Koby_Loby's profile picture. PhD student in NLP @UKPLab
Scientific documents, document structure, linking

Jan Buchmann

@Koby_Loby

PhD student in NLP @UKPLab Scientific documents, document structure, linking

Jan Buchmann reposted

»Attribute or Abstain: Large Language Models as Long Document Assistants« by @koby_loby, @XiaoL558286, @IGurevych 📰 arxiv.org/abs/2407.07799 💻github.com/UKPLab/arxiv20… (4/🧵) #EMNLP2024

UKPLab's tweet image. »Attribute or Abstain: Large Language Models as Long Document Assistants«
by @koby_loby,  @XiaoL558286,  @IGurevych

📰 arxiv.org/abs/2407.07799
💻github.com/UKPLab/arxiv20…
(4/🧵) #EMNLP2024

Jan Buchmann reposted

📢 10 papers authored or co-authored by UKP members have been accepted for publication at #EMNLP2024 in Miami! 🇺🇸 And 2 papers from CL and TACL will be presented at the conference 🎉 Congratulations to everyone involved! 💐 – be sure to check them out in this thread (1/🧵):

UKPLab's tweet image. 📢 10 papers authored or co-authored by UKP members have been accepted for publication at #EMNLP2024 in Miami! 🇺🇸
And 2 papers from CL and TACL will be presented at the conference 🎉

Congratulations to everyone involved! 💐 – be sure to check them out in this thread (1/🧵):

Jan Buchmann reposted

Please drop by our poster tomorrow and talk to my co-authors about how IVON can directly optimize a variational objective with similar performance and cost as Adam but many things on top (uncertainty, model adaptation, diagnosis, ...) Thu. 13:30-15:00, hall c 4-9, poster #1402

🤔 Variational learning is often thought to be impractical 🔥 Plot twist: it actually works better than Adam! Meet IVON, a new optimizer that brings the best out of variational learning – 🧵 (1/9) #NLProc #ICML2024 📰 arxiv.org/abs/2402.17641 youtu.be/TRNYnRRJBRg

UKPLab's tweet card. Variational Learning is Effective for Large Deep Networks [ICML 2024...

youtube.com

YouTube

Variational Learning is Effective for Large Deep Networks [ICML 2024...



In case you are looking for it: @eaclmeeting commitment link is here: openreview.net/group?id=eacl.…


Jan Buchmann reposted

Pointwise V-usable information (PVI) excels in many #NLProc tasks. But fine-tuning #LLMs with it is very time-consuming 🐌 Is in-context PVI the necessary next step? Yes! 🎉Check out our empirical analysis accepted at #EMNLP2023 – and this 🧵 (1/7) 📄 arxiv.org/abs/2310.12300

UKPLab's tweet image. Pointwise V-usable information (PVI) excels in many #NLProc tasks. But fine-tuning #LLMs with it is very time-consuming 🐌 Is in-context PVI the necessary next step?

Yes! 🎉Check out our empirical analysis accepted at #EMNLP2023 – and this 🧵 (1/7)

📄 arxiv.org/abs/2310.12300

Jan Buchmann reposted

Are Emergent Abilities in Large Language Models just In-Context Learning? Spoiler: YES 🤯 Through a series of over 1,000 experiments, we provide compelling evidence: arxiv.org/abs/2309.01809 Our results allay safety concerns regarding latent hazardous abilities. A🧵👇 #NLProc

UKPLab's tweet image. Are Emergent Abilities in Large Language Models just In-Context Learning?

Spoiler: YES 🤯

Through a series of over 1,000 experiments, we provide compelling evidence: arxiv.org/abs/2309.01809 

Our results allay safety concerns regarding latent hazardous abilities.

A🧵👇 #NLProc

Hey @signalapp, why do you keep insisting that I turn on notifications? Being donation funded and tracking-less (which I support very much), signal shouldn’t need to get me hooked, right?

Koby_Loby's tweet image. Hey @signalapp, why do you keep insisting that I turn on notifications? Being donation funded and tracking-less (which I support very much), signal shouldn’t need to get me hooked, right?

Jan Buchmann reposted

Some ppl have asked why we’d expect larger language models to do worse on tasks (inverse scaling). We train LMs to imitate internet text, an objective that is often misaligned w human preferences; if the data has issues, LMs will mimic those issues (esp larger ones). Examples: 🧵


Great tutorial by ⁦@PetarV_93⁩ at ⁦@CambridgeEllis⁩ ML Summer School!

Koby_Loby's tweet image. Great tutorial by ⁦@PetarV_93⁩ at ⁦@CambridgeEllis⁩ ML Summer School!

Jan Buchmann reposted

GPT-3 is amazing at complex tasks like creative writing and summarizing. But it's surprisingly bad at reversing words. 🤔 The reason is that GPT-3 doesn't see the world the way we humans do. 👀 If you teach it to reason, it can get around its limitations to get really good. 💡

npew's tweet image. GPT-3 is amazing at complex tasks like creative writing and summarizing. But it's surprisingly bad at reversing words. 🤔

The reason is that GPT-3 doesn't see the world the way we humans do. 👀

If you teach it to reason, it can get around its limitations to get really good. 💡

Great talks at #SciNLP about exciting applications of NLP research @andycrouse77 @NCBI !


Jan Buchmann reposted

Volksentscheid, WIR KOMMEN! 💛💜 Mit sagenhaften 343.591 Unterschriften🎉 🎈Aufgrund der Anzahl an Unterschriften (schon die vorläufige Zahl ist ein Rekord für Volksbegehren in Berlin) gehen wir davon aus, dass es zum Volksentscheid am 26.09. kommt! 💪

dwenteignen's tweet image. Volksentscheid, WIR KOMMEN! 💛💜 Mit sagenhaften 343.591 Unterschriften🎉

🎈Aufgrund der Anzahl an Unterschriften (schon die vorläufige Zahl ist ein Rekord für Volksbegehren in Berlin) gehen wir davon aus, dass es zum Volksentscheid am 26.09. kommt! 💪

Really interesting, well-researched (and long) article on the origin of SARS-CoV2 and the dangers of gain-of-function research on viruses. Bottomline: The lab escape theory seems to be more plausible than reported by some. thebulletin.org/2021/05/the-or…


Jan Buchmann reposted

I'm back from a week at my mom's house and now I'm getting ads for her toothpaste brand, the brand I've been putting in my mouth for a week. We never talked about this brand or googled it or anything like that. As a privacy tech worker, let me explain why this is happening. 🧵


Jan Buchmann reposted

"An international effort to produce & distribute mRNA-based Covid-19 vaccines worldwide" English version of a petition first proposed by @axelkahn to coordinate an emergency effort to produce and distribute mRNA-based Covid vaccines to the entire world. change.org/p/president-jo…

change.org

Sign the Petition

An international effort to produce & distribute mRNA-based Covid-19 vaccines worldwide.


So ein Blick in die mögliche Zukunft ist doch ein bisschen gruselig... taz.de/Ausbreitung-de… via @tazgezwitscher


Jan Buchmann reposted

Die Lage der Geflüchteten auf Lesbos bleibt so dramatisch wie nach dem Brand von vor einem Monat. Statt Feuer ist es nun der Regen, der die ohnehin prekäre Situation der Menschen nochmal verschlimmert. Die Europäischen Staaten dürfen diese katastrophalen Zustände nicht hinnehmen!

Erster Herbststurm im neuen #Moria. Schaut dieses Video, leitet es weiter. Eigentlich braucht es keine weiteren Kommentare. Es ist so unwürdig. Und viel zu wenige kriegen es mit. #leavenoonebehind



Das Auto ist immer noch viel zu dominant in Deutschlands Städten: taz.de/Verbot-der-Pop…


Jan Buchmann reposted

Heute wurden Menschen in #Moria von einem rechten Mob mit Steinen beworfen. Jetzt hat der Mob ein Feuer gelegt, weil er offenbar die 14.000 Geflüchteten im Camp verbrennen will. Völliger Wahnsinn schon wieder und kaum jemand kriegts mit. #LeaveNoOneBehind


Jan Buchmann reposted

212 ermordete Umweltaktivist*innen: Einer der viele Gründe, gegen Kohleverstromung und Kapitalismus zu kämpfen! Trotz 64 Morden in Kolumbien bezieht Deutschland weiter 1/3 der Steinkohle aus dem Land. Sofortiger Importstopp & Kohleausstieg bis 2030! tagesschau.de/ausland/naturs…


Loading...

Something went wrong.


Something went wrong.