Giang Nguyen @giangnguyen2412
PhD Fellow @AuburnEngineers, Prev. @kaistcsdept Making AIs understandable & friendly to humans via XAI 🤖🤝👨💻 giangnguyen2412.github.io Joined May 2019-
Tweets408
-
Followers189
-
Following362
-
Likes2K
Do models need to reason in words to benefit from chain-of-thought tokens? In our experiments, the answer is no! Models can perform on par with CoT using repeated '...' filler tokens. This raises alignment concerns: Using filler, LMs can do hidden reasoning not visible in CoT🧵
In AI research there is tremendous value in intuitions on what makes things work. In fact, this skill is what makes “yolo runs” successful, and can accelerate your team tremendously. However, there’s no track record on how good someone’s intuition is. A fun way to do this is…
One thing that I started doing at OpenAI is that I created a policy for myself to be *100% transparent* with my manager about everything. It seems obvious and weird to say aloud, but I bet most people don’t actually do this. But once I started doing it, I realized there are a lot…
New Anthropic research: we find that probing, a simple interpretability technique, can detect when backdoored "sleeper agent" models are about to behave dangerously, after they pretend to be safe in training. Check out our first alignment blog post here: anthropic.com/research/probe…
BLINK Multimodal Large Language Models Can See but Not Perceive We introduce Blink, a new benchmark for multimodal language models (LLMs) that focuses on core visual perception abilities not found in other evaluations. Most of the Blink tasks can be solved by humans
The age at which scientists or inventors achieve their moment of genius increasing: Half of all pioneering contributions in science now happen after age 40, it used to be younger. Why? There is much more to master before making a contribution to a field. nber.org/papers/w19866
Our new GPT-4 Turbo is now available to paid ChatGPT users. We’ve improved capabilities in writing, math, logical reasoning, and coding. Source: github.com/openai/simple-…
🔍How can we design neural networks that take neural network parameters as input? 🧪Our #ICLR2024 oral on "Graph Neural Networks for Learning Equivariant Representations of Neural Networks" answers this question! 📜: arxiv.org/abs/2403.12143 💻: github.com/mkofinas/neura… 🧵 [1/9]
Our experiment found that larger, newer AI models tended to be more persuasive - a finding with important implications as LMs continue to scale. Read more about our research here: anthropic.com/news/measuring…, and access the data from our experiment here: huggingface.co/datasets/Anthr…
To assess persuasiveness, we measure the shift in people’s support between their initial view on a claim and their view after reading arguments written by either a human or an LM. We define the persuasiveness metric as the difference between the support scores.
github.com/karpathy/llm.c @karpathy has now implemented training of GPT-2 (CPU, fp32) in C, ~1,000 lines of clean code in a single file ‘What I cannot create, I do not understand’
Btw writing the llm.c training code would imo be a very interesting, impressive, self-contained and very meta challenge for LLM agents. The prompt is: Take the PyTorch code train_gpt2.py And write, compile and unit test a single .c file that reproduces the training: train_gpt2.c…
One of my favorite YouTubers, @3blue1brown, has put out an incredible video explainer about the attention mechanism! I highly recommend checking it out!
[1/5] Introducing VisDiff - an #AI tool that describes differences in image sets with natural language. VisDiff can summarize model failures, compare models, find nuanced dataset differences, discover what makes an image memorable, and so much more! …derstanding-visual-datasets.github.io/VisDiff-websit…
🎉April Fool is the birthday of Unsolvable Problem Detection! UPD examines the VLM’s ability to withhold answers when faced with unsolvable problems. Please enjoy VLMs with unsolvable problems today! paper page: arxiv.org/abs/2403.20331 code: github.com/AtsuMiyai/UPD
🎉April Fool is the birthday of Unsolvable Problem Detection! UPD examines the VLM’s ability to withhold answers when faced with unsolvable problems. Please enjoy VLMs with unsolvable problems today! paper page: arxiv.org/abs/2403.20331 code: github.com/AtsuMiyai/UPD
Smeslesho @smeslesho34918
0 Followers 123 Following Life itself is a journey, we are all worthy and should strive to travel to different lives.Antonin Poché @Antonin_Poche
21 Followers 50 Following Research engineer in XAI at IRT Saint Exupéry in https://t.co/RhQcHCalSF and https://t.co/pBG7yb0Hnw projects. In the developing team of Xplique.Tanmoy Mukherjee @langer_han
644 Followers 2K Following Senior researcher imec.Avid reader. Interested in ML/Explainability/Interpretability/CVShreyas Vaidya @shreyasvaidya23
161 Followers 1K Following Nothing beats the joy of solving interesting problems Third year UG majoring in CS @iitjodhpurAdg_key123 @adg_key123
26 Followers 205 Following PhD Student, Researcher (Explainability in Networks)Weina Jin, MD @weina_jin
149 Followers 115 Following PhD student at Simon Fraser University. A MD and machine learner. #deeplearning #computervision #healthcareOliver Eberle @EberleOliver
195 Followers 615 Following Postdoctoral Researcher @ Machine Learning Group, @TUBerlin 🇩🇪 | 🔮 Explainable AI | 📚 NLP & Humanities | 🧠 Alumni @bccn_berlinEoin Delaney @EoinDelaney_
151 Followers 206 Following Postdoc @UniofOxford @oiioxford 🇮🇪 | Interested in XAI, Interpretability, Evaluation, and Trustworthiness Auditing in Artificial IntelligenceMajeed Kazemi @MajeedKazemi
1K Followers 2K Following PhD student in CS @UofT with @ToviGrossman HCI + Computing Education + Coding / Creativity Support Tools Prev: @MSFTResearch + MSc @HCIL_UMD with @JonFroehlichThao Le @thaole252
63 Followers 206 Following PhD Candidate @Unimelb @cis_unimelb Explainable AI #XAI #ExplainableAILê Bình @jesuislebeauu
27 Followers 33 FollowingShuai Ma | 马帅 @shuaima_hci
181 Followers 410 Following PhD student at @HKUST. #HCI Interests: human-centered system, user modeling/understanding, human-AI collaboration, AI-assisted decision-makingThao Nguyen (Shibe) @thaoshibe
434 Followers 300 Following Hi, I'm a graduate student at @WisconsinCS 🥑Shivam Rai @imsr282
332 Followers 5K Following Tech enthusiast 🚀 | Embarking on a journey through Machine Learning & Data Science 🤖📊 | Curious mind, coding heart ❤️ | Exploring the data-driven frontier 🌐Olioli @Oliolilyx
122 Followers 2K FollowingSloughez @sloughez86994
10 Followers 443 Following Leva apenas 10 minutos para atingir facilmente seu objetivo de ganhar dinheiro usando seu celular.Jiarui Zhang (Jerry) @JiaruiZ58876329
218 Followers 572 Following 张家瑞| @USC CS Ph.D. student @CSatUSC | ex-intern @amazon | B.Eng. @Tsinghua_Uni | MLLM | ReasoningAlbert @ZihengChen1993
49 Followers 565 Following Researcher in XML,Reinforcement learning and recommender system.Yue Dong @ NeurIPS 20.. @YueDongCS
3K Followers 797 Following Assistant Prof @UCRiverside. PhD from @Mila_Quebec @McGillU. Trustworthy NLP+AI safety & Summarization! Former intern @GoogleAI @MSFTResearch @allen_aiJihed Ncib @JihedNcib
2K Followers 3K Following Political Data Scientist @ucddublin | Machine learning | NLP | Manager @Connected_Pol | Member of the https://t.co/rZKw2KrEdS research groupDr. Ulrike Kuhl @DrUlrikeKuhl
507 Followers 723 Following 🥷 Scientific coordinator of the Data-NInJA research training group 🥷 PostDoc @HammerLabML, pondering cognition, explainability, and machine learning. she/herTanya Chowdhury @ta_knee_aa
361 Followers 958 Following Ph.D. Candidate at @umasscs. Prev @genentech @Google @IIITDelhi. Dabbling with Interpretability, Retrieval and some Bioinformatics.Divyansh Agarwal @divyanshaga
353 Followers 2K Following CS grad student @ucla. Prev @glean @uber (Applied Science) @ucberkeley (CS + Statistics). I also do improv theatre/comedyTobias Leemann @t_leemann
77 Followers 100 Following PhD Student in Explainable, Private and Reliable ML | @uni_tue @TU_MuenchenUsha Rengaraju @URengaraju
2K Followers 1K Following Ranked as Top Ten Data Scientists in India 2020 | Ranked as Top ten women data scientist in 2021 (India)| Ranked Top AI leader in 2021|Keynote SpeakerChris Yao Du @yao53513502
446 Followers 3K Following PhD@HKUST Computer Vision, Medical Image AnalysisWENHAN YANG @WenhanYang0315
228 Followers 435 Following P.hD. in CS, UCLA. Interest in self-supervised learning, including exploring Graph CL, CL robustness and multimodal CL robustness.Gabriel Kasmi @gabriel_kasmi
28 Followers 164 Following PhD | XAI for power systems 🎓 @ENSAEparis, @ENS_ParisSaclay & @Mines_Paris 💼@rte_france #deeplearning #energytransitionIvaxi Sheth @ivakshi_s
312 Followers 817 Following PhD student @ CISPA | Prev @Mila_Quebec @imperialcollege ‘20 Organizer @WiCVworkshop @CVPR 2022 and 2023David Carlyn @Carlyn2015
333 Followers 538 Following Computer Science PHD Student @ The Ohio State UniversityTill Beemelmanns @T_Beemelmanns
2 Followers 14 Following PhD Student @RWTH | Computer Vision for Automated Driving | Prev. @MercedesBenz_DEBiagio La Rosa @larosabiagio
91 Followers 230 Following PhD student on Explainable Deep Learning at Cognitive Cooperating Robots Lab (RoCoCo) @SapienzaRoma | Visitor Scholar @ucscVinitra Swamy @vinitra_s
367 Followers 527 Following machine learning PhD @EPFL (explainability, edtech, generalized learning) | formerly at @UCBerkeley @Microsoft @onnxai | ✨👩🏽💻📚🇺🇸✨Michael Hanna @michaelwhanna
263 Followers 309 Following PhD student at the University of Amsterdam / ILLC, interested in computational linguistics and (mechanistic) interpretabilityWeiqiu You @youweiqiu
173 Followers 128 Following PhD @CisPenn in #ML. Former MS @UMassCS, Intern @USC_ISI @IBM . Interested in ML and explainable AI in general. She/herSaundra @murguiasaundra2
130 Followers 3K FollowingMelinda @neal41melinda
194 Followers 3K FollowingZixiang Chen @_zxchen_
986 Followers 2K Following Ph.D. student in CS @UCLA. 📚 B.S. from Tsinghua Univ. 🔍 Interested in Representation Learning, Generative Model & Reinforcement Learning.Duy H. M. Nguyen @DuyHMNguyen1
187 Followers 601 Following Ph.D. Student at Max Planck Research School for Intelligent Systems & University of Stuttgart (@MPI_IS). Working on ML for simulation science.Minseon Kim @kim__minseon
297 Followers 356 Following Ph.D student, Graduate school of AI @KAIST | Adversarial robustness, Self supervised learning, Robustness in Diffusion model |Eoin Kenny @EoinKNNy
133 Followers 79 Following Postdoctoral Associate @MIT 🇮🇪 I am interested in how to deploy understandable and useful XAI.Ali Behrouz @behrouz_ali
912 Followers 848 Following Ph.D. Student @cornell, interested in machine learning.Kirill Bykov @kirill_bykov
349 Followers 1K Following Explainable AI Machine Learning PhD student @UMI_Lab_AI, @bifoldberlin, @TUBerlin; Wir müssen wissen, Wir werden wissenXingyu Fu @XingyuFu2
338 Followers 244 Following PhD student at Upenn @cogcomp. | Focused on Vision+Language Multimodal learning | Previous: B.S. @UIUCAntonin Poché @Antonin_Poche
21 Followers 50 Following Research engineer in XAI at IRT Saint Exupéry in https://t.co/RhQcHCalSF and https://t.co/pBG7yb0Hnw projects. In the developing team of Xplique.Tanmoy Mukherjee @langer_han
644 Followers 2K Following Senior researcher imec.Avid reader. Interested in ML/Explainability/Interpretability/CVMaria De-Arteaga @mariadearteaga
5K Followers 540 Following Asst Professor, IROM Dept @UTAustin | PhD, Machine Learning & Public Policy @CarnegieMellon | Algorithmic fairness, human-AI collab | 🇨🇴 💚 she/her/ella.lmsys.org @lmsysorg
37K Followers 171 Following Large Model Systems Organization. We created Vicuna and Chatbot Arena! Compare 30+ LLMs (GPT-4/Claude/Llamas) side-by-side at https://t.co/IDFeIDIOtmJ.P. Morgan @jpmorgan
761K Followers 49 Following Official account for the latest company news and updates from Asset Management, Private Banking, Commercial Banking, and the Corporate and Investment Bank.Arthur Mensch @arthurmensch
40K Followers 872 Following Co-founder and CEO @MistralAI. Apply https://t.co/yHGRZAtjcxAdg_key123 @adg_key123
26 Followers 205 Following PhD Student, Researcher (Explainability in Networks)Martin Schuessler @martnsch
261 Followers 445 Following Techno±logie Enthusiast, XAI+HCI Researcher @TUBerlin and @JWI_Berlin. Currently evaluating the effectivenes of user-centered XAI technqiues.Negin Golrezaei @NeginGolrezaei
343 Followers 187 Following Associate Professor of Operations Management at the MIT Sloan School of ManagementAbby Smith @flabbysmith
484 Followers 481 Following (Over)enthusiastic, now @NORCnews | alum: @NorthwesternU stats PhD☕, @datascifellows 🌞, @CarnegieMellon 🐪 🏃♀️ | good with people, bad at everything elseLeon Sixt @LeonSixt
181 Followers 362 Following Machine Learning PhD Student | Interpretability | FU Berlin | him/heMartin Pawelczyk @MartinPawelczyk
265 Followers 399 Following Postdoc @Harvard. #reliableML & #recourse. PhD from Tübingen @uni_tue. MScs Stats & Econ @LSE @uni_edinburgh. Previously intern @JP_Morgan AI Research.Gaurav Verma @verma22gaurav
617 Followers 569 Following CS PhD student @GeorgiaTech | JPMorgan AI & Snap Research Fellow | Previously, @MSFTResearch @AdobeResearch; undergrad @IITKanpurOliver Eberle @EberleOliver
195 Followers 615 Following Postdoctoral Researcher @ Machine Learning Group, @TUBerlin 🇩🇪 | 🔮 Explainable AI | 📚 NLP & Humanities | 🧠 Alumni @bccn_berlinEoin Delaney @EoinDelaney_
151 Followers 206 Following Postdoc @UniofOxford @oiioxford 🇮🇪 | Interested in XAI, Interpretability, Evaluation, and Trustworthiness Auditing in Artificial IntelligenceMajeed Kazemi @MajeedKazemi
1K Followers 2K Following PhD student in CS @UofT with @ToviGrossman HCI + Computing Education + Coding / Creativity Support Tools Prev: @MSFTResearch + MSc @HCIL_UMD with @JonFroehlichRowan Cheung @rowancheung
497K Followers 373 Following Founder @therundownai. Sharing the latest developments in the world of artificial intelligence.Furong Huang @furongh
4K Followers 2K Following Assistant professor of @umdcs @umiacs @ml_umd at UMD. Researcher in #AI/#ML, #Trustworthy AI/ML, #EthicalAI, AI #Democratization, AI for ALL.Thao Le @thaole252
63 Followers 206 Following PhD Candidate @Unimelb @cis_unimelb Explainable AI #XAI #ExplainableAIICML Conference @icmlconf
70K Followers 17 Following Int'l Conf on ML • July 21-27, 2024 (Vienna, Austria) • #icml2024 • Contact: https://t.co/6saHKWV01y • https://t.co/sFwmcQNWkEShuai Ma | 马帅 @shuaima_hci
181 Followers 410 Following PhD student at @HKUST. #HCI Interests: human-centered system, user modeling/understanding, human-AI collaboration, AI-assisted decision-makingGuide Labs @guidelabsai
212 Followers 2 Following We are building interpretable foundation/frontier models that can reliably explain their reasoning, and are easy to {align/steer/debug}.Thao Nguyen (Shibe) @thaoshibe
434 Followers 300 Following Hi, I'm a graduate student at @WisconsinCS 🥑jessica dai @jessicadai_
2K Followers 675 Following phd student @berkeley_ai !? also editorial @reboot_hq @kernel_magazine (she/her)Alfredo Canziani @alfcnz
86K Followers 268 Following Musician, math lover, cook, dancer, 🏳️🌈, and an ass prof of Computer Science at New York UniversityAmir-Hossein Karimi @amirhkarimi_
2K Followers 2K Following 🇮🇷 🇨🇦 👨🏻🏫 Asst Prof of ML @UWaterloo & Faculty Affiliate @VectorInst 🔎 Explainable AI, Human-AI Teams 🧠🤖 ex-{@DeepMind, @GoogleAI, @Meta} CHARM Lab👇Jiarui Zhang (Jerry) @JiaruiZ58876329
218 Followers 572 Following 张家瑞| @USC CS Ph.D. student @CSatUSC | ex-intern @amazon | B.Eng. @Tsinghua_Uni | MLLM | ReasoningYue Dong @ NeurIPS 20.. @YueDongCS
3K Followers 797 Following Assistant Prof @UCRiverside. PhD from @Mila_Quebec @McGillU. Trustworthy NLP+AI safety & Summarization! Former intern @GoogleAI @MSFTResearch @allen_aiChhavi Yadav @chhaviyadav_
2K Followers 3K Following Machine Learning Researcher | PhD student @ucsd_cse | @trustworthy_mlYasha Ektefaie @YEktefaie
395 Followers 508 Following Bioinformatics PhD Student @HarvardDBMI | @UCBerkeley 2020 graduate in @BerkeleyBioE + @Berkeley_EECS | Fan of movies, music, and running!Aditya Bhattacharya @adib0073
149 Followers 130 Following Explainable AI Researcher | Ex-Microsoft | Author of Applied Machine Learning Explainability Techniques, Speaker, MentorDr. Ulrike Kuhl @DrUlrikeKuhl
507 Followers 723 Following 🥷 Scientific coordinator of the Data-NInJA research training group 🥷 PostDoc @HammerLabML, pondering cognition, explainability, and machine learning. she/herMax Bain @maxhbain
2K Followers 498 Following multimodal @RekaAILabs | prev: phd @Oxford_VGG hardwork-pilledWENHAN YANG @WenhanYang0315
228 Followers 435 Following P.hD. in CS, UCLA. Interest in self-supervised learning, including exploring Graph CL, CL robustness and multimodal CL robustness.Binxu Wang 🐱 @WangBinxu
826 Followers 819 Following @KempnerInst Fellow; Neuro PhD in Ponce Lab @Harvard; interested in Vision, generative model, optimization. Prev:WUSTL Neuro; PKU Physics, Yuanpei CollegeStefan Kolek @KolekDe
54 Followers 457 Following PhD student at Ludwig Maximilians University Munich - https://t.co/WugLb9O9I0Tolga Bolukbasi @tolgab0
277 Followers 213 Following AI/ML research @GoogleDeepmind, PhD, opinions my own.Gabriel Kasmi @gabriel_kasmi
28 Followers 164 Following PhD | XAI for power systems 🎓 @ENSAEparis, @ENS_ParisSaclay & @Mines_Paris 💼@rte_france #deeplearning #energytransitionIvaxi Sheth @ivakshi_s
312 Followers 817 Following PhD student @ CISPA | Prev @Mila_Quebec @imperialcollege ‘20 Organizer @WiCVworkshop @CVPR 2022 and 2023Till Beemelmanns @T_Beemelmanns
2 Followers 14 Following PhD Student @RWTH | Computer Vision for Automated Driving | Prev. @MercedesBenz_DEVinitra Swamy @vinitra_s
367 Followers 527 Following machine learning PhD @EPFL (explainability, edtech, generalized learning) | formerly at @UCBerkeley @Microsoft @onnxai | ✨👩🏽💻📚🇺🇸✨Biagio La Rosa @larosabiagio
91 Followers 230 Following PhD student on Explainable Deep Learning at Cognitive Cooperating Robots Lab (RoCoCo) @SapienzaRoma | Visitor Scholar @ucscTobias Leemann @t_leemann
77 Followers 100 Following PhD Student in Explainable, Private and Reliable ML | @uni_tue @TU_MuenchenWeiqiu You @youweiqiu
173 Followers 128 Following PhD @CisPenn in #ML. Former MS @UMassCS, Intern @USC_ISI @IBM . Interested in ML and explainable AI in general. She/herMichael Hanna @michaelwhanna
263 Followers 309 Following PhD student at the University of Amsterdam / ILLC, interested in computational linguistics and (mechanistic) interpretabilityDo models need to reason in words to benefit from chain-of-thought tokens? In our experiments, the answer is no! Models can perform on par with CoT using repeated '...' filler tokens. This raises alignment concerns: Using filler, LMs can do hidden reasoning not visible in CoT🧵
Truly enjoyed this interdisciplinary (CS + Psych + Philosophy) workshop today at Princeton as someone doing interdisciplinary research on human understanding of AI systems. Thank you @TaniaLombrozo and Stephen Grimm for organizing and the speakers for the amazing discussion!
One thing that I started doing at OpenAI is that I created a policy for myself to be *100% transparent* with my manager about everything. It seems obvious and weird to say aloud, but I bet most people don’t actually do this. But once I started doing it, I realized there are a lot…
So happy to defend my PhD thesis and couldn't have done it without a 1 of 1 advisor @david_sontag and an incredible committee @erichorvitz @arvindsatya1 @roboticwrestler
Congratulations to Dr. Hussein Mozannar. @HsseinMzannar Strong dissertation research exploring multiple dimensions of human-AI collaboration. @david_sontag @roboticwrestler @arvindsatya1 @MIT
Congrats @HsseinMzannar! It’s been a pleasure to collaborate the past few years and I’m excited to see what you do next!
Congratulations to Dr. Hussein Mozannar. @HsseinMzannar Strong dissertation research exploring multiple dimensions of human-AI collaboration. @david_sontag @roboticwrestler @arvindsatya1 @MIT
📢Excited to share this #CHI2024 paper examining whether and how LLM-powered conversational search creates “generative echo chambers”. We must understand how LLMs and LLM-powered applications may reshape people’s information consumption 🔗arxiv.org/abs/2402.05880
Concerns have long been raised about search systems creating filter bubbles and echo chambers, contributing to a divided society. What if LLM-powered conversational search makes these problems worse? 📢New paper accepted to #CHI2024: arxiv.org/abs/2402.05880 #LLMs #NLProc #HCI
🎉Happy to share our paper on generative echo chamber received a #CHI2024 Best Paper Award🏆 Congratulations to @nikhilsksharma @ZiangXiao
📢Excited to share this #CHI2024 paper examining whether and how LLM-powered conversational search creates “generative echo chambers”. We must understand how LLMs and LLM-powered applications may reshape people’s information consumption 🔗arxiv.org/abs/2402.05880
@UtdParadigm Mind boggling how so many don’t get the sarcasm here 😮💨💀
New Anthropic research: we find that probing, a simple interpretability technique, can detect when backdoored "sleeper agent" models are about to behave dangerously, after they pretend to be safe in training. Check out our first alignment blog post here: anthropic.com/research/probe…
Pardon my self-promotion! In my recent work, I measured how chatbots can drift away from its system prompts within 8 rounds of discourse. Then I introduced a training-free hack in the attention heads to help the model focus on privileged/system prompts. x.com/ke_li_2021/sta…
Introducing the Instruction Hierarchy, our latest safety research to advance robustness for prompt injections and other ways of tricking LLMs into executing unsafe actions. More details: arxiv.org/abs/2404.13208
@giangnguyen2412 Tabular data has the advantage that it often contains highly unique sequences of characters, which helps to detect memorization. Of course, it would be important to extend our analysis to other data modalities.
New work studying memorization and in-context learning capabilities of LLMs. The work proposes a series of tests for memorization and it finds indications of sota models exposing verbatim tabular data during generation. Such tests can be used for decoupling the impact of model…
Should we trust LLM evaluations on publicly available benchmarks?🤔 Our latest work studies the overfitting of few-shot learning with GPT-4. with @HarshaNori Vanessa Rodrigues @besanushi and Rich Caruana Paper: arxiv.org/abs/2404.06209 More details👇 [1/N]
So, it's not true that these models can only perform well on tasks seen during training. However, there seems to be a performance premium for tasks seen during training.
We then use the cutoff date of the training data and compare the few-shot learning performance on datasets seen during training to the performance on datasets released after training.
With this strategy, we are able to show that GPT-3.5 and GPT-4 have seen many tabular datasets during (pre-)training.
@HarshaNori @besanushi Unfortunately, it's challenging to know what data GPT-4 has seen during training. We use the phenomenon of memorization, where the model regurgitates parts of its pre-training data verbatim.
@HarshaNori @besanushi We first need to get our hands on data that the LLM has seen during training and data that it cannot have seen during training.
@HarshaNori @besanushi LLMs have seen tons of data during pre-training. Does this lead to invalid performance estimates? We study this question in the context of few-shot learning with tabular data.
Should we trust LLM evaluations on publicly available benchmarks?🤔 Our latest work studies the overfitting of few-shot learning with GPT-4. with @HarshaNori Vanessa Rodrigues @besanushi and Rich Caruana Paper: arxiv.org/abs/2404.06209 More details👇 [1/N]