Hanna Yukhymenko (@a_yukh) header
Hanna Yukhymenko (@a_yukh) profile picture

Hanna Yukhymenko

@a_yukh

post-training/evals @apertusllm @eth_ai_center | ex-🤗 @eth | EEU ambassador @insaitinstitute

 🇨🇭Zurich|🇺🇦
linktr.ee/ayukh
Joined September 2017
  • Tweets496
  • Following595
  • Followers770
128Media
Photo by Hanna Yukhymenko (@a_yukh)Photo by Hanna Yukhymenko (@a_yukh)Photo by Hanna Yukhymenko (@a_yukh)Photo by Hanna Yukhymenko (@a_yukh)Photo by Hanna Yukhymenko (@a_yukh)Photo by Hanna Yukhymenko (@a_yukh)Photo by Hanna Yukhymenko (@a_yukh)Photo by Hanna Yukhymenko (@a_yukh)Photo by Hanna Yukhymenko (@a_yukh)Photo by Hanna Yukhymenko (@a_yukh)
Apertus 1.5 Technical Report is out! 134 pages of fully open multimodal reasoning model development from Swiss AI Initiative (ETH Zurich/EPFL) with data recipe, training experiments and many interesting insights 🇨🇭 Thread with my personal coverage ⬇️
Image from Hanna Yukhymenko (@a_yukh)
2
13
3
56
2,864
and you get a benchmark and you get a benchmark given the recent news looks like benchmarks need some refreshing, more coming this week👀
Hanna Yukhymenko (@a_yukh) profile pictureHanna Yukhymenko
@a_yukhJul 28
so many nice benchmarks cooking rn, maybe I am an eval engineer after all
Image from Hanna Yukhymenko (@a_yukh)
2
22
490
Hanna Yukhymenko retweeted
Angelika Romanou (@agromanou) profile picture
Angelika Romanou@agromanouOct 25
We're excited to release INCLUDE-2.0! 🌍 An expanded multilingual benchmark evaluating models' regional factual knowledge using questions from local exams across 91 languages, 22 scripts and 86 countries.
4
8
2
32
3,504
Btw who is at #EMNLP2026 right now, you can check out CulturalRiddles poster at MRL workshop on October 28th! 🇭🇺🎉
Hanna Yukhymenko (@a_yukh) profile pictureHanna Yukhymenko
@a_yukhOct 22
Language models claim to know many languages, but do they understand many cultures? 🤔 Very honored to contribute to newly released CulturalRiddles benchmark from @Cohere_Labs 🌐 #culturalriddles
1
1
13
377
looking at some agentic and IF benchmarks my point still stands 🫠
Hanna Yukhymenko (@a_yukh) profile pictureHanna Yukhymenko
@a_yukh27 Jun 2025
hot take: doing a multilingual paper just for top 5-10 most common languages is not really multilingual
Image from Hanna Yukhymenko (@a_yukh)
9
322
Hanna Yukhymenko retweeted
Cohere Labs (@Cohere_Labs) profile picture
Cohere Labs
@Cohere_LabsOct 23
Open science is only possible because of the people who make it a collective effort. 💙 A huge thank you to the 160+ riddle writers who shared their time, knowledge, and cultures to help build CulturalRiddles. By contributing riddles in their own languages, they helped create a benchmark that challenges AI models to go beyond words and facts to better understand cultural context. 🌟 Thank you to all our contributors for making this work possible! #CulturalRiddles
Image from Cohere Labs (@Cohere_Labs)
2
8
1
54
2,257
Language models claim to know many languages, but do they understand many cultures? 🤔 Very honored to contribute to newly released CulturalRiddles benchmark from @Cohere_Labs 🌐 #culturalriddles
3
2
1
27
1,080
more replies
This is a very important and timely release, which I hope will promote even more related work in this area! People may think multilinguality is solved, but benchmarks like these show that the work is far from done!
Image from Hanna Yukhymenko (@a_yukh)
1
3
52
Paper: arxiv.org/abs/2610.29968 Try beat the AI yourself: culturalriddles.ai Dataset: huggingface.co/datasets/Cohe… Dashboard: huggingface.co/spaces/Cohere…
CulturalRiddles: A Multicultural Benchmark of Riddles

CulturalRiddles: A Multicultural Benchmark of Riddles

As LLMs expand their language coverage, evaluating how well they understand and represent diverse cultures becomes increasingly important. However, many benchmarks remain English-centric, with...

arxiv.org
3
57
Hanna Yukhymenko retweeted
Cohere Labs (@Cohere_Labs) profile picture
Cohere Labs
@Cohere_LabsOct 22
Can you solve a riddle rooted in another culture? 🧩 Our latest research, CulturalRiddles, explores whether AI models can understand the cultural context behind human-written riddles from around the world. Test your knowledge with our quiz! culturalriddles.ai/
4
15
7
42
3,237
what if i told you we got one paper per week to release👀 tomorrow's one was fun to help with! fastcompany.com/91627566/ai-…
AI chatbots struggle to understand cultural context, a new study finds

AI chatbots struggle to understand cultural context, a new study finds

In an exclusive look at new research from Cohere Labs, even the best-performing AI models failed to solve riddles rooted in local cultures. Worse, they rarely admitted when they didn’t know the...

fastcompany.com
Hanna Yukhymenko (@a_yukh) profile pictureHanna Yukhymenko
@a_yukhOct 15
Apertus 1.5 Technical Report is out! 134 pages of fully open multimodal reasoning model development from Swiss AI Initiative (ETH Zurich/EPFL) with data recipe, training experiments and many interesting insights 🇨🇭 Thread with my personal coverage ⬇️
Image from Hanna Yukhymenko (@a_yukh)
1
19
733
arxiv.org/abs/2602.16800 🔊🔊🔊
Bryan Johnson (@bryan_johnson) profile pictureBryan Johnson
@bryan_johnsonOct 21
I have an AI agent that logs every negative comment someone makes about me online. It then creates a profile of the person, maps all of their social connections, and finds all their emails. Tomorrow morning, my AI agent will sent out an email blast to everyone's extended social network, detailing the horrible things they've said about me.
2
96
34,854
since the tweet is doing some views, sharing two of my favorite reads on this topic and the blog for paper above worth checking out simonlermen.substack.com/p/l… arxiv.org/abs/2310.07298 arxiv.org/abs/2604.21211
1
3
1,242
Hanna Yukhymenko retweeted
tweet davidson (@andyreed) profile picture
tweet davidson
@andyreedOct 19
another week at the prompt factory
Image from tweet davidson (@andyreed)
10
412
17
5,420
100,551
what in the beyond memorization paper is going on is this done on real data or am i missing something?
Image from Hanna Yukhymenko (@a_yukh)
AI at Meta (@AIatMeta) profile pictureAI at Meta
@AIatMetaOct 19
Today we published a technical report on Muse Glimmer, our 30B open-weight model built for agentic tasks on consumer hardware. It runs at practical speeds on a Mac or PC with a single consumer GPU at 4-bit precision, and leads comparably-sized open models on tool use and multi-step planning. Read about the model architecture, distillation recipe, quantization and more: go.meta.me/4Bcvxol
Image from AI at Meta (@AIatMeta)
11
434
many PRESENTING authors at neurips 🇫🇷 (not even main location) workshops are waitlisted, how small is that venue jesus
3
26
1,238
Update: larped my way to workshop ticket 🎉
Image from Hanna Yukhymenko (@a_yukh)
1
98
Hanna Yukhymenko retweeted
Avijit Ghosh (@evijit) profile picture
Avijit Ghosh
@evijitOct 15
Apertus 1.5 is the (checks notes) first ever model to come with a full eval card embedded in the tech report/blog! Check it out and would be super interesting to see repoducibility nerds deep dive into the data and uncover cool insights: huggingface.co/blog/swiss-ai…
Apertus 1.5 Release Notes

Apertus 1.5 Release Notes

A Blog post by Swiss AI Initiative on Hugging Face

huggingface.co
EvalEval Coalition (@evaluatingevals) profile pictureEvalEval Coalition@evaluatingevalsOct 15
Apertus 1.5 Tech report is out! It is the first model release with every eval on Evaluation Cards and HF Community Evals, reproducible down to the generation settings. 48 benchmarks for the 8B, 47 for the 70B. Rerun them yourself. Congrats Swiss AI ❤️ evalevalai.com/infrastructur…
Image from EvalEval Coalition (@evaluatingevals)
2
12
510
Load more