Saibo-Creator

PhD @EPFL | Reliable Efficient LLM Inference | ex-intern @MSFT

[Saibo-Creator @SaiboGeng]

Arrived at #NeurIPS ! Happy to chat :)

[Saibo-Creator @SaiboGeng]

Author here. Zip2zip is a cool work and I’m proud of it :)

[Alexia Jolicoeur-Martineau @jm_alexia]

Cool new work on progressively growing a dynamic vocabulary that merges tokens using LZW compression.

https://t.co/106V4lmvU9

[Bob West @cervisiarius]

📄✨Excited to share our new paper accepted to #EMNLP ’25:

Combining Constrained and Unconstrained Decoding via Boosting: BoostCD and Its Application to Information Extraction
https://t.co/ljsWULBHEA

(led by #EPFL PhD student Marija Šakota -- soon on the job market, hire her!!)

[Julian Minder @jkminder]

New paper: Finetuning on narrow domains leaves traces behind. By looking at the difference in activations before and after finetuning, we can interpret what it was finetuned for. And so can our interpretability agent! 🧵

[Saibo-Creator @SaiboGeng]

Definitely something that we urgently need!

[Bob West @cervisiarius]

🚨New paper alert! 🚨

Tandem Training for Language Models
https://t.co/Emzcgf1KHx

Actions & thoughts of AI w/ superhuman skills will be hard for humans to follow, undermining human oversight of AI. We propose a new way to make AI produce human-understandable solutions. How?👉🧵

[Saibo-Creator @SaiboGeng]

Our oral presentation of zip2zip at ICML-ES-FoMo is now available online! https://t.co/m334GpVM6m

[Saibo-Creator @SaiboGeng]

🚀 Excited to share our latest work at ICML 2025 — zip2zip: Inference-Time Adaptive Vocabularies for Language Models via Token Compression!

Sessions:

📅 Fri 18 Jul 
- Tokenization Workshop

📅 Sat 19 Jul 
- Workshop on Efficient Systems for Foundation Models (Oral 5/145)

[Saibo-Creator @SaiboGeng]

Great honor to present our work zip2zip at @ESFoMo ! Had discussion with many people and got some great feedback! Big thanks to the organizers ! @nathanrchn @realDanFu @SonglinYang4

[ES-FoMo@ICML2025 @ESFoMo]

@wanchao_ @bariskasikci @EyubogluSabri Our next contributed talk, zip2zip!

[Saibo-Creator @SaiboGeng]

📄 Paper: https://t.co/KxAxCHKvJh
💻 Code: https://t.co/mQ6EoWCAAh
🤖 Model: https://t.co/JcNEekimc5

[Saibo-Creator @SaiboGeng]

🚀 Excited to share our latest work at ICML 2025 — zip2zip: Inference-Time Adaptive Vocabularies for Language Models via Token Compression!

Sessions:

📅 Fri 18 Jul 
- Tokenization Workshop

📅 Sat 19 Jul 
- Workshop on Efficient Systems for Foundation Models (Oral 5/145)

[Saibo-Creator @SaiboGeng]

💡 The result: up to 60% fewer tokens, faster inference, and lower cost — all while preserving strong performance across domains like code, biomedical text, and multilingual inputs.

[Saibo-Creator @SaiboGeng]

It feels quite unreal that after I return from my holiday, the company I'm interning got acquired! CRAZY TIMES !!!😎
Congrats again to the cofounders!!!

[Invariant Labs @InvariantLabsAI]

We’re thrilled to officially join forces with @snyksec!

Together, we’re changing the landscape of the agentic AI future. More to come!

[Saibo-Creator @SaiboGeng]

Tried GPT4o, Claude 4 and DeepSeek, all give 27

[Saibo-Creator @SaiboGeng]

OpenAI now has an improved Structured Output mode, with support from Guidance(llguidance) :)

[OpenAI Developers @OpenAIDevs]

We've made some improvements to Structured Outputs:

🎣 Parallel function calling now works with strict mode—ensuring calls reliably adhere to schema

⚙️ Many more keywords are now supported, letting you specify:
- Output string lengths and formats via regex or formats like email
- Min/max ranges for numbers
- Min/max elements in arrays
- and more!

[Luca Beurer-Kellner @lbeurerkellner]

🔴 New MCP attack leaks WhatsApp messages via MCP, side-stepping WhatsApp security. 1/n

We show a new MCP attack that leaks your WhatsApp messages if you are connected via WhatsApp MCP.

Our attack uses a sleeper design, circumventing the need for user approval.

More 👇

[Saibo-Creator @SaiboGeng]

Great opportunity!

[Akhil Arora @aroraakhilcs]

I am recruiting 2 PhD students for Fall'25 @csaudk to work on bleeding-edge topics in #NLProc #LLMs #AIAgents (e.g. LLM reasoning, knowledge-seeking agents, and more).

Details: https://t.co/JiYUnM2n38
Deadline: May 1, 2025

Please boost!
cc: @WikiResearch @AiCentreDK @CPH_SODAS

[Chris Wendler @wendlerch]

R1 is exciting and provides unique challenges from a mechanistic interpretability point of view. Come join us at ARBOR — an open research collective aimed at collaboratively reverse engineering LLM based reasoning models.

[Julian Minder @jkminder]

Can we understand and control how language models balance context and prior knowledge? Our latest paper shows it’s all about a 1D knob! 🎛️
https://t.co/698wbJ0IDZ
Co-led with @kevdududu, as well as @niklas_stoehr, @giomonea, @wendlerch, @cervisiarius & Ryan Cotterell.

[Akhil Arora @aroraakhilcs]

Disappointed to not be at #EMNLP owing to a dislocated shoulder 😢
@DebjitPaul2 will present our poster on Multilingual Entity Insertion (cf. https://t.co/PrYp4owMQ9). Swing by our poster in session #6 on Wed 13@10:30 EST

🚀 PS: I am hiring PhD students @csaudk
#LLM #GNNs #CSS