Lin Gui

NLP lecturer (assistant professor)@King's College London, also do a bit IR, ML, and Stats for fun.

[Lin Gui @LinGui_KCL]

Thanks to #ICLR2026 for giving me the chance to expand my horizons. Just got assigned some papers way outside my domain…

[Lin Gui @LinGui_KCL]

Thanks for the suggestion. BTW, do remember to estimate the reviewer's time zone. Don’t time your reply to land exactly 12 hours after their comment. No one likes openreview notifications at midnight.

[Ravid Shwartz Ziv @ziv_ravid]

ICLR rebuttal strategy meeting:
🕐 30 min discussing the OPTIMAL time to ping reviewers
📊 Must be after Thanksgiving but before the weekend
🧠 Late enough they can't ask for more experiments
⏰ Early enough they still have time to respond

Lead author right after the call: "I just sent it"
Everyone: "WHAT HAVE YOU DONE. THIS IS NOT WHAT WE AGREED ON. YOU RUINED EVERYTHING. WE'RE COOKED 💀"

5 minutes later
Two reviewers raise their scores by 4 points each 😱😱😱

[Lin Gui @LinGui_KCL]

@iclr_conf In my batch, it seems like the judgement of AI geneated reviews is purely based on the decisions from pangram. Do you think it is reliable evidence to desk reject so many submissions? At least in my batch, 2 out of 7 reviews flagged by pangram are not from AI

[Lin Gui @LinGui_KCL]

@alec_helbling May I include this in my teaching slides? Very comprehensive!

[Rosanne Liu @savvyRL]

Remember the world is bigger than your ICLR reviews

[Lin Gui @LinGui_KCL]

@xuandongzhao Who is Summary?

[Lin Gui @LinGui_KCL]

@gan_chuang @miiche_yang Congratulations! May I invite him to share this work with our group? We have many students who are working on the topic of latent reasoning.

[Lin Gui @LinGui_KCL]

New work from our KCL NLP group @kclnlp in collaboration with the Centre for AI @ASTRAZENECAUK !

We introduce Latent Refinement Decoding (LRD), a two-stage parallel generation framework that achieves up to 10.6× faster decoding while improving accuracy across coding and reasoning tasks.

LRD addresses key limitations of diffusion-inspired approaches by maintaining distributional mixtures and iterative feedback for more globally consistent generation.

🧠 A strong and versatile alternative to standard autoregressive decoding!

Paper: https://t.co/s2bx2pLDG1
#AI #NLP #LLMs #GenerativeAI #Research

[KCL NLP @kclnlp]

🚀 The KCL NLP Group is heading to #ACL2025
We’re excited to present our latest work on LLM reasoning, interpretability, personalization, and many other topics.

📍 Catch us in Vienna — stop by, say hi, and let’s talk!

[Lin Gui @LinGui_KCL]

@CSProfKGD @mabdelhack Our coauthor, who is Chinese but completed her education in the UK since high school and has no prior research experience in China, is still awaiting the outcome of her visa application.

[Lin Gui @LinGui_KCL]

Steer LLM reasoning using a single continuous token, without relying on SFT or RL. Check out our #ICML2025 poster E-2024 from the KCL NLP group: Soft Reasoning: Navigating Solution Spaces in Large Language Models through Controlled Embedding Exploration.

Paper: https://t.co/OGNHSPdv9t
GitHub: https://t.co/RCSIItFqqr