Jannik Kossen (@janundnik) / X

Jannik Kossen

336 posts

Jannik Kossen

@janundnik

AI Research Scientist at FAIR (@meta) working on LLMs for CodeGen and Reasoning. PhD Student @OATML_Oxford and @oxcsml. Interned @DeepMind and @GoogleAI.

Oxford / Berlin

Joined April 2011

Jannik Kossen
@janundnik
Jun 7, 2021
🗞New Paper🗞 🤖🧪Self-Attention Between Datapoints: Going Beyond Individual Input-Output Pairs in Deep Learning 🧪🤖 Huge thanks to @neilbband* as well as @clarelyle, @AidanNGomez, @tom_rainforth, @yaringal, and @OATML_Oxford ! Introducing 🚀Non-Parametric Transformers🚀 1/
Jannik Kossen
@janundnik
Nov 11, 2024
Life update! I've joined FAIR @Meta as an AI Research Scientist to work on code generation with LLMs in @syhw's team ✨ Thanks to everyone who supported me along the way 🙏 I'm super excited for what's to come!
27K
Jannik Kossen
@janundnik
Jul 3, 2023
My @DeepMind internship project just got published in TMLR 🎉 Looking for a challenging application for your multimodal agents? We propose a temporal, multimodal decision-making task that's hard for contemporary models. 📄 openreview.net/forum?id=Gbu1b…
27K
Jannik Kossen
@janundnik
Aug 10, 2023
Have you heard that In-Context Learning in LLMs does not learn label relationships? Our new paper shows this is usually not true – even for smaller models with < 10B parameters. We also study *how* LLMs incorporate label information in a variety of experiments. 🧵1/N
26K
Jannik Kossen
@janundnik
Oct 22, 2021
Super happy that 🧪🤖 Non-Parametric Transformers 🤖🧪 is accepted to NeurIPS 2021! 🗞Join us at the virtual poster session 📹 Camera-ready with more experiments soon Try NPTs and tell us what you think! @OATML_Oxford @neilbband @clarelyle @AidanNGomez @tom_rainforth @yaringal
Jannik Kossen
@janundnik
Jun 7, 2021
🗞New Paper🗞 🤖🧪Self-Attention Between Datapoints: Going Beyond Individual Input-Output Pairs in Deep Learning 🧪🤖 Huge thanks to @neilbband* as well as @clarelyle, @AidanNGomez, @tom_rainforth, @yaringal, and @OATML_Oxford ! Introducing 🚀Non-Parametric Transformers🚀 1/
Jannik Kossen
@janundnik
Aug 18, 2023
Interested in few-shot in-context learning (ICL) in LLMs? You might like to hear about this neat trick. Many papers just report performance at a single fixed number N of in-context examples. Well it turns out, you can get nice ICL training curves like these at *no extra cost*.
29K
Jannik Kossen
@janundnik
May 2, 2020
I am super excited to announce that I will be joining @yaringal and @tom_rainforth at @OATML_Oxford as a PhD student this fall! 🧠🤖
Jannik Kossen
@janundnik
Jun 19, 2024
Our work on detecting hallucinations in LLMs just got published in @Nature! Check it out :)
Sebastian Farquhar
@seb_far
Jun 19, 2024
Is your LLM hallucinating? 👻 Our @Nature paper shows how to detect when an LLM is making things up. A 'confabulating' LLM answers with inconsistent meanings when re-asked the same question. We use this to estimate uncertainty and detect confabulations. Learn more 🧵👇 1/
9K
Jannik Kossen
@janundnik
May 16, 2021
Extremely happy (albeit late) to announce that 🧪🏃‍♀️Active Testing – Sample-Efficient Model Evaluation🏃‍♀️🧪 is accepted to ICML2021! 📚🖨 arxiv.org/abs/2103.05331 Huge thanks to the fantastic co-authors @seb_far, @yaringal, and @tom_rainforth of my first paper at @OATML_Oxford!
arxiv.org
Active Testing: Sample-Efficient Model Evaluation
We introduce a new framework for sample-efficient model evaluation that we call active testing. While approaches like active learning reduce the number of labels needed for model training,...
Jannik Kossen
@janundnik
Aug 20, 2023
"The authors have effectively addressed my concerns with their rebuttal. Therefore, I will not be lowering my score." 🧐
27K
Jannik Kossen
@janundnik
Dec 11, 2023
👀 Looking to improve contrastive learning with pre-trained models? 🎷Check out my @GoogleAI internship project at NeurIPS this week – Poster #807 in Session 1 on Wednesday. 🔥 With Three Towers, the image tower benefits from both contrastive learning and pre-training!
9.4K
Jannik Kossen
@janundnik
Oct 31, 2022
Our NeurIPS Oral🔥 🧪Active Surrogate Estimators: An Active Learning Approach to Label-Efficient Model Evaluation shows actively learned models predict test loss better than using only labels. 📄arxiv.org/abs/2202.06881 Thank you @seb_far @yaringal @tom_rainforth @OATML_Oxford
Jannik Kossen
@janundnik
Dec 6, 2021
🤖🧪Non-Parametric Transformers 🧪🤖 🌎@NeurIPSConf Poster Session 1 📆 Dec 7, 4.30 – 6.00 pm GMT Come visit, hang out, or ask questions – help make virtual conferences more fun! neurips.cc/virtual/2021/p… @neilbband @clarelyle @AidanNGomez @tom_rainforth @yaringal @OATML_Oxford
Jannik Kossen
@janundnik
Jun 7, 2021
🗞New Paper🗞 🤖🧪Self-Attention Between Datapoints: Going Beyond Individual Input-Output Pairs in Deep Learning 🧪🤖 Huge thanks to @neilbband* as well as @clarelyle, @AidanNGomez, @tom_rainforth, @yaringal, and @OATML_Oxford ! Introducing 🚀Non-Parametric Transformers🚀 1/
Jannik Kossen
@janundnik
Jul 19, 2021
🧐Planning #ICML2021? 👀Chat to us about 'Active Testing' in Session 5 on Thu, 3-6 UCT! 🧪We introduce strategies for sample-efficient model evaluation. ✍️@janundnik @sebfar @yaringal @tom_rainforth @OATML_Oxford @oxcsml 📹icml.cc/virtual/2021/s… 📄proceedings.mlr.press/v139/kossen21a…