
Data & AI Ethics @LBGplc | #greentech @cityoflondon | systemic #AI risks @CambridgeJBS | ex- @UniofOxford, @uniofwarwick, @DefenceHQ
Joined December 2014
- Tweets6.1K
- Following3.5K
- Followers924
- Likes6.9K
Thanks to all my favourite people who contributed, and special thanks to @sarameghanbeery & @mmbronstein 🙏
#AIHelpsUkraine was only possible thanks to the contribution of an incredible group of speakers. Big thanks to all of them!
7
3,969
Nataliya Tkachenko, PhD in #LLMs retweeted
just spotted on the Singapore skyline, coming to an iclr near u tmrw... #superdiff #ICLR2025 🦹♀️
(Hall 3 + Hall 2B #141 @ 3 pm)
🧵(1/5) Have you ever wanted to combine different pre-trained diffusion models but don't have time or data to retrain a new, bigger model?
🚀 Introducing SuperDiff 🦹♀️ – a principled method for efficiently combining multiple pre-trained diffusion models solely during inference!
We provide a new approach for estimating density without touching the divergence. This gives us the control to easily interpolate concepts (logical AND) or mix densities (logical OR), allowing us to create one-of-a-kind generations! ⚡🌀🤗
This is all due to an amazing team: @martoskreto @lazar_atan @bose_joey @AlexanderTong7
📄Paper: arxiv.org/abs/2412.17762
💻Code: github.com/necludov/super-di…
🤗HuggingFace: huggingface.co/superdiff
Nataliya Tkachenko, PhD in #LLMs retweeted
Google has never been good at marketing its products, but boy is it good at technology. Very bullish on Gemini, the Gemma line of open models, and the JAX/Keras/TPU stack
Nataliya Tkachenko, PhD in #LLMs retweeted
The colours of spring at Exeter College 🌸🌼
📷 Instagram | ObservingOxford
Nataliya Tkachenko, PhD in #LLMs retweeted
New paper from FAIR+NYU:
Q: Is language supervision required to learn effective visual representations for multimodal tasks?
A: No.
⬇️⬇️⬇️
Nataliya Tkachenko, PhD in #LLMs retweeted
Reasoning models offer a groundbreaking approach to interpretability by analyzing the content of their Chain-of-Thoughts.
These models reveal undesired behaviors within their own Chain-of-Thoughts.
This is galaxy brain — the models themselves lay bare their misalignment in plain English!
Monitoring Chain-of-Thoughts represents an entirely new branch of safety!
openai.com/index/chain-of-th…
Nataliya Tkachenko, PhD in #LLMs retweeted
Oxford University has announced plans to expand its artificial intelligence (AI) offering and capabilities with @OpenAI.
Students and faculty staff will have access to research grant funding and cutting-edge AI tools that will enhance teaching, learning, and research.
Nataliya Tkachenko, PhD in #LLMs retweeted
We are looking for three AI Research Engineers and Scientists to join us in our London office at @SilurianAI to build physics foundation models for weather and energy. Apply below and feel free to reach out if you have any questions:
ycombinator.com/companies/si…
Nataliya Tkachenko, PhD in #LLMs retweeted
This is cool work on mechanistic interpretability through graph theory (it was about time!), coming out as one of the mini-projects done by students of our Geometric DL Master's course @Cambridge_Uni.
(@NeelNanda5 @ArthurConmy you may enjoy this one!)
Our first attempts at mechanistic interpretability of Transformers from the perspective of network science and graph theory!
A wonderful collaboration with superstars @elb4tu, Deepro Choudhury, @pl219_Cambridge as part of the Geometric Deep Learning class at @Cambridge_CL!
Nataliya Tkachenko, PhD in #LLMs retweeted
🌊 STURM-Flood: a curated dataset for deep learning-based flood extent mapping leveraging Sentinel-1 and Sentinel-2 imagery
📢 [New Article] STURM-Flood a curated dataset for deep learning-based flood extent mapping leveraging Sentinel-1 and Sentinel-2 imagery by Nicla Notarangelo @NiclaMN , Charlotte Wirion & Frankwin van Winsen
👉Article link: doi.org/10.1080/20964471.202…
💌 #Flooding is a growing challenge, requiring precise, data-driven solutions. STURM-Flood provides an open-access, AI-ready dataset, integrating Sentinel-1 & Sentinel-2 imagery with Copernicus EMS ground truth to improve flood mapping and #risk assessment.
#AIforGood #FloodMapping #EarthObservation #ClimateResilience #RemoteSensing #STURMProject #STURMFlood #OpenData #AI #disaster
Nataliya Tkachenko, PhD in #LLMs retweeted
We have another exciting #EthicsinAI Lunchtime Research Seminar this week, presented by @IasonGabriel Senior Staff Research Scientist at @GoogleDeepMind 🌟
🗓 When: Wednesday, 29 January @ 11:30am
📍 Where: In-Person or Online
📋 Title: A Matter of Principle? #AI Alignment as the Fair Treatment of Claims
Join us for an insightful seminar where Iason delves into the critical challenges of AI alignment. Registration link below!
Nataliya Tkachenko, PhD in #LLMs retweeted
Federico Barbero @fedzbar: "Sheaf Neural Networks with Connection Laplacians" (ICML'22 TAGML)
w/ @crisbodnar @ocariz__ @mmbronstein @pl219_Cambridge
proceedings.mlr.press/v196/b…
Nataliya Tkachenko, PhD in #LLMs retweeted
The Otters are back!
We were sent this gorgeous video of the otters in the Cherwell by Magdalen Fellow John Nightingale - they look like little Loch Ness Monsters 🥹🥹
Nataliya Tkachenko, PhD in #LLMs retweeted
1/ Today the Prime Minister published my AI Opportunities Action Plan for the UK and committed to implement all its recommendations. A quick thread on why this is important, what we need to do now - and how you can help
Nataliya Tkachenko, PhD in #LLMs retweeted
The @PyTorch team are working on a new super important tool:
github.com/pytorch-labs/torc…
This repository implements techniques for doing a per-step fault tolerance so you can keep training if errors occur without interrupting the entire training job.
Some big companies already have this as a proprietary solution, so it's great this super important solution is worked on for the rest of us.
It's a prototype at the moment that already works for DDP
Nataliya Tkachenko, PhD in #LLMs retweeted
Dear friends and followers, I'm sharing here the Nobel prize acceptance speech I gave on behalf of Simon Johnson, James Robinson and myself. Thank you everybody.
Your Majesties, Your Royal Highnesses, Excellencies, Dear laureates, Ladies and gentlemen,
On behalf of Professor Johnson, Professor Robinson, and myself, I wish to thank the Royal Swedish Academy of Sciences and the Nobel Foundation for recognizing our work on the origins of economic and political institutions and the ways in which they shape global inequality.
We are heartened by this recognition not just because of the great honor it confers on our work, but also because of the encouragement it provides to many young scholars working on these topics. This has been a collective effort by many people, including some in this room, demonstrating how combining economics and history can shed light on the big questions of social science.
Our research showed that inclusive institutions – such as democratic participation, the rule of law, secure property rights, and broad access to economic opportunities – are a major contributor to shared prosperity in all parts of the world.
Today, 25 years later, the same institutions are in danger. Democracy is threatened almost everywhere, and support for democratic institutions is at a modern low. And now the world must confront the problems brought by a warming climate, new global power relations, and rapidly aging populations. Perhaps most importantly, AI promises to disrupt everything, everywhere, all at once.
History again has lessons for us. Shared prosperity has only occurred during some periods of human history, and it has never been an automatic process. Inclusivity is key. If, in the name of progress, the politically powerful trample on people’s rights and voices, if the privileged start to see the rest of society as expendable, if the elite mistakenly convince themselves that only their ideas and talent matter, this will shatter the institutions that underpin shared prosperity.
Institutions are always about choices.
What worries us also gives us hope. We can build better institutions and choose a direction for technology that creates more good jobs. But this has to be a collective effort, too – by all of us in this room and beyond, including young researchers as they venture into new areas to ask the big questions about institutions, technology, inequality, and how we can secure shared prosperity in the age of AI.
Nataliya Tkachenko, PhD in #LLMs retweeted
Mining eats into more of the world’s forests. Nearly 1.4mn ha of tree cover was lost between 2001 and 2020 in areas where mining and related activities have taken place on.ft.com/3Yuy4aH
Nataliya Tkachenko, PhD in #LLMs retweeted
I'll be at #NeurIPS2024 next week!
- Reach out to chat about the state of AI regulation, data and data transparency, public resources, and mechanisms for AI accountability,
- Come to our EvalEval workshop on Sunday where we talk about what's needed to make evaluation meaningful!
Nataliya Tkachenko, PhD in #LLMs retweeted
And so it starts:
JD Vance says US could drop support for NATO if Europe tries to regulate Elon Musk’s platforms independent.co.uk/news/world…



































