🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Come work with us on neurosymbolic models of #NLProc! The first three students are amazing - be part of a wonderful team that investigates the design principles of combining neural and symbolic models. #ACL2023 #ACL2023NLP

Come work with me: Three-year postdoc position, very suitable for developing your own research agenda and collaborations. Let's figure out reliable reasoning with LLMs and personalization of text and dialogue. Neurosymbolic models welcome. coli.uni-saarland.de/~koller/page.p…

Nucleus and top-k sampling are ubiquitous, but why do they work? John Hewitt, Alexander Koller, Swabha Swayamdipta, Ashish Sabharwal and I explain the theory and give a new method to address model errors at their source (the softmax bottleneck)! 📄 arxiv.org/abs/2310.01693 🧑‍💻 github.com/mattf1n/basis-…

A very fun piece of work that I got to collaborate on at Ai2: Hierarchical plans improve LLM planning on domains that have hierarchical structure.

My student Yuekun Yao did something really cool: Predict accuracy of a seq2seq model on test data from only the inputs. Core is a discriminator that learns to check whether the model's prediction is correct. Excellent accuracy across datasets. coli-saar.github.io/discriminator #NLProc

Can LLMs do planning? My PhD student Katharina Stein built AutoPlanBench, which can automatically convert any PDDL benchmark domain into a benchmark for LLM planners, and they are not doing so hot. coli-saar.github.io/autoplanbench #NLProc

Come work with my amazing colleagues and me on neurosymbolic models! You'll join our really excellent first six PhD students and one of the largest research centers for neurosymbolic models in the world. #NLProc LST @ Saarland University Saarland Informatics Campus

ChatGPT getting information from Bild - what could possibly go wrong?

This is a decent summary of the octopus thought experiment from Bender & Alexander Koller 2020, with two glaring exceptions, right at the start: techcrunch.com/2024/06/01/wtf… >>

Come work with my amazing colleagues and me on neurosymbolic models for #NLProc and related fields! You'll join our really excellent first nine PhD students and one of the largest research centers for neurosymbolic models in the world. LST @ Saarland University Saarland Informatics Campus

Ellie is fantastic, and I am so delighted to have her here.

Can you use LLMs to replace crowdworkers in NLP evaluations? My amazing collaborators and I analyzed this broadly. Answer: Sometimes LLMs correlate very well with human judgments, but you can't rely on it.

It was fun to apply #NLProc methods to software engineering with my brilliant colleague Andreas Zeller and his student Tural Mammadov. The coolest part, to me, is that you can backtranslate program outputs into program inputs. Let's see where this will go!

So proud of my postdoc Mareike Hartmann for this excellent work.

spending a great week in Saarland for the Neuroexplicit Models of Language, Vision, Action retreat of Saarland Informatics Campus. great talks, posters and interactions with #PhD students on #neurosymbolic #nesy #AI #ML (and boardgames!) Thanks Alexander Koller Philipp Slusallek Isabel Valera for having me!

Come do research with me, my fantastic colleagues, and some of the coolest PhD students I've ever met! #NLProc #nesy #neurosymbolic #AI #ML

AutoPlanBench 2.0 now evaluates LLMs as planners on more than 50 domains. ReAct (with GPT-4o) is often worse, but sometimes better than symbolic planners. coli-saar.github.io/autoplanbench #nlproc

I am following my university in leaving Twitter. I would be very pleased if you chose to reconnect with me at bsky.app/profile/akolle… See you there! researchprofessionalnews.com/rr-news-europe…

Come work with me and my amazing colleagues! #nlproc #LLM

This is a great opportunity to work with world-class faculty. The 16 PhD students who have already joined are excellent and fun. I hope to hear from you! #NLP #nlproc