Pinned Tweet
Hallelujah!
I’m excited to share that I’ve been selected as a 2025 AI2050 Early Career Fellow by @schmidtsciences
This year’s fellows represent 42 institutions across eight countries, working to ensure AI benefits humankind.
Learn more at: lnkd.in/eZA5FHci
I’m excited to share that I’ve been selected as a 2025 AI2050 Early Career Fellow by @schmidtsciences
This year’s fellows represent 42 institutions across eight countries, working to ensure AI benefits humankind.
Learn more at: lnkd.in/eZA5FHci
We're excited to welcome 28 new AI2050 Fellows! This 4th cohort of researchers are pursuing projects that include building AI scientists, designing trustworthy models, and improving biological and medical research, among other areas. buff.ly/riGLyyj
43
15
303
18,832
David Ifeoluwa Adelani 🇳🇬 retweeted
TACL now only allows one new submission from first authors every three months (excluding invited resubmissions), with some exceptions. See
transacl.org/index.php/tacl…
transacl.org/index.php/tacl…
0
1
29
1,588
David Ifeoluwa Adelani 🇳🇬 retweeted
Starting in an hour at #AACL2026 @aaclmeeting: our tutorial on Measurement Science for Language Model Evaluation, with @sangttruong, @_rodolfocorona_, and Serena Wang.
We take six claims people often make about benchmarks, spanning three kinds of error (sampling, specification, strategic), and ask the room to vote:
1. Averaging safety benchmarks robustly measures safety
2. To compare models fairly, every model must answer the same full test
3. A well-translated benchmark measures the same thing in every language
4. The leaderboard winner will fit my product best
5. Evaluation gaming is mostly cheating or data leakage
6. A private, uncontaminated test set stops gaming
Most of these are partly right. We’ll show where each one holds, where it breaks, and a better claim to make instead.
14:00, Muhe Hall B, Hengqin
Reading list: aims.stanford.edu/tutorial
(slides uploaded after tutorial)
We take six claims people often make about benchmarks, spanning three kinds of error (sampling, specification, strategic), and ask the room to vote:
1. Averaging safety benchmarks robustly measures safety
2. To compare models fairly, every model must answer the same full test
3. A well-translated benchmark measures the same thing in every language
4. The leaderboard winner will fit my product best
5. Evaluation gaming is mostly cheating or data leakage
6. A private, uncontaminated test set stops gaming
Most of these are partly right. We’ll show where each one holds, where it breaks, and a better claim to make instead.
14:00, Muhe Hall B, Hengqin
Reading list: aims.stanford.edu/tutorial
(slides uploaded after tutorial)
4
12
132
6,337
David Ifeoluwa Adelani 🇳🇬 retweeted
Thrilled to give a tutorial at #AACL2026 @aaclmeeting with @juhyunohh @gentaiscool @ruochenz_ @thamar_solorio @aliceoh
👉 𝘾𝙤𝙙𝙚-𝙎𝙬𝙞𝙩𝙘𝙝𝙞𝙣𝙜 𝙛𝙤𝙧 𝙈𝙪𝙡𝙩𝙞𝙡𝙞𝙣𝙜𝙪𝙖𝙡 𝙇𝙇𝙈𝙨
🔗 Website: sites.google.com/view/code-swit…
📄 Slides: docs.google.com/presentation/d…
👉 𝘾𝙤𝙙𝙚-𝙎𝙬𝙞𝙩𝙘𝙝𝙞𝙣𝙜 𝙛𝙤𝙧 𝙈𝙪𝙡𝙩𝙞𝙡𝙞𝙣𝙜𝙪𝙖𝙡 𝙇𝙇𝙈𝙨
🔗 Website: sites.google.com/view/code-swit…
📄 Slides: docs.google.com/presentation/d…
1
9
43
2,395
David Ifeoluwa Adelani 🇳🇬 retweeted
Super excited for Chutong Meng's new paper (in collaboration with Barron Associates): arxiv.org/pdf/2610.40233
We substantially improve translation for extremely low-resource languages by extracting parallel data from linguistic documentation materials.
Thread 👇
We substantially improve translation for extremely low-resource languages by extracting parallel data from linguistic documentation materials.
Thread 👇
1
9
27
2,418
David Ifeoluwa Adelani 🇳🇬 retweeted
(1/7) 🚀 Introducing DR Pixel, the first open-source deep research agent that searches and reasons directly over page images, no OCR. On 1.06M pages, our trained 27B model matches GPT-5.4.
📄 arxiv.org/abs/2611.03097
Joint work from @BerkeleyNLP, @rtnldynamics!
📄 arxiv.org/abs/2611.03097
Joint work from @BerkeleyNLP, @rtnldynamics!
13
39
191
28,054
David Ifeoluwa Adelani 🇳🇬 retweeted
We @huggingface reproduced this paper!!
It's built with ML Intern, an open-source harness to do AI research, train models, and more
Model: huggingface.co/ML-Intern-lab/…
Demo: huggingface.co/spaces/ML-Inte…
It's built with ML Intern, an open-source harness to do AI research, train models, and more
Model: huggingface.co/ML-Intern-lab/…
Demo: huggingface.co/spaces/ML-Inte…
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: fronely read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weight-sp…Community noteLLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
64
181
1,638
95,741
David Ifeoluwa Adelani 🇳🇬 retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: fronely read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weight-sp…
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: fronely read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weight-sp…
Community note
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
2,365
7,276
55,949
9,931,408
Congratulations Prosper @francoisrmeyer and @janmbuys 🎉
Announcing the Best Paper for this year's MRL: RunyaNER - Auxiliary Language Selection for Runyankore NER!
Congratulations to the authors: Prosper Asiimwe, Francois Meyer, Jan Buys!
#MRL26 #EMNLP2026
0
1
18
718
David Ifeoluwa Adelani 🇳🇬 retweeted
And that's a wrap! Thank you so much for attending, contributing, and participating in this year's MRL workshop! 👏
See you next year at EMNLP 2027! 🎉 🌴
Kudos to our in-person organizers: @ljvmiranda @omerNLP @linguist_cat Senyu Li @Jirui_Qi @davidstap !
See you next year at EMNLP 2027! 🎉 🌴
Kudos to our in-person organizers: @ljvmiranda @omerNLP @linguist_cat Senyu Li @Jirui_Qi @davidstap !
0
3
11
234
David Ifeoluwa Adelani 🇳🇬 retweeted
Slated to give a tutorial on “Multilingual and Multicultural LLMs” 🌍 at EMNLP 2026 with @davlanade @vinodkpg @sarahookr @alan_ritter.
See you in beautiful Budapest — looking forward to connecting with everyone! 👋
#emnlp
1
2
31
2,140
David Ifeoluwa Adelani 🇳🇬 retweeted
I’m excited to be presenting our research, TukaBench, at EMNLP 2026 here in Budapest, Hungary! 🇭🇺
I’ll be giving an oral presentation today, and I’d love to connect with fellow researchers and anyone interested in AI safety, multilingual NLP, and culturally inclusive AI.
Our ant question: How well do large language models handle safety when prompts are grounded in African languages and cultural contexts?
Presentation details:
Date: Monday, October 26, 2026
Time: 4:45 PM to 6:15 PM
Venue: Hall G-15-16, HUNGEXPO, Budapest
Session: 9 (Orals/Poster E)
If you’re attending EMNLP, please come around. I’d be happy to share our findings, answer your questions, and hear your thoughts.
Looking forward to some interesting conversations!
Paper: aclanthology.org/2026.emnlp-mai…
#EMNLP2026 #AISafety #NLP #AfricanLanguages #MultilingualAI #AIResearch
I’ll be giving an oral presentation today, and I’d love to connect with fellow researchers and anyone interested in AI safety, multilingual NLP, and culturally inclusive AI.
Our ant question: How well do large language models handle safety when prompts are grounded in African languages and cultural contexts?
Presentation details:
Date: Monday, October 26, 2026
Time: 4:45 PM to 6:15 PM
Venue: Hall G-15-16, HUNGEXPO, Budapest
Session: 9 (Orals/Poster E)
If you’re attending EMNLP, please come around. I’d be happy to share our findings, answer your questions, and hear your thoughts.
Looking forward to some interesting conversations!
Paper: aclanthology.org/2026.emnlp-mai…
#EMNLP2026 #AISafety #NLP #AfricanLanguages #MultilingualAI #AIResearch
6
8
43
939
David Ifeoluwa Adelani 🇳🇬 retweeted
New publication - AfriScience-MT: Towards Decolonizing Science in Africa through Text Translation <> A parallel corpus covering six African languages (Amharic, Hausa, Luganda, Northern Sotho, Yorùbá, and isiZulu) across 11 scientific domains. aclanthology.org/2026.emnlp-mai… #EMNLP2026
0
12
32
1,434
David Ifeoluwa Adelani 🇳🇬 retweeted
🚀 I am hiring a Postdoc in AI Safety & Alignment @imperialcollege to work on ⚡️Detecting Unverbalized Dangerous Capabilities in Frontier Models⚡, as part of a grant I received.
🗓️ Application Deadline: 22nd November
🔗 Link: imperial.ac.uk/jobs/search-jo…
📅 Start date: ASAP
🗓️ Application Deadline: 22nd November
🔗 Link: imperial.ac.uk/jobs/search-jo…
📅 Start date: ASAP
3
16
128
9,946
David Ifeoluwa Adelani 🇳🇬 retweeted
I turned @googlegemma EmbeddingGemma 2, an embedding model, into a speech-to-text model, and it turned out to be one of the fastest STT models you can run on device.
On a base M3 MacBook Air it transcribes an hour of audio in about 13 seconds: 288× real time.
The idea came in ding 2. It puts text, images, audio and video into one shared space, so a spoken word lands right next to the written word. If that's true, you should be able to read a transcript straight out of the embedding model. I couldnt tried it then, but last week EmbeddingGemma 2 came out as an open model, so I tried it.
On a base M3 MacBook Air it transcribes an hour of audio in about 13 seconds: 288× real time.
The idea came in ding 2. It puts text, images, audio and video into one shared space, so a spoken word lands right next to the written word. If that's true, you should be able to read a transcript straight out of the embedding model. I couldnt tried it then, but last week EmbeddingGemma 2 came out as an open model, so I tried it.
125
274
3,520
298,034
David Ifeoluwa Adelani 🇳🇬 retweeted
As we approach @emnlpmeeting next week, we're excited to share with you the MRL Workshop Programme on October 28! See you in-person / virtually at MRL 🎉🌏
Stay tuned in the coming days for more details about our Keynote Talks ;)
#EMNLP2026 #nlp #nlproc
Stay tuned in the coming days for more details about our Keynote Talks ;)
#EMNLP2026 #nlp #nlproc
0
3
12
689
David Ifeoluwa Adelani 🇳🇬 retweeted
Super excited to share SnapJudge -- a free, open detector trained for scientific and math-heavy writing, and can handle multiple modalities, including screenshots of paper text, LaTex, and text extracted from PDFs... 🧵
Introducing SnapJudge: a free, open AI detector for scientific writing, including math!
Drop in a PDF screenshot, paste text or LaTeX, or just put in an arXiv link.
We ran it on arXiv: ~1 in 3 new math papers are flagged as mostly AI, up from ~0 in early 2025. CS is 45%.
🧵
1
14
67
4,174
David Ifeoluwa Adelani 🇳🇬 retweeted
⌘C ⌘V: Introducing Copy-Paste Language Models!
CPLMs learn to copy tokens better, improving downstream accuracy, CoT, long-context performance, and memory-efficiency for RAG at <3% overhead
PS: they (partially) mitigate the LM head bottleneck
Paper: arxiv.org/abs/2610.19247
CPLMs learn to copy tokens better, improving downstream accuracy, CoT, long-context performance, and memory-efficiency for RAG at <3% overhead
PS: they (partially) mitigate the LM head bottleneck
Paper: arxiv.org/abs/2610.19247
25
65
555
61,818
David Ifeoluwa Adelani 🇳🇬 retweeted
how do you fine-tune a decision model on your own data?
in these two notebooks, you'll learn:
• how to prepare the training data
• how to fine-tune our d1-3B decision model
• using @UnslothAI and @huggingface TRL
all within 30 minutes on a free-tier GPU.
unsloth: github.com/Liquid4All/coo…
TRL: github.com/Liquid4All/coo…
ps: how adorable are these halloween kitties? 👻
in these two notebooks, you'll learn:
• how to prepare the training data
• how to fine-tune our d1-3B decision model
• using @UnslothAI and @huggingface TRL
all within 30 minutes on a free-tier GPU.
unsloth: github.com/Liquid4All/coo…
TRL: github.com/Liquid4All/coo…
ps: how adorable are these halloween kitties? 👻
42
242
1,986
85,870
David Ifeoluwa Adelani 🇳🇬 retweeted
Mila’s annual supervision request process is open for MSc and PhD applications for Fall 2027 admission. While I won’t be taking on new students for supervision this cycle, I encourage you to apply to join Mila’s brilliant deep learning research community.
For more information and guidelines, you can visit the link below or contact the @Mila_Quebec team directly.
mila.quebec/en/prospective…
For more information and guidelines, you can visit the link below or contact the @Mila_Quebec team directly.
mila.quebec/en/prospective…
4
22
169
8,408
Recruiting 1-2 PhD students for Fall 2027, please apply if you are interested in Multilingual LLMs
Mila's annual supervision request process is now open to receive MSc and PhD applications for Fall 2027 admission! For more information, visit mila.quebec/en/prospective…
3
28
115
4,902