Leon Derczynski ✍🏻🌹☀️
@LeonDerczynski
Followers
6,351
Following
1,155
Media
1,902
Statuses
25,081
NLP/ML/language/security. Principal research scientist @NVIDIA , & Prof @ITUkbh . Views ostensibly professional. llmsec stan acct
Copenhagen / Seattle
Joined January 2012
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Cheney
• 352089 Tweets
Colombia
• 227368 Tweets
Packers
• 129498 Tweets
श्री गणेश
• 84928 Tweets
ENGFA ACTRESS 100M
• 82030 Tweets
Perú
• 73869 Tweets
1 Day Left For Avataran Diwas
• 71886 Tweets
Saquon
• 71462 Tweets
गणपति बप्पा
• 63373 Tweets
#GaneshChaturthi
• 55859 Tweets
Hope Love
• 55775 Tweets
Jalen
• 55322 Tweets
#ATIPASHOPxCHARLOTTESNACK
• 50187 Tweets
Giants
• 50176 Tweets
Green Bay
• 37328 Tweets
Philly
• 33073 Tweets
भगवान गणेश
• 29280 Tweets
Jordan Love
• 27735 Tweets
Duke
• 25989 Tweets
Lord Ganesha
• 24569 Tweets
#二度と撮れない画像を貼れ
• 21852 Tweets
#GanpatiBappaMorya
• 20846 Tweets
Fritz
• 16838 Tweets
Durán
• 16640 Tweets
Lucho
• 14548 Tweets
Tiafoe
• 13627 Tweets
Jayden Reed
• 13614 Tweets
ビッグラン
• 12914 Tweets
イコラブ
• 12093 Tweets
Pinned Tweet
Proud to announce:
💫 garak - an LLM vulnerability scanner💫
🔎 Check if a model is susceptible to common attacks
🦜 Supports HuggingFace, OpenAI, ggml, Cohere, ...
🔧 >70 probes: prompt injection, false claims, toxicity, encoding evasion, ..
6
77
341
'I don't really trust papers out of "Top Labs" anymore'
47
554
4K
machine learning researchers learn to optimise their own best paper rate through collusion and other unregulated mechanisms
42
205
1K
Thanks, reviewer 2
𝚁𝚎𝚊𝚜𝚘𝚗𝚜 𝚝𝚘 𝚛𝚎𝚓𝚎𝚌𝚝
--------------------------
𝙸 𝚌𝚊𝚗𝚗𝚘𝚝 𝚏𝚒𝚗𝚍 𝚊𝚗𝚢 𝚛𝚎𝚊𝚜𝚘𝚗 𝚝𝚘 𝚛𝚎𝚓𝚎𝚌𝚝 𝚝𝚑𝚒𝚜 𝚙𝚊𝚙𝚎𝚛 𝚊𝚝 𝚊𝚕𝚕 (𝚝𝚑𝚎𝚛𝚎 𝚒𝚜 𝚊 𝟷0 𝚠𝚘𝚛𝚍 𝚖𝚒𝚗𝚒𝚖𝚞𝚖 𝚏𝚘𝚛 𝚝𝚑𝚒𝚜 𝚜𝚘 𝙸 𝚓𝚞𝚜𝚝 𝚔𝚎𝚎𝚙 𝚠𝚛𝚒𝚝𝚒𝚗𝚐)
5
14
662
using a 3d printer to "write" homework generated by chatgpt. your move, examiners
21
104
562
machine learning researchers debating theories of consciousness
11
91
537
ChatGPT not best at many language tasks. It's outranked by other systems on many NLP benchmarks in current evaluation. For 77.5% of tasks examined, other systems are better than ChatGPT.
38
82
371
it's 2024 and we're still having to say this lol
cos sucks if you have more than say three or four dimensions
Is Cosine-Similarity of Embeddings Really About Similarity?
Netflix cautions against blindly using cosine similarity as a measure of semantic similarity between learned embeddings, as it can yield arbitrary and meaningless results.
📝
28
398
2K
12
22
332
ELIZA designer Joseph Weizenbaum observed: “What I had not realized is that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.”
4
88
312
@qntm
@SwiftOnSecurity
i know a head of national sales at google who only knows how to use "recent" docs. that's their filesystem
7
18
313
Efficient NLP methods - an up-to-date survey, to appear in TACL. We cover efficiency wrt:
* Data
* Model design
* Pre-training
* Fine-tuning
* Inference & compression
* Hardware utilization
* Evaluation
* Model selection
This was a blast to co-produce!
6
81
307
How can NLP be more efficient? We can't blindly scale forever. This survey I'm lucky to be part of presents current efficiency research at many stages of the NLP process, and actionable advice to practitioners for making their NLP more efficient.
#nlproc
8
78
306
the replicability crisis has been real for over a decade, but replicability still is nowhere near a publishing requirement in our field. even hugging face are ahead of our publishing venues in this and they're just.. 3? 4? years old (from here )
4
22
265
Reddit bans use of Reddit content for machine learning unless each participating user gives individual explicit informed consent
Updated API ToS:
11
54
224
"Dear Leon, I’m happy to say that the CS department has decided to offer you a position as associate professor" -- What a privilege! Only 6.5 hours til the bubbles can be opened :)
32
0
205
@JeffDean
Submitting insufficiently PR-washed work *for review* is a summary terminable offence at Google, especially if you have a minority background? Got it, thanks for the clarification
2
3
172
We release T5 large language models for Danish, the biggest monolingual models for the language. We had to use a bunch of tricks to get this to work with modest data and hardware; here's how:
#dktech
7
18
182
There's an NLP task, "detoxification", where the goal is to rephrase a statement that uses toxic language into one that doesn't use toxic language. This is a really odd task laden with harms. ⬇️
9
26
180
if a paper's title starts "Towards", it means they didn't get there
16
23
174
you can even get strong papers with new results by doing replicability studies - here's an example, - and there are more every year. so why do we accept so many results on trust, with no replication? because there's no incentive to confirm findings
2
10
159
Streamers trying to game location-based content recommendation algorithms by going to a wealthy neighbourhood to stream. For donations. Under a bridge. Sat on the ground.
10
31
152
Updated with now 90 openly-available abusive language datasets covering 17 languages!
#nlproc
2
35
151
New paper! 📝
Text data can contain harmful things, but unlike other disciplines, we don't have protocols for handling or presenting harmful material🛢️. This paper with
@hannahrosekirk
@Abebab
@bertievidgen
presents the risks & some solutions. 🥼🥽
3
23
146
What toxic nonsense Google have chosen to autoinclude in their search results. Irresponsible use of AI with Google then taking no ownership over the consequences of their actions when told about them - doubly embarrassing.
8
24
143
@WendaXu2
Oh yeah, I've had this a few times. Officer is correct. I always make sure I have a book, now, to pass the time. LAX is particularly rude, ORD has an even less comfortable waiting area - recommend SFO or SEA for ingress, they tend to be polite and efficient
5
3
143
Love this
* attacker can inject backdoors by issuing very few malicious instructions among thousands of gathered data
* attacker can achieve > 90% success
* instruction attacks show resistance to existing inference-time defense
Any mTurker has control!
4
30
140
Toxic & misinformative top-level google summaries: 🧵
5
42
140
Integrating electricity consumption into loss during hyperparameter search finds LLMs that are pareto-efficient for power and benchmarks, doing the same as bigger models for less.
"Hyperparameter Power Impact
in Transformer Language Model Training" -
7
23
131
Debunking the idea that using linear activations means a computational neural net simplifies to an affine function of the input layer. We don't compute with continuous value representations, so the simplification is missing. Quantisation is powerful!
6
13
131
@RealSexyCyborg
@pixelgem
Please understand that changing the question derails the conversation -- a technique frequently used in classic Soviet propaganda -- and deliberately derailing a conversation about gender equity is inescapably sexist. That's what you're doing here.
4
1
128
@adityas129
oh man i used to do this!
* dedicate time to working on your emotional and mental health
* drop every single extra project
* close the computer and turn off the phone every now & then
* give focus *only* to non-work things outside of 9-5
* don't have kids. DO NOT have kids
9
0
126
Here's a list of hate speech datasets for
#nlproc
- many non-English, many completely open access.
8
52
123
New project funded: Verif-AI, on misinformation detection across languages; 30 months. I'll be hiring in Copenhagen.
#nlproc
#misinformation
4
26
122
Imagine being able to conduct your entire career in your first language, and in just one language! It must be amazing
6
7
119
wow how do people who are parents also manage to have professional careers and also survive. i am in awe
20
13
120
Women in Denmark experience the most violence out of *all EU states*, including both from their partners and from outside a relationship. Analysis is thorough and addresses reporting biases etc. well:
#twitterhjerne
#dkpol
22
60
110
If GPT3 took 50 petaflop-days to train , w. GPUs at 10^8 flops per watt , so those 1.2E18 flop-hours used 12 GWh to train? E.g. 12 hours of a whole nuclear reactor? At 0.73kg per kWh that's.. 8.8 megatons of CO2?!
#sanitycheck
#nlproc
11
33
116
@thechosenberg
To get more back when they die. Paying down principal over longer period works perfectly for the bank
3
0
111
We can't do reproducible science on closed models
Update: A day later and I can't reproduce it using the very same prompt. This sucks 🫤
13
1
56
8
4
108
@ClaudiaStokes
This casts the net inappropriately broad, doesn't it. Met my wife at a university bar while she was a PhD candidate and I was a postdoc, in different departments; now I'm a professor, she's a former grad student, and nothing was wrong.
23
0
107
Modeling the rhetoric of persuasion, with application to propaganda and disinfo detection -- accepted to NLP for Positive Impact at EMNLP, with Amalie Brogaard Pauli and Ira Assent. Really excited about this direction
5
7
99
@SwiftOnSecurity
Right. From our years of work on fake news (e.g.
@PhemeEU
), it become evident that planted stories that "feel" right and sound edgy are those consumed most readily
0
14
89
Seems the majority of
#COLING2018
papers are presented with code/data, on github or even "pip install <x>". This is the first
#nlproc
conference I've seen where sharing is the overwhelming norm; not sharing is now the unusual, minority behavior.
#openscience
0
32
97
Clever Hans: the horse we thought could do arithmetic, but really was relying on other signals. Really enjoyed this blog post on NLP, clever hands and what to do instead of leaderboards.
#nlproc
2
44
95
Possibly the most compelling call for papers I've ever seen
8
1
91
🌶️🌶️🌶️
from collections import defaultdict
nested_dict = lambda: defaultdict(nested_dict)
nest = nested_dict()
nest[0][1][2][3][4][5] = 6
7
3
93
@bryancsk
there is ice *everywhere* here and I don't know why. some random floor of a hotel? you can get rooms, stairs, elevators, a fire hose, and ICE. gas station? we have gas, diesel, and ICE
#respectallcultures
7
2
84
1
0
84
Our paper on data for abusive language detection is here. Highlights:
- Best practices for making abusive language data
- Advice for machine learning practitioners
- Emerging research problems
- Comprehensive corralling of the state-of-the-art
#nlproc
0
19
77
Language detection is hard, especially between similar languages. For example, the best tools still mix up Danish and Norwegian. Here's a new challenge for distinguishing between six similar Nordic languages, with dataset and baseline. Enjoy!
#nlproc
3
18
77
@fToRrEeEsSt
@mina_hamedi
@awkwardlefty_cj
Yeah, usually people are super grateful for uninvited criticism! :)
1
0
74
Always nice to have something local to celebrate! Another hyggelig day in Denmark :)
3
2
77
Funeral today. That leaves us with 0/8 grandparents, 1/4 parents (disabled and abroad), two us, and 1/n children. People with families that help out, you are fortunate!
17
0
75
To the fragile men arguing against the findings' validity: What's your end goal here? Who are you trying to silence, and what good would that serve? Sometimes the best first step to improving a problem is just to listen to and absorb the facts. Read the report.
3
3
74
@ChrisMurphyCT
* It was explicitly built into the model.
* The people building it programmed it to use data including complicated chemistry.
* It has zero agency and didn't decide a single thing for itself.
2
0
73
acl: just saw that our author response lifted a 2 -> 3 and a 2.5 -> 3.5, and then the paper was accepted to the main conf (scores out of five). this pleases me.
0
0
76
Time after time, datasets without documentation turn out to be stuffed with toxic, unjust, and illegal data. The models trained on them propagate the harms. This time, it's the Stable Diffusion data that amazingly has no docs. -- Really thought this lesson had been learned by now
Would be great if there was an effort to create datasheets, data statements or related documentation for the entire dataset.
4
22
106
4
13
75
Slides for my talk on Structured LLM Red-teaming: How do people do this new activity?
(Spoiler: it's qualitative NLP/AI research)
w/
@NannaInie
3
16
75
Working on offensive speech / hate speech? Don't stick to just English; look at the huge range of resources & languages here: .
#nlproc
#hatespeech
3
27
75
@generativist
read this as "i REFUSE to deal with the case where they turn two during the trip but land one again"
3
0
73
@spectatorindex
The consequences are expected to be that Sweden and Finland will be NATO members
1
0
67
@jaschasd
Love how hitting the fidelity limit at the end makes it look like reality is breaking down
1
1
70
The Impossible Safety of Large AI Models
thank you. even with my AI Ethics hat off and my CS/infosec hat on, this seemed super likely.
"this paper systematizes our knowledge about the fundamental impossibility of building arbitrarily accurate and secure machine learning models"
In this paper,
@L_badikho
@Sadegh_Farhad
@rlt_s
other colleagues and I, who have written many of the AI security papers published at top venues, reviewed the AI security literature and conclude that securing large models is mathematically
#impossible
.
5
56
164
1
17
69
Back to full time industry work today for the first time in 19 (!) years. I don't have a "quit lit" about academia, it was and is a weird, special place where people do wonderful things. Besides, I'll keep my uni affiliation! OK, see you out there in the usual places :)
5
1
69
@yoavgo
reward dataset papers, so people bother producing them. i've made a ton of smaller datasets to enable research on new problems and framings - it's way too much effort, you won't get a good paper, and it'll work against one's career. this is a cultural problem.
5
7
67
nothing made me more grateful for academia than moving to industry. the amount of academic research we consume, implement, and rely on to grow is astounding. i've never pored over such Volumes of papers before. thank you, profs postdocs grad students and undergrads, you rock! ❤️
2
5
66
Our research at
#acl2021nlp
covered in Wired. Really great article with plenty of the required nuance.
1
10
68
New paper: "The Danish Gigaword Corpus" - a clean, open, billion-word corpus of Danish stretching across domains, modalities, time, dialect, ... To be presented at
#nodalida
w/
@ManuelCiosici
#nlproc
2
17
67
"Efficient Transformers with Dynamic Token Pooling"
A family of language models that pool variable-size segments of tokens in intermediate layers to enhance the efficiency & performance
1
19
65
We looked at misogyny online. Getting the process right is important. We present: a detailed annotation methodology & codebook; a comprehensive taxonomy of labels for annotating misogyny; and a high-quality dataset of annotated posts; in this ACL paper.
3
11
65
@_akhaliq
A bit is literally a binary digit, not a ternary digit, but I appreciate avoiding the alternative portmanteau
7
2
63
Big text datasets should be cleaned up a bit, for e.g. hate and abuse, before distribution. People who collate and create datasets are responsible for the selection of content in those datasets. So, they should do the cleaning.
6
14
63
@pmddomingos
ACL were happy to publish techniques for automatically unveiling Chinese dissidents despite ethical objections during review.
3
1
60
Super happy to announce ACL SIGSEC, a Special Interest Group on NLP & LLM Security
* Everyone is welcome
* Regular invited talks (see the schedule!)
* Explicitly connecting industry and academia
* Annual open research workshop
Join us!
2
24
61
1. LLMs are very useful
2. LLMs present concrete dangers & risks
-> These two claims don't conflict
People who state one of these claims don't actually refute the other, despite the assumptions around in the discourse here (this IS twitter, right?). Most people accept both.
4
5
59
Online hate pushes people out of conversations and polarises online societies. We took a scientific approach using machine learning to analyse online hate against women, the largest target group.
Video ->
#hatespeech
#misogyny
3
14
60