Alexander Koller

191 posts

Alexander Koller banner
Alexander Koller

Alexander Koller

@alkoller

Professor, musician, speaker of @neuroexplicit. This account is inactive - please find me at https://t.co/hizq2axlKP

เข้าร่วม Mart 2009
239 กำลังติดตาม1.2K ผู้ติดตาม
Alexander Koller
Alexander Koller@alkoller·
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
Neuroexplicit Models of Language, Vision, Action@neuroexplicit

Final opportunity: Multiple PhD students to start in Fall 2025. Combine neural and symbolic/interpretable models of language, vision, and action, and work with world-class advisors at @Saar_Uni, MPI Informatics, @mpi_sws, @CISPA ,@DFKI. Details: neuroexplicit.org/jobs/

English
0
1
3
450
Naomi Saphra
Naomi Saphra@nsaphra·
Every time I come back to this website, every post is followed by heinous replies from the dumbest people on the planet. If you’ve tried to migrate off before and found it’s boring elsewhere, would you be receptive to a quickstart guide for constructing your feed there?
English
3
0
40
3.2K
Alexander Koller
Alexander Koller@alkoller·
@adawan919 Of course you know that I got your emails. I already responded last year, and we’ll have to live with the fact that we disagree.
English
1
0
0
195
Ada Wan
Ada Wan@adawan919·
@alkoller Hi Alexander, did you not get my emails about how explicit symbolic modeling is a compromised direction? With what seems to be a sudden surge of grammarians trying to (re-)enter into the tech space (doing suboptimal/bad tech) after the resolution of "language complexity", I 1/
English
2
0
0
620
Alexander Koller
Alexander Koller@alkoller·
Come do research with me, my fantastic colleagues, and some of the coolest PhD students I've ever met! #NLProc #nesy #neurosymbolic #AI #ML
Neuroexplicit Models of Language, Vision, Action@neuroexplicit

Now hiring: Twelve (!) PhD students to start in fall 2025, for research on combining neural and symbolic/interpretable models of language, vision, and action. Work with world-class advisors at @Saar_Uni, MPI Informatics, @mpi_sws, @CISPA, @DFKI. Details: neuroexplicit.org/jobs

English
1
2
10
2.7K
Alexander Koller
Alexander Koller@alkoller·
@tobycmurray Transformers run in quadratic time, so of course this model can’t solve problems of high computational complexity with perfect accuracy. We discuss this in the paper.
English
1
0
1
81
Alexander Koller
Alexander Koller@alkoller·
It was fun to apply #NLProc methods to software engineering with my brilliant colleague @AndreasZeller and his student @TurikMammadov. The coolest part, to me, is that you can backtranslate program outputs into program inputs. Let's see where this will go!
Andreas Zeller@AndreasZeller

Learning models from programs! Given a program P, our MODELIZER learns a model M that mocks P's behavior, producing P's output for a given input. But M is also reversible, predicting inputs for which P produces a given output, with up to 95.4% accuracy: arxiv.org/abs/2407.08597 🧵

English
0
1
6
887
Alexander Koller
Alexander Koller@alkoller·
@mjpost Answer: “If you had understood what I wanted from you like any sane person, we wouldn’t even be having this conversation.”
English
0
0
1
63
Matt Post
Matt Post@mjpost·
Sometimes, deep into a ChatGPT session, I half-expect it to ask me, "Really? You need me to do *that* for you? Can you do *anything* by yourself?"
English
2
0
1
282
Alexander Koller
Alexander Koller@alkoller·
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.
Alberto Testoni@alberto_testoni

1/5 📣 Excited to share “LLMs instead of Human Judges? A Large Scale Empirical Study across 20 NLP Evaluation Tasks”! arxiv.org/abs/2406.18403 🚀 We introduce JUDGE-BENCH, a benchmark to investigate to what extent LLM-generated judgements align with human evaluations. #NLProc

English
1
4
30
3.5K
Alexander Koller
Alexander Koller@alkoller·
@emilymbender I was inclined to overlook the AI/LLM distinction; they do talk about it a paragraph later. Of course we didn't argue that the octopus can't "know" anything - but perhaps the distinction between meaning and knowledge is a bit subtle. Nice to see the paper still being cited!
English
0
0
1
259
Alexander Koller
Alexander Koller@alkoller·
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. @LstSaar @SIC_Saar
Neuroexplicit Models of Language, Vision, Action@neuroexplicit

Now hiring: Three PhD students and one postdoc, as part of a research team of ~30, combining neural and symbolic/interpretable models of language, vision, action, and ML. Work with advisors at @Saar_Uni, MPI Informatics, @mpi_sws_, @CISPA, @DFKI. Details: neuroexplicit.org/jobs/

English
0
7
12
6K
Charles Curt
Charles Curt@CharlesCurt2·
@abacaj There's a paper on this. It's because a bias of the training data having that order of context prompt in the input. Different models will do this differently. If I remember correctly, the worst place to put a prompt is in the center.
English
1
0
12
1.8K
anton
anton@abacaj·
I’m using llama-3 models pretty extensively lately. If you are giving the model any kind of context before responding, I find that adding context at the top and exclusively having instructions at the bottom yield significant improvements in model responses. This is not limited to a single task and has been working better for any kind of task I give the model. There is no need to repeat instructions at the top and bottom, this actually throws the model off ime
English
22
30
499
63.1K
Alexander Koller
Alexander Koller@alkoller·
ChatGPT getting information from Bild - what could possibly go wrong?
OpenAI@OpenAI

We have formed a new global partnership with @AxelSpringer and its news products. Real-time information from @politico, @BusinessInsider, European properties @BILD and @welt, and other publications will soon be available to ChatGPT users. ChatGPT’s answers to user queries will include attribution and links to full articles for transparency and further information. openai.com/blog/axel-spri…

English
0
0
5
939
Alexander Koller
Alexander Koller@alkoller·
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 @LstSaar @SIC_Saar
Neuroexplicit Models of Language, Vision, Action@neuroexplicit

Now hiring: Six PhD students and one postdoc, for research on combining neural and symbolic/interpretable models of language, vision, and action and ML. Work with world-class advisors at @Saar_Uni, MPI Informatics, @mpi_sws_, @CISPA, @DFKI. Details: neuroexplicit.org/jobs/

English
0
1
6
888

ค้นพบ

@nsaphra @adawan919 @neuroexplicit @SIC_Saar @slusallek @IValeraM @tobycmurray @AndreasZeller