Matthew Finlayson ✈️ NeurIPS
@mattf1n
Followers
944
Following
662
Statuses
137
First year PhD at @nlp_usc | Former predoc at @allen_ai on @ai2_aristo | Harvard 2021 CS & Linguistics
Los Angeles, CA
Joined October 2013
Wanna know gpt-3.5-turbo's embed size? We find a way to extract info from LLM APIs and estimate gpt-3.5-turbo’s embed size to be 4096. With the same trick we also develop 25x faster logprob extraction, audits for LLM APIs, and more! 📄 Here’s how 1/🧵
6
79
360
RT @RobertTLange: Loving the #NeurIPS2024 'Beyond Decoding: Meta-Generation Algorithms for LLMs' workshop ❤️ by @wellecks @mattf1n @hailey…
0
25
0
I didn’t realize when making these diagrams that my Taylor example would be so timely 😂
In Vancouver for NeurIPS but don't have Taylor Swift tickets? You can still spend the day going through our tutorial reading list: - Tuesday December 10, 1:30-4:00pm @ West Exhibition Hall C, NeurIPS
0
0
9
RT @wellecks: We're incredibly honored to have an amazing group of panelists: @agarwl_ , @polynoamial , @BeidiChen, @nouhadziri, @j_foerst…
0
3
0
RT @jaspreetranjit_: Thank you so much @SpecNews1SoCal @jaskang21 for featuring our work on OATH-Frames: Characterizing Online Attitudes to…
0
7
0
RT @xiangrenNLP: Arrived in Philadelphia for the very 1st @COLM_conf! Excited to catch up w/ everyone & happy to chat about faculty/phd pos…
0
6
0
RT @harsh3vedi: I had a fantastic time visiting USC and talking about 🌎AppWorld ( last Friday!! Thank you, @swabhz,…
0
1
0
Just landed in Philly for @COLM_conf where I’ll be presenting my work on extracting secrets from LLM APIs at the Wednesday afternoon poster sesh. Please reach out if you wanna hang and talk about sneaky LLM API hacks, accountability, and the geometry of LLM representations!
Wanna know gpt-3.5-turbo's embed size? We find a way to extract info from LLM APIs and estimate gpt-3.5-turbo’s embed size to be 4096. With the same trick we also develop 25x faster logprob extraction, audits for LLM APIs, and more! 📄 Here’s how 1/🧵
0
10
55
@nthngdy @agarwl_ @COLM_conf @andreasgrv First off, really cool paper @nthngdy I’m excited to see you at COLM and talk about it! I agree that I don’t see a direct link, as there are many other confounding vars affecting small LM performance on math reasoning. If using greedy-like decoding I’d expect no effect from SMB.
0
0
2
1
0
0
@alisawuffles @JonathanHayase I wonder, could this be combined with identifying undertrained tokens to figure out how the training mixture differs from the bpe mixture?
Fishing for Magikarp: Automatically Detecting Under-trained Tokens in Large Language Models abs: code: Cohere presents a method for consistently identifying glitch tokens, across both open-source and closed-source LLMs.
0
0
6
RT @xiangrenNLP: Congratulations to the GDM @GoogleDeepMind team on their best paper award at #ICML2024 & Appreciate @afedercooper's shout…
0
8
0
Congratulations on this well-deserved award for a brilliant paper! And thank you for the shout-out :)
0
3
33
@andreasgrv @StatMLPapers Very cool! Mapping this to the problem in my paper their technique can be used to reconstruct a model’s embedding matrix knowing only a few of its entries and the model image. Funnily enough I just learned about ADMM while working on some follow up work
1
0
1
Jealous that Will gets to mess around with Rust as part of his research.
0
0
4
Grateful to be part of this multi-institution effort to document the state of LLM generation. Interested in decoding algorithms? Start by reading our survey paper 🤓
What do nucleus sampling, tree-of-thought, and PagedAttention have in common? They're all part of our new survey: "From Decoding to Meta-Generation: Inference-time Algorithms for Large Language Models"
0
1
31
RT @jaspreetranjit_: 📢 Can LLMs 🤖 assist social workers 👩🏻💻 in characterizing discourse on social issues? We introduce OATH-Frames: a res…
0
17
0