Data/Compliance lawyer in Mainland China. CEO of Pathway Compliance, an ALSP based in Shanghai. Views my own and not legal advice.
Shanghai
Joined April 2011
- Tweets4K
- Following3.2K
- Followers28.3K
- Likes18.7K
Jared T Nelson retweeted
Why do legal AI systems fail when cases unfold over months, not moments?
Researchers from Fudan University, Shanghai Innovation Institute, and Northwest University of Political and Law present LegalWorld – a life-cycle legal environment that models Chinese civil litigation as a causally connected chain of five stages, not isolated tasks.
Built on 75,309 real judgments with persistent memory and skills, it outperforms prior simulators by preserving cross-stage dependencies.
Tests across consultation, drafting, and courtroom advocacy reveal sharp capability gaps – no single model leads across all stages, proving that today's aggregated scores hide critical weaknesses.
LegalWorld: ALife-Cycle Interactive Environment for Legal Agents
Paper: arxiv.org/abs/2606.18728
Project: github.com/sii-research/Lega…
Our report: mp.weixin.qq.com/s/E3e8Sp0BT…
📬 #PapersAccepted by Jiqizhixin
Jared T Nelson retweeted
I’m an AI PhD at Oxford wanted to move to the US for years, but it never worked out. Maybe I can share why.
To me, it feels like the US is making it harder and harder for Chinese talent to come or stay.
First, Trump signed PP10043, which meant I couldn’t go to the US for any graduate school. That’s why I had to turn down my Stanford offer and come to Oxford.
Later, I qualified for an EB-1A (the extraordinary ability green card). But because I was born in China, I’m stuck in a country-specific backlog that could take another 4–5 years.
Meanwhile, on the other side, I’m getting 3–4 emails from China almost every week. They offer high salaries, free housing, generous research funding, and even offered an astonishing amount of money just for me to come back and have a conversation.
I’ve always loved the American spirit. I truly believe my abilities could have a bigger impact there. But honestly, the whole process has been incredibly frustrating.
All of this makes me wonder whether I should just give up.
Why am I spending so much energy trying to go to a country that doesn’t seem to want me, when another country is doing everything it can to bring me back?
Jared T Nelson retweeted
A lawyer friend on Legora:
"Legora really helped with this latest motion, which was dense with past issues in the case. Now my paralegal says she cannot live without it. It is just too remarkable and timesaving."
Jared T Nelson retweeted
Today I published The China Tech Field Guide, and it is free.
Over the past year, Hello China Tech has covered China's AI, semiconductor, robotics, and EV sectors essay by essay. The Field Guide consolidates that year into six principles. A sample of the argument: in 2025, 764 Chinese AI companies raised a combined Rmb 83bn, and the top 20 captured 30 per cent of it. Capital structure, not technology, decides which firms survive. The state does not referee these markets. It designs them.
If your work touches China tech, the Guide is designed to be the document you send to a colleague who asks where to start. Web and PDF, both free. hellochinatech.com/p/china-t…
Jared T Nelson retweeted
The good news is that this has really brought my husband and I together by uniting his interests (soccer) and mine (rules, what are the rules, who decides when the rules are broken, what is the punishment for breaking the rules, what is the process for evaluating the fairness of
Jared T Nelson retweeted
China’s Meituan just released LongCat 2.0, 1.6T params, trained entirely on 50k domestic chips (zero NVIDIA).
This isn’t just another Chinese model catching up. It’s a 1.6T MoE beast that’s already matching or pressuring GPT-5.5, Gemini 3.1 Pro, and Claude Opus on benchmarks, all while being fully open-sourced and running on non-US hardware.
Western labs relying on NVIDIA moats and closed models just got a loud wake-up call. If China can scale frontier training on local silicon at this level, the compute arms race is wider open than ever.
Innovation doesn’t wait for export licenses!
Introducing LongCat-2.0 🐱
1.6T parameters · MoE with ~48B active · 1M context
The full model behind Owl Alpha on @OpenRouter — now available.
Built for agentic coding from the ground up:
◆ LongCat Sparse Attention (LSA) — scales efficiently for 1M-context tokens
◆ Zero-Compute Experts — dynamic activation 33B–56B per token, zero wasted compute
◆ MOPD — three specialized expert groups (Agent / Reasoning / Interaction), gate-routed per task
How it stacks up:
→ Terminal-Bench 2.1: 70.8
→ SWE-bench Pro: 59.5 (GPT-5.5: 58.6)
→ SWE-bench Multilingual: 77.3
→ FORTE: 73.2 · RWSearch: 78.8 · BrowseComp: 79.9
📖 Tech Blog: longcat.chat/blog/longcat-2.…
Try it across different scenarios 🧵👇
Jared T Nelson retweeted
And for the “just check the cites” crowd:
“In addition, the brief completely misrepresented what was decided or discussed in two real cases cited therein.”
Hallucinated cases are just the tip of the iceberg.
More lawyer AI misuse, from a NY state appellate decision today:
"Sanders later admitted in his affirmation, he was not being candid at the time of oral argument, as he should have said that the cases were hallucinated by artificial intelligence.
Sanders 'expla[ined]' that his decision to be dishonest with the Court was due to being 'scared.' This may be true.
However, Sanders's decision not to be honest with the Court during oral argument also demonstrated his professional immaturity, arrogance, and profound lack of respect for the judicial system."
Yikes.
Jared T Nelson retweeted
China has overtaken Europe in drug innovation, a top Pfizer exec just warned. The numbers are wild: 40% of all oncology clinical trials worldwide are now in China;
Drug development runs 3x faster and ~half the cost vs. Europe; In 2024, 28 of 81 new innovative medicines came from China — vs. just 18 from Europe. The US is responding too — the FDA just launched "Operation TrialBlazer" to cut development timelines. China outpacing Europe in drug innovation and development, Pfizer executive says | Reuters
Jared T Nelson retweeted
There's a database every lawyer should bookmark, and then be scared by.
It's Damien Charlotin's AI hallucination database. A live, public record of lawyers and firms whose AI-generated fake citations got caught in court. Real cases, real names, real outcomes.
As of this week it holds 1,636 cases. The growth is the part that should stop you cold:
16 cases in 2023.
59 in 2024.
822 in 2025.
739 already in the first half of 2026.
The most recent entry is from three days ago. The fines run into six figures. One US sanction topped $110,000. And 120 of these carry a professional sanction, the disciplinary referral that decides whether you keep your license. It spans every practice area, contract to criminal. Nobody is too senior or too specialized to land on it.
Spend ten minutes in there and one thing becomes obvious. These were not stupid lawyers. Most were competent, busy people who made one specific mistake. They trusted what the AI gave them.
That's the whole trap, in one sentence. The model produces a citation that looks perfect. Formatted correctly, confident, plausible. And sometimes it simply does not exist. The danger was never that AI is obviously wrong. It's that AI is convincingly wrong, exactly when you're tired and the deadline is tonight.
The advice you usually hear is "train your team" and "write a policy." Both fine. Neither is the thing that actually keeps you off the list.
The thing that keeps you off it is one rule, applied without exception. You never file a citation you have not personally opened and read. Not "the AI seemed reliable." Not "it's usually right." Opened. Read. Confirmed it exists and says what you're claiming. Every time.
Here's the uncomfortable part though. Almost every lawyer on that list knew, in theory, that they should verify. Knowing was never the problem. Not having a system was.
The firms that stay off it don't rely on willpower at 11pm. They have it built in. The team is trained on what AI is actually for. There's a policy on which tools, for which tasks, and what gets checked before anything leaves the firm. The verification isn't a good intention. It's a step in the process that doesn't get skipped, because the process won't let it.
That's the work I do. I help firms set up AI the right way, with the verification discipline and the confidentiality layer built into how they actually work. Not by avoiding AI. By putting a system around it.
This doesn't make you immune. Nothing does. A careless lawyer can still skip the step. But it's the difference between a firm that's exposed and a firm that's protected, and right now most firms have nothing in place at all.
If yours is one of them, that's the gap, and it's worth closing before a court closes it for you.
Comment "safe" or send me a message, and I'll walk you through what setting up a system actually looks like for a firm your size.
Jared T Nelson retweeted
Zhu Zhisong, ex-Party chief of Shanghai's Pudong New Area, was given a suspended death sentence on Tuesday for taking more than 139 million yuan (US$19 million) in bribes between 2003 and 2024.
A court in eastern China's Jiangxi province also stripped him of political rights for life and ordered all his personal property confiscated.
Jared T Nelson retweeted
New paper: every law in America is technically public. But not really, until now!
With @DenisPeskoff at UC Berkeley, we built a corpus of ~every publicly accessibly city and county law, and released a huge chunk of it!
2.2 million laws, you're (probably) covered in it!
🧵
Jared T Nelson retweeted
I got thousands of rare Cultural Revolution-era photos from a historical archive here in China.
I believe some of these have never been posted online, or only exist behind extremely expensive licenses.
So here they are, for free. Open the thread for more.
Jared T Nelson retweeted
Crowded Shanghai lilong alleys, summer of 1990.
Back then, many of Shanghai’s narrow lanes were so tight that only one or two people could pass at a time.
On sunny days, residents would drag their washing machines out into the alley. There were no dedicated water hookups inside the homes yet — they’d connect the hose straight to the public tap, and the dirty water would simply be poured straight into the gutter.
📸 Gong Jianhua龚建华
Jared T Nelson retweeted
> Your boss asks you to lead a project
> You need a legal agreement
> You open your LLM of choice
> “Draft me an agreement.”
> “Make it airtight.”
> “Add a custom provision for this weird edge case.”
> LLM delivers a 25-page, hyper-protective agreement
> You don’t fully understand all of it, but it sounds right
> You send it
> Counterparty opens their LLM
> “Find every issue.”
> “Rewrite this in our favor.”
> “Also add protections so we don’t get burned.”
> You get back a 35-page redline
> Half the comments contradict yours
> Some provisions now interact in ways you don’t understand
> Your boss asks:
“Are we covered here?”
> You pause
Because now the real question isn’t:
“Can an LLM draft an agreement?”
It’s:
“Do I actually understand the risk I’m signing up for?”
And:
“If this goes sideways… who owns that decision?”
What’s your next move?
Do you get it now??
We will likely have more lawyers in the future than today, because:
1) AI will cause so many more people to ask legal questions which will encourage them to need to verify or execute through an actual lawyer.
2) AI will cause an explosion of more and more exotic legal terms that lawyers will be spending even more time reviewing redlines or new cases around.
3) All the new areas of law that now are emerging around the use of AI itself in every single industry. AI introduces an explosion of IP, privacy, and regulatory compliance challenges across all verticals.
This has historical precedent as well. Between the creation of the PC and the internet (both technologies that made the legal profession far more efficient), the ABA pegs active attorneys having gone from roughly 400,000 in 1975 to roughly 1,375,000 in 2025.
When we make professions more efficient and automated, often demand for them goes up not down.
Jared T Nelson retweeted
Alibaba Group introduces Wukong, an AI-native enterprise platform engineered for agentic execution in business environments, available as a standalone desktop application or embedded within the latest version of DingTalk.
Developed by the Wukong Business Unit under the newly established Alibaba Token Hub (ATH) Business Group, Wukong can coordinate multiple agents to handle complex tasks within a single interface 👇
Jared T Nelson retweeted
Mapping of GenAI use cases in a lifecycle of legal work | Singapore 🇸🇬 Ministry of Law
mlaw.gov.sg/files/Guide_for_…
TY @Og_Dickson #AIGovernance #LegalTech
