Li Nanbo @nbl_ai

Postdoc Fellow at @AI_KAUST (with @SchmidhuberAI) | Previously PhD at @EdinburghUni | Working on Machine Learning and World Models nanboli.github.io Saudi Arabia Joined October 2020

Tweets

16
Followers

126
Following

82
Likes

62

Li Nanbo @nbl_ai

a month ago

Never get tired of self improvement🚀

Jürgen Schmidhuber @SchmidhuberAI

a month ago

Meta learning and recursive self-improvement are old ideas. Foundation models breathe new life into them. Our new survey, “Self-Improvements in Modern Agentic Systems,” reviews how the concepts are continuing to evolve. Paper: arxiv.org/abs/2607.13104 Project:

24 159 764 54K 700

0 0 7 273 0

View Details

Li Nanbo @nbl_ai

4 months ago

Come and check this out 🤖！！！

Piotr Piękos @PiotrPiekosAI

4 months ago

Hey #ICLR2026, We're in Rio! Today we're presenting Huxley-Gödel Machine on the Oral 1C session (10:30 am-12, room 202 A/B. Apart from that, we also have a poster session later at 3:15 pm in Pavilion 3!

3 4 23 9K 1

0 0 4 139 0

View Details

Li Nanbo @nbl_ai

4 months ago

@braza_faouzi Hi - Yes. SlotSPE can uncover structural pathological patterns, not restricted to cancers, so we believe that it’d generalizes to other pathologies

0 0 1 46 0

View Details

Li Nanbo @nbl_ai

4 months ago

Excited to share our ICLR 2026 paper, SlotSPE! 🎉 Modular programming for healthcare to unlock efficiency, reveal interpretable structures, identify prognostic events, and advance trustworthy survival analysis. Honored to work on research with a higher purpose. See u in Rio🇧🇷

6 10 95 12K 52

View Details

Li Nanbo @nbl_ai

4 months ago

Honored to work with: Yilan Zhang, @ChangchunYangCS, @SchmidhuberAI, Xin Gao Arxiv: arxiv.org/abs/2512.01116 Github: github.com/zylvemvet/Slot…

0 0 3 323 2

View Details

Li Nanbo @nbl_ai

4 months ago

[3/3] More accurate, more interpretable, and more structured: not just risk prediction, but event-level morphology–molecular correspondence.

1 0 0 329 0

View Details

Li Nanbo @nbl_ai

10 months ago

We code only the agents that code themselves🤖🤖🤖! Wenyi Wang, @PiotrPiekosAI, @nbl_ai, Firas Laakom, @Beastlyprime, @MatOstasze, @MingchenZhuge, @SchmidhuberAI

Piotr Piękos @PiotrPiekosAI

10 months ago

🚨Time to let agents code themselves! Meet Huxley-Gödel Machine (HGM), a game changer in coding agent development🚨 [🤖vs.🧑‍💻]HGM evolves by self-rewrites to match the best officially checked human-engineered agents on SWE-Bench Lite despite being optimized on a different dataset.

2 11 62 8K 27

0 0 4 478 0

View Details

Piotr Piękos @PiotrPiekosAI

10 months ago

Authors: Wenyi Wang, @PiotrPiekosAI , @nbl_ai , Firas Laakom, @Beastlyprime , @MatOstasze, @MingchenZhuge , @SchmidhuberAI ArXiv: arxiv.org/abs/2510.21614 Github: github.com/metauto-ai/HGM

1 2 15 772 7

View Details

Piotr Piękos @PiotrPiekosAI

10 months ago

2 11 62 8K 27

View Details

Li Nanbo @nbl_ai

a year ago

Dear #ICLR2025 attendees, pls drop by our poster tomorrow if you’re interested in RNNs, world models, or some random chats. Looking forward to seeing you! Sat 26 Apr 15:00-18:00 At Hall 3 + Hall 2B #317.

0 0 3 132 0

View Details

Li Nanbo @nbl_ai

a year ago

@SchmidhuberAI [3/3] This design offers sophisticated state-space expressiveness, adaptability to data modularity, robustness to input permutation changes, and supports parallel linearisation. Check our paper below: arxiv.org/abs/2410.20922

1 4 23 2K 12

View Details

Li Nanbo @nbl_ai

a year ago

@SchmidhuberAI [2/3] FACTS advances further: It introduces attentive routing to augment input-memory selectivity, enabling permutable memory structures such as graphs and sets.

0 1 4 1K 0

View Details

Li Nanbo @nbl_ai

a year ago

@SchmidhuberAI [1/3] LSTMs utilise input-memory interactions (the famous gating mechanism) for selective history compression. Mambas simplify such compression with input-only selectivity and linearised recurrence for enhanced parallelisation.