Liam Liang Ding

building agentic ai @AlibabaGroup & ex-@AIstartup @JD_Corporate @TencentGlobal @Sydney_Uni. opinions are my own.

[Liam Liang Ding @liangdingNLP]

@wzenus @YejinChoinka @jiajunwu_cs @ManlingLi_ @LINJIEFUN @chi_gui_1 @DeimosGN @qineng_wang @James_KKW @shiqi_chen17 @zhengyuan_yang Congrats 🎊

[Liam Liang Ding @liangdingNLP]

6/6) Call to Action: Current LLMs still leave substantial headroom on IndustryBench (top score is 2.083/3). Industrial LLM evaluation must move beyond aggregate accuracy and prioritize source-grounded, safety-aware diagnosis.

[Liam Liang Ding @liangdingNLP]

1/6) Excited to share our latest work from the Multimodal and Industrial AI team at Alibaba: IndustryBench! 🚀⚙️

In industrial procurement, an LLM's answer is only useful if it survives strict standards checks. Partial correctness can mask safety-critical contradictions.

Check out the full paper for deep dives into capability dimensions and model comparisons! Feedback and PRs are highly welcome. 👇

Data: https://t.co/8ZflFcHw5W
Code: https://t.co/iTRMcJQhDr
Paper: https://t.co/reTXgWdDrf

#Alibaba #Gemini #Qwen #GPT #Claude #Kimi #GLM #Mimimax

[Liam Liang Ding @liangdingNLP]

5/6) The Multilingual Blindspot: We released 2,049 items with aligned renderings in EN, RU, VI, and ZH.

Across 17 models, "Standards & Terminology" is the most persistent capability weakness. This weakness survives across all four language translations—proving this is a structural knowledge gap, not just a translation artifact.

[Liam Liang Ding @liangdingNLP]

@xieenze_jr Big congrats 🎉

[Liam Liang Ding @liangdingNLP]

@bingyikang @amilabs @sainingxie @ylecun Congrats

[Liam Liang Ding @liangdingNLP]

qwen is nothing without its people, best wishes for whatever happens next ❤️

[Junyang Lin @JustinLin610]

me stepping down. bye my beloved qwen.

[Liam Liang Ding @liangdingNLP]

@prajdabre lol exactly 😹

[Liam Liang Ding @liangdingNLP]

@xuhaiya2483846 Great works! GUI is still an important I/O interface for human-machine👏

[Liam Liang Ding @liangdingNLP]

@jungokasai nice product and can’t wait to try 🎉

[Liam Liang Ding @liangdingNLP]

Legally it acts as a liability shield, but strictly speaking it's a UX innovation. It bridges the gap and brings the tech closer to the average person.

[Jay ⛽️ @jelanifuel]

Openclaw hype is kinda wild to me.

It’s literally just a wrapper to Claude code with more risk.

The fact that it can run autonomously and call apis is the same thing Claude code can do…

Not sure what I’m missing

[Liam Liang Ding @liangdingNLP]

Setting temperature = 0 ❄️🇮🇸

[Liam Liang Ding @liangdingNLP]

@MikaStars39 @huggingface Big congrats

[Liam Liang Ding @liangdingNLP]

Welcome to check out our recent reality checking work on whether Diffusion LLMs are suitable for agentic tasks.

[Qingyu-Lu @SiriusLu1]

❓ Do diffusion-LLMs truly generalize in agentic tasks?

We reveal systematic failure modes in causal reasoning & tool use, and introduce DiffuAgent for comprehensive evaluation ⚡

📄 https://t.co/lDI7YhXP8l
🔗 https://t.co/0KZhAYAD5L

#AgenticAI #LLM #EmbodiedAI #ToolUse

[Liam Liang Ding @liangdingNLP]

@wangchunshu @dreamina_ai Congrats 🥂

[Liam Liang Ding @liangdingNLP]

Milan moments🇮🇹