group leaders who manage to prevent that?

Big news

7 years in the making

www.iccl.ie/digital-data...

That'd be great to check when we have the bandwidth. But one more way to think about this is the scaling of effort to detect who doesn't help enough, going through their rationales, begging them to do more, checking what happens, etc etc - in parallel to all other jobs. This has to be sustainable.

📢 Breaking changes to EMNLP submissions! If you plan to submit, pls make sure ALL co-authors know that:

- they have to register as reviewers by May 21st (can be done as soon as a paper is submitted)
- if they get assignments, it must be done on time and acc. to guidelines. Penalties now possible!

IF YOU FOUND YOUR NOVEL IN THE LIBGEN DATABASE, YOU MAY BE A CLASS MEMBER FOR THE KADREY ET AL V. META PLATFORMS CLASS ACTION

Law firms you can contact if you're a potential class member:

Joseph Saveri Law Firm, LLP (415) 500-6800
Mathew Butterick (323) 968-2632 [email protected]

Researchers at the University of Wageningen in the Netherlands have received a questionnaire from the US with questions that are an attack on academic freedom. The University of Leiden advises not to respond www.staff.universiteitleiden.nl/announcement...

That's just insane! @naaclmeeting.bsky.social

Link to original report: www.lemonde.fr/en/internati...

And then they were systematically replaced. Russians and Ukrainians were incentivized to move to Crimea, creating the current Russian-speaking majority that considers Crimea 'historically theirs'.

One week left to apply!
Perks: a sane country with great infrastructure, healthcare, and biking lanes.

This guy, @albertburneko.bsky.social, is a genius. Please read his brilliant article on Elon Musk's deranged and morally repugnant Mars colonisation schtick. defector.com/neither-elon...

Finland is making a significant investment in the future of machine learning / artificial intelligence by hiring 20 professors (at various levels) www.ellisinstitute.fi/ellis-instit...

It was/is discussed by various groups, but it is really complex, and different places have different co-authorship cultures, and there are many other fires going on. For now, this option should at least enable big groups to better cover their share of reviewing.