@ShahriarGolchin Profile picture

Shahriar Golchin

@ShahriarGolchin

Student Researcher @Google | CS PhD in ML/NLP @UArizona (opinions are mine)

Similar User
Woojin Ko photo

@woojin_ko

DuckWack photo

@duckwackIND

Adnan Harun DOĞAN photo

@adnanharundogan

DeepReel photo

@DeepReel_AI

Nyla Worker photo

@nyla_worker

Arjun Ashok photo

@arjunashok37

Swagatam Haldar photo

@swagatam2198

Nikhil Barhate photo

@nikhilbarhate99

Nassar Mohamed photo

@Qdvvdu

Luis Castano photo

@luis_castano_95

Adil Sheraz photo

@adilsheraz_

Shubham Bhardwaj photo

@4245Shubham

irfan bulu photo

@irfanbulu

Aritra Bhowmik photo

@AritraBhowmik6

Mohammad Raihanul Bashar photo

@rbfahim

Pinned

The mechanism behind in-context learning (ICL) has been unclear for a long time. We studied the behind-the-scenes memorization in ICL and found that there is a very strong correlation between this memorization and improved performance. arxiv.org/abs/2408.11546


Excited to share I've joined @Google as a Student Researcher! Ready to learn, innovate, and make an impact!


We (w/ @msurd) will be presenting our Spotlight paper at #ICLR2024 in Vienna next week. Drop by for some amazing intellectual exchanges on data contamination. Paper: arxiv.org/abs/2308.08493 Code: github.com/shahriargolchi… Media: thenewstack.io/how-to-detect-…

Tweet Image 1

Shahriar Golchin Reposted

Are your LLMs highly accurate, or simply contaminated? As the race to build the best LLM intensifies, clean evaluation is becoming more important than ever, yet contaminated LLMs and benchmarks obfuscate the real performance of models. Checkout our new work (comprehensive survey…


Shahriar Golchin Reposted

Do you have a paper about data contamination with some evidence reported on it? Consider submitting that evidence to the Data Contamination Evidence Collection. Any evidence is very welcome!

Can you imagine having all the evidence of data contamination gathered in one place? 📢As part of the CONDA workshop, we present the Data Contamination Evidence Collection, a shared task on reporting contamination. Available as a @huggingface space: hf.co/spaces/CONDA-W…



Shahriar Golchin Reposted

1/ Many cool papers coming from our lab recently! Here are just a few:


Shahriar Golchin Reposted

The #nlproc group at University of Arizona has grown too big, and my account doesn't do it justice. Please follow @LabCLU for updates about our group!


Shahriar Golchin Reposted

A brief but nice post on data contamination that discusses the work that @ShahriarGolchin, one of our great PhD students :), did: thenewstack.io/how-to-detect-…


Surprisingly, Andrej Karpathy revealed that GPT-4 isn't truly 67% proficient in coding (HumanEval). The lesser-known problem we uncovered earlier is that GPT-4 has higher levels of coding and reasoning data contamination than officially reported. arxiv.org/abs/2311.06233

Claude 3 takes on the Tokenization book chapter challenge :) context: twitter.com/karpathy/statu… Definitely looks quite nice, stylistically! If you look closer there are a number of subtle issues / hallucinations. One example there is a claim that "hello world" tokenizes into 3…



Shahriar Golchin Reposted

📢 Excited to announce that our Workshop on Data Contamination (CONDA) will be co-located with ACL24 in Bangkok, Thailand on Aug. 16. We are looking forward to seeing you there! Check out the CFP and more information here: conda-workshop.github.io


Thrilled to share that our "Time Travel in LLMs" paper has been accepted to #ICLR2024 as a Spotlight! w/ my awesome advisor @msurd #LLMs #DataContamination @iclr_conf

Data contamination suggests LLMs have possibly seen test data from downstream tasks. Our recent study introduces a novel method to replicate LLMs' training data, including downstream dataset instances, to aid in detecting data contamination. Read more: arxiv.org/abs/2308.08493



Shahriar Golchin Reposted

Thankfully, several works have been published on this topic since we submitted our paper for review: Time Travel in LLMs: Tracing Data Contamination in Large Language Models By: @ShahriarGolchin @msurd Paper: arxiv.org/abs/2308.08493


Loading...

Something went wrong.


Something went wrong.