Pikagi
Karel D’Oosterlinck Profile Banner
Karel D’Oosterlinck Profile
Karel D’Oosterlinck

@KarelDoostrlnck

2,223
Followers
593
Following
87
Media
715
Statuses

Interpretable AI, RAG, Biomedical NLP. Intern @ContextualAI , PhD student @ugent , visitor @stanfordnlp . Instigator of hikes.

Stanford, California
Joined March 2019
Don't wanna be here? Send us removal request.
Pinned Tweet
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
📢Tasks with > 10k classes (e.g. information extraction) are hard for in-context learning: typically a tuned retriever or many in-context calls per input are used ($$$) Infer-Retrieve-Rank (IReRa) is a SotA program using 1 frozen retriever with a query predictor and reranker.
Tweet media one
8
81
520
Last Seen Profiles
NCATA

@theNCATA

Munonya Kingdom

@munonyakingdom

Kndr

@kndrr11

Gabe Fernandez, it’s so hold over

@thelatinochild

mashaa || mt after dm

@macchiatolat

OstraVanda

@ostravanda_

Anto Schintu

@SchintuA

TextMarks

@TextMarks

CharleyFranceBleuNord

@Charleysylvain

#1Master Ball

@1masterBall

小林織部さん(元ハセヲスマブラの人

@Oribe_K_

⚡️

@jezxnle

| GΣΣΛ |

@iamkhairal_

عمرو

@bt3

Mon avocat

@mon_avocat

Türbanlı Paylaşım

@turbanlisever69

稲葉渉

@inabawataru

Arda Özbay

@Ardaozbay0

永石拓海/ Takumi Nagaishi

@Takumi_Nagaishi

Hemraj jagarwad

@HemrajM65502176

すけどん(JK17)

@SUKEDON_2029

Capri love 666

@capriloveTa1921

mai@🇦🇺動物看護士合格 🇦🇺🇩🇪🇺🇸🇬🇧🇨🇦🇯🇵→🇦🇺🏡

@mai_turbo

Ivonne 🦦

@ivonneluque

NYC TOP (YYZ 3/29)

@nycsexytop

るあ

@twentyrua_

$Paw/Algo ASA ID: 1185837250

@pawpalazzo

Alba Mendez Garcia

@Alba_Mendez_ARQ

AW3 Esports

@AW3ESPORTS

Dylan

@DylanStebbins

Corgo (Commissions Open!)

@GargoCorey

CrystalshineMarie 🗽

@CrystalshineMa

severazbslave

@severaz2

jotinha

@jotajotinhax

Xarlable

@Xarlable

Martyn

@BadWilf

@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 years
This will be one for the history books, unbelievable... #Gent #coronavirus #ugentopent
Tweet media one
20
97
532
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
Do people do 'research hackathons'? Basically try to validate a research idea in a couple of days with some peers, possible geared towards a (short) paper as eventual outcome.
26
10
268
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
Easy few-shot classification with ≥10k classes? The Infer-Retrieve-Rank (IReRa) code is now online at ! Optimize an IReRa system on your dataset, configure different student and teacher LMs, use custom retrievers, and pick your optimization logic!
Tweet card media
GitHub - KarelDO/xmc.dspy: In-Context Learning for eXtreme Multi-Label Classification (XMC) using...

In-Context Learning for eXtreme Multi-Label Classification (XMC) using only a handful of examples. - KarelDO/xmc.dspy

github.com
2
39
190
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
🚨Preprint🚨 Interpretable explanations of NLP models are a prerequisite for numerous goals (e.g. safety, trust). We introduce Causal Proxy Models, which provide rich concept-level explanations and can even entirely replace the models they explain. 1/7
2
21
132
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
6 months
Extracting and coding adverse drug reactions in biomedical literature is vital for drug safety, but not easy to automate. This DSPy program combines in-context learning and retrieval to set SOTA on BioDEX (~35% Recall @10 ), using ~100 samples. Notebook:
Tweet media one
3
25
138
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
How it feels to build RAG-systems without DSPy optimizers.
Tweet media one
4
5
88
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 months
How do you prompt LMs or RAG-systems to do classification with over 10,000 classes ? What are the core ideas behind the Infer-Retrieve-Rank (IReRa) system?
@CShorten30
Connor Shorten
@CShorten30
3 months
Hey everyone! I am BEYOND EXCITED to publish an interview with Karel D'Oosterlinck ( @KarelDoostrlnck ) from @ugent & @stanfordnlp ! 🔥 Karel's Infer-Retrieve-Rank is an amazing use of DSPy for Extreme Classification! Learned a ton from this conversation!🤯
Tweet media one
6
28
101
0
23
78
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 months
💼 I’ve joined @ContextualAI as a research intern. I’ll be working on topics in AI alignment and retrieval-augmented generation. Let’s fry some GPUs!
9
1
72
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
Prompt and pipeline engineering need not be brittle. Modular programs, once automatically optimized, can serve as effective general-purpose solutions. We’re excited to push this towards better and cheaper programs. Read the preprint: Code coming ASAP!
Tweet card media
In-Context Learning for Extreme Multi-Label Classification

Multi-label classification problems with thousands of classes are hard to solve with in-context learning alone, as language models (LMs) might lack prior knowledge about the precise classes or how...

arxiv.org
5
10
66
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
5 months
Thanks everyone for stopping by! See you next year in Miami #EMNLP2024 ? 🌴
Tweet media one
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
Drug monitoring (PharmacoVigilance) is incredibly important for public safety. We set out to improve it using NLP! Introducing BioDEX, a dataset for Biomedical adverse Drug Event Extraction, containing 19k papers and 256k expert-created drug reports. 📄
Tweet media one
3
18
60
1
10
64
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
Drug monitoring (PharmacoVigilance) is incredibly important for public safety. We set out to improve it using NLP! Introducing BioDEX, a dataset for Biomedical adverse Drug Event Extraction, containing 19k papers and 256k expert-created drug reports. 📄
Tweet media one
3
18
60
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
With ≅50 labeled inputs and a minimal prompt, we bootstrap prompts left-to-right, using different LMs for each module–this is crucial to achieve the cheapest program with best performance. This optimization is instantiated from the logic directly, which is *tiny* (thanks DSPy!)
Tweet media one
1
4
44
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
9 months
I’m back at Stanford for 6 months, working with @stanfordnlp and specifically professor @ChrisGPotts ! Let’s meet up to talk about explainable AI, biomedical NLP or interpretable model editing (my new project 👀).
Tweet media one
3
1
38
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
#AI gebruiken om op basis van bestaande @ugent vakken nieuwe vakken te verzinnen? Dat kan uiteraard: dit is hoe "Advanced Procrastination" of "Introduction to Milking Cows" er zou uit zien aan de UGent.
Tweet media one
Tweet media two
2
8
38
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
@lateinteraction This could not be possible without the rapid development-cycle DSPy permits! Can't wait to explore further, so much low-hanging fruit: chunking, ensembling, hierarchical ontologies, etc. -- all backed by bootstrapping and data-driven optimization.
2
3
35
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
People who fly from #NeurIPS straight to #EMNLP , I commend you for your insane stamina
1
1
35
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 months
I had a TON of fun discussing Infer-Retrieve-Rank (IReRa) with @CShorten30 on the @weaviate_io podcast! We talked about extreme classification, DSPy, biomedical NLP, advanced program optimization strategies, and much more!
@CShorten30
Connor Shorten
@CShorten30
3 months
Hey everyone! I am BEYOND EXCITED to publish an interview with Karel D'Oosterlinck ( @KarelDoostrlnck ) from @ugent & @stanfordnlp ! 🔥 Karel's Infer-Retrieve-Rank is an amazing use of DSPy for Extreme Classification! Learned a ton from this conversation!🤯
Tweet media one
6
28
101
1
9
33
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
@tomgoldsteincs I've been playing with something similar! Complex character manipulations are still quite hard, here are some negative results.
Tweet media one
Tweet media two
4
1
32
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
We optimize the program on 4 datasets. We set SotA on 3 HR benchmarks (labeling job vacancies) and get meaningful traction on BioDEX (extracting medical reactions from full biomedical papers), despite these being very different in shape. cost to optimize <<< cost to train.
Tweet media one
1
6
31
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
This person tried really hard to find our work and got rate limited from executing too many queries 🥲 Glad you ended up finding it !
@paul_cal
Paul Calcraft
@paul_cal
4 months
I just got rate limited trying to search for a tweet. Does anyone remember the one flying around about good multi-label classification performance with LLMs? I feel like it said 10,000 labels or something surprising (And can twitter *please* add a bookmark search option)
Tweet media one
1
0
9
4
1
28
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
(1/7) Als laatstejaarsstudent burg. ir. computerwetenschappen aan de @ugent @ugent_fea heb ik afgelopen semester bijgehouden hoe veel uur ik geïnvesteerd heb in mijn studies, thesis, 'passion projects' en vakantiejobs. Het resultaat is deze grafiek 👇.
Tweet media one
3
0
27
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
If I read a paper that is a few months old and the Twitter thread has already died out, is it weird to "revive" it if I have some questions / want to engage in a discussion? @PhDVoice #AcademicChatter #AcademicTwitter
8
2
25
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
👀Sneak peek of the the new study aid application we are developing together with @GentseStud for all @ugent students. You will be able to easily share notes, documents and tips&tricks within the confines of the new UGent copyright rules. 💪Beta-release end of September
1
1
24
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
Awesome little win: IReRa made it to the PapersWithCode Trending Research page🎉!
Tweet media one
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
📢Tasks with > 10k classes (e.g. information extraction) are hard for in-context learning: typically a tuned retriever or many in-context calls per input are used ($$$) Infer-Retrieve-Rank (IReRa) is a SotA program using 1 frozen retriever with a query predictor and reranker.
Tweet media one
8
81
520
2
4
23
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 months
Looks like I picked the correct place to do an internship 😅. Great results by @winniethexu
@winniethexu
Winnie Xu
@winniethexu
3 months
Excited to share a new model with @ContextualAI that tops the AlpacaEval 2.0 leaderboard! How did we manage to rank higher than models like GPT4, Claude 3 and Mistral Medium? Enter iterative alignment… 🧵
Tweet media one
11
35
203
1
3
24
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
People have been asking where I’ll share the code. I’ll put it here:
Tweet card media
GitHub - KarelDO/xmc.dspy: In-Context Learning for eXtreme Multi-Label Classification (XMC) using...

In-Context Learning for eXtreme Multi-Label Classification (XMC) using only a handful of examples. - KarelDO/xmc.dspy

github.com
@KarelDoostrlnck
Karel D��Oosterlinck
@KarelDoostrlnck
4 months
📢Tasks with > 10k classes (e.g. information extraction) are hard for in-context learning: typically a tuned retriever or many in-context calls per input are used ($$$) Infer-Retrieve-Rank (IReRa) is a SotA program using 1 frozen retriever with a query predictor and reranker.
Tweet media one
8
81
520
0
3
22
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
Just had my first #opensource contribution merged on @github (to ) 🥰🤩. My contribution wasn't much, but I am amazed how easy the entire process was. I think I might do this more often 😉.
2
1
23
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
Happy to announce our paper 'Causal Proxy Models for Concept-based Model Explanations' was accepted to #ICML2023 !
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
🚨Preprint🚨 Interpretable explanations of NLP models are a prerequisite for numerous goals (e.g. safety, trust). We introduce Causal Proxy Models, which provide rich concept-level explanations and can even entirely replace the models they explain. 1/7
2
21
132
2
0
23
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
Slightly late announcement, but I'm thrilled that the first paper of my PhD has been accepted to #NeurIPS2022 ! Many thanks to everyone who made this possible!
1
0
22
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
👀Sneak peek of a weekend project I'm working on: intuitive video editing via language. I hope to add many NLP-based features soon that will allow for insanely quick text-based video editing, without ever opening actual editing software.
3
5
22
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 months
How can you optimize RAG-like systems? One way is to have a teacher model generate demonstrations for a student model. However, you can go much further in this teacher - student interaction. We discussed some advanced optimization ideas in this clip:
2
1
22
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
The people in my building are organizing a French (!) chocolate tasting event. I am deeply offended and have no choice but to retaliate by organizing a Belgian wine & cheese evening.
0
0
21
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
@YiMaTweets As a first year PhD student, I feel it’s challenging to get a good grip on the history of the field. How would you advise a PhD student in ML to balance “catching up” with the field on one hand, and having productive output on the other hand?
2
0
21
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 years
Professoren die hun les via livestream geven maar toch weigeren die dan op te nemen en achteraf online te plaatsen, waarom??? Ik probeer het oprecht te begrijpen. Lesopnames zijn goud waard.
0
3
21
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
6 months
🥳🍾 Congratulations to my awesome collaborators Jing Jiang Huang, Atticus Geiger, @ZhengxuanZenWu and @ChrisGPotts (and me) for winning a @BlackboxNLP best paper award!!
@BlackboxNLP
BlackboxNLP
@BlackboxNLP
6 months
🏆 Rigorously Assessing Natural Language Explanations of Neurons by Jing Huang, Atticus Geiger, Karel D'Oosterlinck, Zhengxuan Wu and Christopher Potts
1
1
9
1
2
19
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
7 months
Neural coref is an important step in many NLP pipelines. SOTA coref methods are inefficient, using >= O(n) passes of an LM per document. We revisit ~efficient~ coref, identify and fix 2 of its routine failure cases and close the gap with SOTA by 34%!
Tweet card media
CAW-coref: Conjunction-Aware Word-level Coreference Resolution

State-of-the-art coreference resolutions systems depend on multiple LLM calls per document and are thus prohibitively expensive for many use cases (e.g., information extraction with large...

arxiv.org
1
2
17
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
Awesome list of DSPy resources by @KamaraiCode !
@KamaraiCode
Jason
@KamaraiCode
4 months
DSPy Github Repo - Intro Notebook - DSPy - Assertions - Computational Constraints for Self-Refining Language Model Pipelines (Dec 2023) - In-Context Learning for Extreme Multi-Label Classification (Jan
3
14
61
1
6
18
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 years
@rvdwalle @HLN_BE De studenten die blij zwaaien naar de camera alsof er helemaal niets aan de hand is maken het echt helemaal af...
0
1
18
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
7 months
Interpretability claims should be rigorously measured, else we might run the risk of deceiving ourselves. In this work led by Jing Huang, we assessed a recent OpenAI interpretability proposal and found explanations of neurons to generally not align with actual behavior.
@ChrisGPotts
Christopher Potts
@ChrisGPotts
7 months
Jing Huang, Atticus Geiger, @KarelDoostrlnck @ZhengxuanZenWu & I found this OpenAI proposal inspiring and decided to assess it. We find that the method has low precision and recall, and we find no evidence for causal efficacy. To appear at BlackboxNLP:
5
29
239
1
3
17
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
Ik houd van vakken die samen met de theorie ook een portie geschiedenis voorzien: "Door wie is iets uitgevonden? Wanneer? Wat was de tijdsgeest toen, welk probleem wouden ze oplossen? Welke impact had de uitvinding?". Helpt alles in een context te plaatsen en te onthouden.
0
0
17
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
Awesome and detailed explanation of our Infer-Retrieve-Rank (IReRa) work by @BotDeepLearning ! I appreciate how they walk through the actual code and show actual examples of bootstrapped demonstrations. Subscribed!
@BotDeepLearning
Awesome Papers Review
@BotDeepLearning
4 months
DSPy on ICL RAG Classification: Code explained 🎥 Watch here:
0
8
27
5
4
16
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
This Wednesday at 8pm, we are hosting a ‘ #datascience bar’ on Clubhouse about recent trends and advancements in #NLP . Come join us for a fun chat! Some people joining: @yvespeirsman ( @nlptown ) @JVHautte (Techwolf) @FsMatt ( @ml6team ) @OxyKodit ( @explosion_ai )
2
5
16
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
Trots om één van de genomineerden voor Student Van Het Jaar @Stadgent te mogen zijn! Bedankt aan de vriendjes van @VTKGent en @GentseStud om niet vies te zijn van een ambitieus projectje of twee😉. Lees de interviews met mij en de andere kandidaten hier:
0
1
16
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
Come to Ghent Belgium, I’ll show you around, you won’t regret it 🍫🍺🇧🇪
Tweet media one
Tweet media two
Tweet media three
Tweet media four
2
0
16
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
California is pretty sick
Tweet media one
Tweet media two
Tweet media three
Tweet media four
0
0
15
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 months
Check out all these people working on awesome projects in the RAG / DSPy / Prompt optimization space🤯
@lateinteraction
Omar Khattab
@lateinteraction
3 months
Incomplete list of DSPy internals experts. Follow them all! Optimization: @kristahopsalong @michaelryan207 @KarelDoostrlnck @JoshPurtell Modules & Assertions: @arnav_thebigman @ShangyinT @slimshetty_ Abstractions: @Lambda_freak @HaozeHe_Hector @krypticmouse @cyrusofeden
6
17
129
0
2
15
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
Why and how to build a constrained🔒language model, using @huggingface (notebook with code in the @TDataScience article). 🧵👇
Tweet card media
The power of constrained language models.

Why and how to build constrained language models, using a custom beam search algorithm. A guide with Hugging Face code.

towardsdatascience.com
1
5
15
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
Loved my time at #CLeaR2023 . People have been posting many pictures of the conference, so enjoy some shots of Tübingen instead! See you all next year at @UCLA ? @CLeaR_2022
Tweet media one
Tweet media two
Tweet media three
0
1
15
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 years
Disclaimer: I was nowhere near that street
1
0
14
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
@wduyck en lesopnames voor elk vak die de studenten op eigen tempo kunnen verwerken
0
0
15
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
I really enjoyed discussing some recent advances in AI / NLP at #NordicsDataConference2023 in Oslo! Thanks @strandedinoslo for the wonderful event!
@strandedinoslo
Patricio Lobos
@strandedinoslo
1 year
@ChrisGPotts @KarelDoostrlnck @stanfordnlp Presenting at the #NordicsDataConference2023 Thank you so much for presenting at our conference!, it is a #KodakMoment for the banking industry after #BloombergGPT and @openai work with @MorganStanley I guess it is our turn😊😎😎. Our
0
1
16
0
0
14
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
Not one but two Infer-Retrieve-Rank releases today??? What a time to be alive. Great job by the @llama_index team 🚀
@llama_index
LlamaIndex 🦙
@llama_index
4 months
Infer-Retrieve-Rerank ( @KarelDoostrlnck et al.) is a simple but powerful paradigm to use LLMs for complex classification problems with thousands of classes. Examples include medical reactions 🥼 and job skills/qualifications 👷. Use an LLM to infer a set of predictions, use
Tweet media one
4
70
353
0
1
14
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
Cloud computing ☁️
Tweet media one
1
1
14
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
@trappology @lateinteraction 1. Follow @lateinteraction and read some of his threads, they are great starting points. 2. Check out some of the tutorials and notebooks over at 3. Keep your eyes peeled for a new paper we’re releasing very soon 🤫
Tweet card media
GitHub - stanfordnlp/dspy: DSPy: The framework for programming—not prompting—foundation models

DSPy: The framework for programming—not prompting—foundation models - stanfordnlp/dspy

github.com
0
0
14
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
10 months
I’m presenting this work at ICML in Hawaï next week! Feel free to reach about it you want to talk explainable, causal AI!
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
🚨Preprint🚨 Interpretable explanations of NLP models are a prerequisite for numerous goals (e.g. safety, trust). We introduce Causal Proxy Models, which provide rich concept-level explanations and can even entirely replace the models they explain. 1/7
2
21
132
1
1
14
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 months
I like this paper
@ericzelikman
Eric Zelikman
@ericzelikman
2 months
Language models today are trained to reason either 1) generally, imitating online reasoning data or 2) narrowly, self-teaching on their own solutions to specific tasks Can LMs teach themselves to reason generally?🌟Introducing Quiet-STaR, self-teaching via internal monologue!🧵
35
190
973
0
0
13
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
IReRa is a general RAG-type system which can be efficiently optimized towards a range of Information Extraction datasets. Prompt and pipeline engineering should be scalable. This code provides the IReRa building blocks; use it to build (&optimize) all kinds of RAG-type systems.
Tweet media one
1
4
12
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
Gaan de zaken wat te goed in je leven? Introducing geautomatiseerde demotivatie™! Net enkele unmotivational quotes gegenereerd met #AI , dit zijn mijn favoriete:
Tweet media one
Tweet media two
Tweet media three
1
0
13
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
Exploring California🐻☀️
Tweet media one
Tweet media two
Tweet media three
Tweet media four
0
0
12
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
Today, I'm officially a *second-year* PhD student. The first year was even more fun than I ever could have imagined. Many thanks to all friends, colleagues and mentors along the way!
0
0
12
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
The D in DSPy stands for "Deal with it" when DSPy snipes the top of the leaderboard.
@lateinteraction
Omar Khattab
@lateinteraction
4 months
You should read our new guide to LLM abstractions, a stack with 5 layers! To randomly help this tweet reach ppl, see how happy DSPy power users feel when they get state-of-the-art scores using DSPy optimizers—so happy, in fact, they make fancy slack emojis. cc: @KarelDoostrlnck
Tweet media one
0
4
38
0
1
11
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
10 months
Want to learn more about Causal Proxy Models and concept based model explanations? Swing by poster #706 , session 2 (2pm HST) in exhibit hall 1! #ICML2023
Tweet media one
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
🚨Preprint🚨 Interpretable explanations of NLP models are a prerequisite for numerous goals (e.g. safety, trust). We introduce Causal Proxy Models, which provide rich concept-level explanations and can even entirely replace the models they explain. 1/7
2
21
132
1
0
11
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
Many thanks to my wonderful collaborators @lateinteraction , @FremyCompany , @thomeestr , Chris Develder, and @ChrisGPotts .
1
1
11
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
Week 1-10: vliegt voorbij in 2 weken Week 11-13: duurt 10 mentale weken Blok: Wat is tijd? Wat is een week? Welke dag is het vandaag?
0
0
11
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
6 months
Crucially, we apply discrete prompt optimization to the in-context module, with the grounding in-the-loop. No human-expert prompting work required! Easy, intuitive, reproducible, and modular; this is how in-context learning should be done.
Tweet media one
2
1
12
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
@TristanThrush Exciting work! Also, this is based @aryaman2020 .
Tweet media one
0
0
11
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
6 months
We’re working on a more rigorous study; in the meantime enjoy our notebook and feel free to build your own state-of-the-art reaction-extraction systems for biomedical literature! Get some more information on the BioDEX dataset:
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
Drug monitoring (PharmacoVigilance) is incredibly important for public safety. We set out to improve it using NLP! Introducing BioDEX, a dataset for Biomedical adverse Drug Event Extraction, containing 19k papers and 256k expert-created drug reports. 📄
Tweet media one
3
18
60
1
2
12
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
@raphaelmilliere Wait this is actually so good and funny
0
0
11
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
6 months
I'll be talking about NLP for pharmacovigilance, and our BioDEX dataset, January 10th. See you there?
Tweet card media
Next in AI: AI For Pharma Pioneer Spotlight | LinkedIn

Topic: How AI disrupts Pharma: Natural Language Processing for Drug Event Extraction The timely and accurate extraction of Adverse Drug Events (ADE) is not just a technical challenge—it’s a vital...

www.linkedin.com
0
2
10
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
10 months
Pre-conf was a big success #ICML2023
Tweet media one
Tweet media two
Tweet media three
0
0
11
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 months
This Monday, I'm giving a talk on how NLP can help parse biomedical documents and how to efficiently bootstrap RAG-like systems for such tasks! If you're in Ghent, come join! It's at the @ml6team office.
Login to Meetup | Meetup

Not a Meetup member yet? Log in and find groups that host online or in person events and meet people in your local community who share your interests.

www.meetup.com
1
2
11
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
Eat Sleep Thesis Repeat
0
0
10
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
Who wants to give me some constructive feedback on my personal website? All kinds of ideas welcome! I feel it might not be 'academic' enough? @PhDVoice @OpenAcademics @AcademicChatter @PhD_Genie #AcademicChatter
11
1
10
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
Treat yourself to this video, you’ve earned it
Tweet card media
Microservices

it's because of the way our backend works // unlock exclusive deleted scenes: https://www.patreon.com/KRAZAM // merch: https://merch.krazam.tv // https://www...

www.youtube.com
2
0
9
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
Our non-profit platform got featured in the local news😄! @GentseStud
@DeGentenaar
De Gentenaar
@DeGentenaar
3 years
Eén website om notities en samenvattingen te delen: studenten willen eigen deelplatform lanceren
Tweet media one
0
0
2
0
0
10
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
Now that I have my engineering degree, I can finally say "I'm a professional" every time I fix my mom's computer. Worth it.
1
0
10
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
@clin_dev_1 Well, not solved yet! I think we still need some work to get stronger and cheaper few-shot performance on a range of biomedical tasks. If cheap enough, I'd love to run IReRa over every public access biomedical paper for a bunch of ontology-tasks and opensource the results.
1
0
9
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
Mijn favoriete nieuwe vakken zijn "Computational Procrastination", "Advanced Juggling" en "Introduction to Twitter".
Tweet media one
Tweet media two
Tweet media three
1
0
9
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
The only thing worse than not knowing what caused a bug is not knowing what fixed it.
0
0
9
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
CC-ing some people who may be interested: @bclavie , @matei_zaharia , @srush_nlp , @jeremyphoward
1
0
9
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
@llama_index This is amazing, thanks for helping make Infer-Retrieve-Rank accessible. We've also just released our official repository. Exciting times for RAG🤯
Tweet card media
GitHub - KarelDO/xmc.dspy: In-Context Learning for eXtreme Multi-Label Classification (XMC) using...

In-Context Learning for eXtreme Multi-Label Classification (XMC) using only a handful of examples. - KarelDO/xmc.dspy

github.com
1
1
9
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
@tomgoldsteincs However, I can't say I'm not impressed with some of the results on simpler problems.
Tweet media one
Tweet media two
2
0
9
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
We're on the cusp of general RAG-type programs, but a ton of exciting research and engineering still needs to be done. The design spaces for the program logic and the optimization flow are vastly underexplored. I've listed some ideas below. Happy to collaborate, DMs are open!
Tweet media one
1
1
8
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
We believe that both training on counterfactual data and localizing hidden representations through intervention training could be a valuable avenue towards the development of more robust, explainable, and malleable neural networks. 📄 Read the paper here:
Tweet card media
Causal Proxy Models for Concept-Based Model Explanations

Explainability methods for NLP systems encounter a version of the fundamental problem of causal inference: for a given ground-truth input text, we never truly observe the counterfactual texts...

arxiv.org
1
0
8
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
6 months
I've been selected as a highlighted reviewer for the @XAI_in_Action workshop at @NeurIPSConf ! I'm glad we are putting some spotlight on reviewing, since this is such an important but under appreciated part of our field. Thanks!
0
0
7
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
Discrete optimization is back again???
@lateinteraction
Omar Khattab
@lateinteraction
4 months
Prompts are the new parameters. LM programs are the new DNNs. Compiling is the new training.
4
5
74
0
0
7
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
Goedemorgen aan iedereen behalve mensen die 3 GPU's claimen voor een interactieve job en die job dan 10 dagen laten runnen.
0
0
8
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
Ervaring met web-development? Nog op zoek naar een betaalde vakantiejob? Met @GentseStud en @VTKGent zijn we op zoek naar een vakantiejobber om deze zomer mee te werken aan een ambitieus studentenproject met enorme impact! Zie foto's of DM voor meer info.
Tweet media one
Tweet media two
Tweet media three
Tweet media four
0
0
8
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
Joint work with @ZhengxuanZenWu , @atticus_geiger , Amir Zur, and @ChrisGPotts at @stanfordnlp . 📄 Read the paper here: 👇 Read the summary in this thread. 2/7
Tweet media one
1
0
7
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
6 months
Previous fine-tuned and in-context attempts struggled on BioDEX because of long contexts, the biomedical domain, and extreme classification (~26k classes). Combining grounding and in-context learning is a great first step, but many pipelines are possible!
1
0
8
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
Preach! 🙏
Tweet media one
@ZhengxuanZenWu
Zhengxuan Wu
@ZhengxuanZenWu
4 months
🧐In our new commentary: we argue the notion of "illusion" in this paper labels correct explanation as illusory, & that avoiding "illusion" would require unwarranted constraints on NNs. The "illusions" are, though, instructive about how models work. 1/
2
12
54
0
1
7
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
@hvbris_ Should you use DSPy to scaffold the interaction between LMs and retrievers?id say so! If you have a metric you want to optimize and some data, you can get a lot of value out of using DSPy’s optimization. If you know gpt4 can do the task zero-shot, bootstrapping can be strong.
0
1
7
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
Here is the code:
Tweet card media
GitHub - KarelDO/xmc.dspy: In-Context Learning for eXtreme Multi-Label Classification (XMC) using...

In-Context Learning for eXtreme Multi-Label Classification (XMC) using only a handful of examples. - KarelDO/xmc.dspy

github.com
0
1
7
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
4 months
This talk about our "NLP for Pharmacovigilance" work is tomorrow, 9am PST!
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
5 months
Can AI help drug safety research? Find out January 10, 9am PST (6pm CET). Come listen to my talk, no expert AI-knowledge required to attend! A healthy mix of Pharma, AI, and Venture Capital people have already signed up.
1
0
2
0
1
4
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
6 months
Sometimes you’ve got to go and capture yourself a fresh set of zoom backgrounds.
Tweet media one
Tweet media two
Tweet media three
Tweet media four
0
0
7
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
@courteauxm Thanks, looking forward to learn a lot more about NLP and research in general🤓!
0
0
7
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
1 year
All our code and resources are open-source, check them out!
Tweet card media
GitHub - KarelDO/BioDEX: BioDEX: Large-Scale Biomedical Adverse Drug Event Extraction for Real-Wo...

BioDEX: Large-Scale Biomedical Adverse Drug Event Extraction for Real-World Pharmacovigilance. - KarelDO/BioDEX

github.com
1
0
7
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
3 years
13.5 uur aan lesopnames gekeken op 9 uur vandaag => lesopnames zijn fantastisch
0
0
6
@KarelDoostrlnck
Karel D’Oosterlinck
@KarelDoostrlnck
2 years
@tomgoldsteincs We've just released a benchmark for explanation methods in NLP! While many different kinds of methods should still be added, we hope that our benchmark will lead to a more principled evaluation of explanation methods in AI.
0
0
6