@benbenhh Profile picture

Benjamin Heinzerling

@benbenhh

Postdoc @RIKEN_AIP, working on Natural Language Programming and Neuro-Linguistic Processing.

Similar User
Anna Rogers photo

@annargrs

Yonatan Belinkov photo

@boknilev

Barbara Plank photo

@barbara_plank

Ivan Vulić photo

@licwu

Ivan Titov photo

@iatitov

Rico Sennrich photo

@RicoSennrich

Julia Kreutzer photo

@KreutzerJulia

Lucie Flek photo

@lucie_nlp

Goran Glavaš photo

@gg42554

Jonathan Berant photo

@JonathanBerant

Marek Rei photo

@MarekRei

Mor Geva photo

@megamor2

Omer Levy photo

@omerlevy_

Edoardo Ponti photo

@PontiEdoardo

AdapterHub photo

@AdapterHub

Pinned

We've released a new version of our pretained byte pair embeddings in 275 languages. Now with pip install, automatic download of embeddings and sentencepiece models, convenient subword segmentation, and tons of pretty UMAP visualizations. nlp.h-its.org/bpemb


Benjamin Heinzerling Reposted

ITT: an OAI employee admits that the text-davinci API models are not from their papers. Until @OpenAI actually documents the connection between the models in their papers and the models released via APIs, #NLProc researchers need to stop using them to do research.

I agree. While OpenAI doesn't like talking about exact model sizes / parameter counts anymore, documentation should definitely be better. text-davinci-002 isn't the model from the InstructGPT paper. The closest to the paper is text-davinciplus-002. twitter.com/janleike/statu…



There is nothing "sudden" about the increase in validation accuracy. The apparent sudden increase is just an artifact of the x-axis log scale. A linear scale reveals a steady increase. That's still a really cool finding, but "grokking" is a bit of a misnomer

Tweet Image 1

Read a bit about Grokking recently. Here's some learnings: "Grokking" is a curious neural net behavior observed ~1 year ago (arxiv.org/abs/2201.02177). Continue optimizing a model long after perfect training accuracy and it suddenly generalizes. Figure:

Tweet Image 1


It's kinda ridiculous to have these million dollar LMs and then NLP researchers are like: the current SOTA prompt for recipe generation is "I want to cook a yummy meal" but we found that average recipe tastiness increased 1.3% if you append "#potatosalad" and a cow emoji


Benjamin Heinzerling Reposted

Our dataset, Semi-structured Explanations for Commonsense Reasoning (COPA-SSE), has been accepted to #LREC2022! HUGE thanks to my co-authors, @benbenhh, @pkavumba, and @inuikentaro 🙌 J'ai bien hâte de vous voir à Marseille 😄 (espérons...!)


A paper I'm reviewing unironically describes a particular deep learning method as "traditional" and cites work from 2017 to 2019 as evidence for the old age of this method


Solving NLP apparently involves global world domination. I think we'll have to stack quite a few more transformer layers until ACL supersedes the United Nations :( github.com/andreasvc/disc…

Tweet Image 1
This post is unavailable.

Papers on automatic summarization or information extraction usually imagine users with "information needs" facing "information overload", but I think the real use case is to obviate doomscrolling by condensing long feeds into screen-sized doses of existential dread. #nlp #nlproc


"lack of recognition for meta-research as a valid part of NLP, which, as we learned in writing this paper, makes it difficult to publish on it. In a way, NLP peer review... prevents research on NLP peer review." 😂 Great read by @annargrs and @IAugenstein arxiv.org/abs/2010.03863


Spam email: "You're being investigated by the United States of America for war crimes and torture involving ai google abuse" Ok, I'll admit that my code isn't the cleanest, but "war crimes and torture" is a bit harsh


"The history of AI [...] can be seen as a prolonged deconstruction of our concept of intelligence." Insightful article by @togelius AI will be yet another scientific discipline that shows humans are not as special as we might like to think.

A very short history of some times we solved AI togelius.blogspot.com/2020/08/a-very…



I have a joke about the observer effect but it's only funny if you don't read it.


Loading...

Something went wrong.


Something went wrong.