iiinova — where your digital self keeps evolving

Danny Wilf-Townsend @dannywt.bsky.social · 09/02/2026

Of course, that doesn't tell you what proportion of lawyers are "current users." Last spring, Bloomberg found that a majority of lawyers "have used" generative AI for work, which is at least something, but isn't quite the same as "currently use."

120

Danny Wilf-Townsend @dannywt.bsky.social · 09/02/2026

A periodic update about the frequency and intensity of AI use in legal practice: Thomson Reuters reports that 55% of generative AI users at law firms use it at least daily, with 30% multiple times a day:

110

Danny Wilf-Townsend @dannywt.bsky.social · 18/12/2025

For people who are pretty current with conversations about regulating AI, these two paragraphs are the crux of the post. This is responding in part to writing by @deanwb.bsky.social, Ketan Ramakrishnan, and @milesbrundage.bsky.social on entity-based paradigms for AI regulation.

120

Danny Wilf-Townsend @dannywt.bsky.social · 01/10/2025

Overall, GPT-5-Pro was good enough to use for my (informal) approach here—it was both internally consistent and looked good in accuracy spot checks. Its grades show some models scoring in the A- to A range, consistent with what others have found, too.

000

Danny Wilf-Townsend @dannywt.bsky.social · 01/10/2025

It turns out that some models are deeply inaccurate, and some are frequently inconsistent, but a few are reasonably consistent and accurate. And along the way, I learned that human graders are sometimes less consistent than we might hope.

Text describing consistency rates in human graders

100

Danny Wilf-Townsend @dannywt.bsky.social · 17/06/2025

I think this is one of the more common mistakes I see with people trying AI—the idea that if you go to a free chatbot, quickly run a question by it, and it does a bad job, then you've learned that AI cannot do a good job on that question.

A screenshot excerpting a blog post with the following text: Relatedly: if your goal is learning “what AI can do,” there is an asymmetry in how much information you get when AI fails at a task versus when it succeeds. If you try giving an AI tool a task, and it does a bad job at it, you have some evidence that it’s not good at that task—but it could very well be that a different model, a different prompt, or giving the tool more context (such as uploading relevant documents) could give you a different result. In contrast, if you figure out a way to get the AI to successfully do a task, you have strong evidence that the tool at least can be successful at that task, even if the tools’ probabilistic nature means that you should be cautious about assuming equally good results will occur on every run.

Because of this asymmetry, it is particularly worth varying up your approach and trying different ways of tackling a problem if you don’t get success on the first attempt. Try telling the tool that it’s wrong, or what it did wrong, or giving it more information. It is often possible that ten minutes and half a dozen different attempts at a task will move the result from bad to good.

020

Danny Wilf-Townsend @dannywt.bsky.social · 12/05/2025

The use of generated legal texts raises familiar issues in the world of AI (e.g. bias, inaccuracy) but also some distinct concerns (insincerity, a flood of documents). Check out our paper here: papers.ssrn.com/sol3/papers....

010

Danny Wilf-Townsend @dannywt.bsky.social · 12/05/2025

I've got a new paper up with the inimitable @kevintobia.bsky.social: "Generated Legal Texts"—about texts generated by AI and used in legal institutions. These texts are arising frequently in legal contexts around the world, perhaps faster than many realize. And, we argue ...

Abstract
Generative AI’s sudden growth has transformed the production of data,
images, videos, and especially text. And with that impact on text comes
impact on law. From contracts to judicial opinions, from legislation to
litigation, nearly every aspect of the law operates through text. Yet
while there is much legal scholarship on AI, little work focuses on legal
texts as the relevant unit of analysis.
We introduce the concept of “generated legal texts,” arguing that how
these texts are deployed across different legal institutions creates
common patterns and concerns—making them worthy of study in their
own right. Through a broad empirical survey, we document the breadth
and speed of generated legal texts’ integration into our institutions,
revealing how they are already reshaping the legal landscape.
Next, we classify the types of these new generated legal texts, the forms
of human co-production of these documents, the texts’ audiences, and
the spectrum of regulations concerning generated legal texts. Generated
legal texts raise concerns such as bias and inaccuracy that are common
with AI. But they also raise distinctive new issues, like “floodgates”
problems and concerns about sincerity. We identify and elaborate on
those concerns, and then discuss the emerging putative solution of
“ratification,” in which an individual or body assumes responsibility for
the text by “ratifying” it. By drawing on embedded legal norms,
ratification has many strengths as a tool for shaping the responsible use
of generated legal texts. But it also has limitations, reflecting the need
to pay close attention to generated legal texts going forward.

111

Danny Wilf-Townsend @dannywt.bsky.social · 22/04/2025

This blog post also contains my favorite AI+Law anecdote so far from the teaching-and-talks circuit.

Text from the linked blog post: One of my favorite anecdotes about AI and legal practice comes from a talk I was giving last spring to a group of lawyers. In the Q+A after the talk, one law firm partner raised his hand and told me about his week. Early in the week, his firm had received a notice from its malpractice insurance provider. The insurance company was giving notice that its malpractice policy did not cover any task on which generative AI was used. Fair enough—with legal news coverage replete with stories of fake cases and other hallucinations, it seems like an insurance provider might want to be cautious about the technology. But then, later in the week, his firm got a letter from one of their biggest clients, a large technology company. The tech company said that it would no longer pay for associates’ time on routine tasks that could be handled by generative AI—in other words, asking the law firm to start using generative AI instead of associates on a number of tasks, to save money.

052

Danny Wilf-Townsend @dannywt.bsky.social · 17/04/2025

I haven't spent much time with the new o3 model from OpenAI, but it is the first model to get all of the questions right on the informal testing of legal questions that I've been doing for a while when new models come out.

A table of results showing how different generative AI models perform on a range of questions about the law of civil procedure

010

Danny Wilf-Townsend @dannywt.bsky.social · 09/04/2025

Several good things about this order. A small, technical detail that is good but easy to miss is the court's ruling on FRCP 65(c). In other times this would be unremarkable (or just absent from the order), but the Trump admin has tried to use Rule 65 to make this kind of civil action harder.

A screenshot of the court's order, stating "Based on the record and the applicable law, the Court also finds that no security is required at this time under Federal Rule of Civil Procedure 65(c). See Kaepa, Inc. v. Achilles Corp., 76 F.3d 624, 628 (5th Cir. 1996) (noting that a court “may elect to require no security at all” under Rule 65(c))."

120

Danny Wilf-Townsend @dannywt.bsky.social · 20/03/2025

A mandatory arbitration agreement that requires subjecting disputes to an AI system. What could go wrong?

010

Danny Wilf-Townsend @dannywt.bsky.social · 11/03/2025

And here is a slightly different kind of problem, from another almost-a-real-case hypothetical involving copyright and disgorgement: 7/

110

Danny Wilf-Townsend @dannywt.bsky.social · 11/03/2025

And one from the consumer finance context involving a customer chatbot: 6/

110

Danny Wilf-Townsend @dannywt.bsky.social · 11/03/2025

I give a few examples in the paper from different contexts, and discuss a few different problems and implications. Here's a hypothetical from the antidiscrimination context, inspired by a real case involving a content moderation tool: 5/

110

Danny Wilf-Townsend @dannywt.bsky.social · 11/03/2025

In the growing world of AI litigation, class actions are going to be particularly important. In a new paper forthcoming in the Wash. U. law review, I look at how class actions will influence the efficacy of AI regulations, and how they might even open up some new options for regulators. 🧵

194