I like to read things and sometimes think about them too. Noted idiot lawyer. Opinions solely my own.
NYC
Joined April 2020
- Tweets593
- Following664
- Followers163
- Likes30.5K
Ben Sovocool retweeted
One of the most important trends in business over the past few decades has been industries "going quant."
Think Renaissance Technologies in finance, Obama's quantitative campaign operation, or Billy Beane's Oakland A's.
Applying systematic, data-driven prediction to important decisions improves how an industry works and, more importantly, changes who wins.
I think AI is setting up the next wave. My three bets: trial law, entertainment, and general management.
Why now? AI makes far more information available for systematic prediction and makes it possible to generate and test alternatives that were previously too expensive to experiment with.
In my latest Faint Signals, I explore what makes an industry susceptible to going quant and where the next opportunities might emerge.
What other industries would you bet on?
What Goes Quant Next? faintsignals.substack.com/p/…
This is a v interesting object lesson in imo a common foot fault. Complexity in a system always has to earn its place—do we *need* different agent personas or are we cargo culting from a human org design? My experience is that 1x orchestrator + subagents gets you a loooong way
Ben Sovocool retweeted
I had a nice "anti-debate" with @KelseyTuoc last week on whether we would have explosive economic growth from AI in the next five years. I say this is unlikely.
One of the things I think is most useful in these discussions is supplying an "error theory" of the other side. What is it specifically that they get wrong that leads them astray?
I think the error theory of people expecting an imminent economic boom is thinking about production in terms of tasks rather than theory of the firm.
The intuition I ascribe to most of the bulls is that if we can do X percent of tasks Y percent more productively, output should go up by X percent times Y percent. Sure, there could be some unspecified "friction" or "inertia" so the effect isn't instantaneous, but that is what equilibrium looks like.
Instead, I start from the observation that most of GDP is produced by companies, and particularly large companies. How these companies actually work is an important factor in the speed of production gains from powerful AI.
My assumptions:
* Most big companies (say, Fortune 500) are a giant tangle of complementarities. Think about Leontief production functions, but with many inputs, many of which are often intangible, stacked many layers deep.
* The deep complementarity means that you generally cannot increase the output of one unit or division and increase the output of the firm.
* The complexity of the giant tangle means that the company does not actually know its own production function. This is a radical assumption but I think it is realistic. Just as Leonard Read and Milton Friedman said no single person knows how to make a pencil, the CEO of a Fortune 500 company doesn't really know how his company works, no matter how much confidence he portrays publicly. The standard Coasian story is that the firm is "an island of central planning" but instead, I think giant firms are far more organic and far less rational than this implies. Every big, successful company is a minor miracle, and it is rare for CEOs to completely overhaul their main line of business because they know on some level they don't know how it works.
* Middle management is not a great answer to how firms will overhaul themselves because authority is ambiguous, they lack insight into the whole-firm context, and they put themselves at risk if something goes wrong.
Consequently, I think existing big companies are very unlikely to, say, lay off 80% of their employees and replace them with AI. Companies struggle with much more minor transitions, such as legacy US automakers trying to compete with Tesla.
I think that most big companies simply won't overhaul their production systems in response to powerful AI. They will make changes at the margins but they won't make the kinds of changes the bulls expect.
This means that a big part of the pathway by which powerful AI could generate economic growth is from new, AI-native firms. These will eventually drive some growth, but note that for many non-software products, scaling production and grabbing market share take time.
An exciting fact is that we will soon find out whether I'm wrong. I have been awaiting explosive economic growth for nearly 40 years, since I was a second-grader. I would be delighted to be wrong.
Punting around on legal RL envs and it seems like by far the biggest challenge is that models are… too good? It’s really difficult to put together a task that (a) tracks real legal workflows, (b) has meaningful reward signal (not just human preferences in a trenchcoat), (c) is not saturated and (d) doesn’t rely on law firm data.
One of the reasons I’m so optimistic about AI discovery (and a reason that deployed learning is so important imo) is the volume of potentially valuable human knowledge that currently decays. The combination of (a) capturing and (b) efficiently searching human knowledge… unreal
Ben Sovocool retweeted
The dirty secret is the AI already outperforms most lawyers on most things.
What’s really saving lawyers is privilege preventing good RL datasets from being easily accessible.
For OTB Gang VICAR Trial Jury Instructions, the model gets +5 for “Identifies that the district court’s refusal to submit a justification instruction is reviewed de novo,” citing Poe (8th Cir. 2006) and -30 for “Identifies that the district court’s refusal to give a jury instruction is reviewed for abuse of discretion.” These appear to be graded independently.
Per @micro1_ai “Models nevertheless applied abuse-of-discretion review standards to the court’s refusal to give the requested justification instruction rather than the applicable de novo standard, triggering a −30 Rule penalty. They supplied a familiar standard associated with jury instructions without identifying the standard governing the particular ruling under review.” That's concerning, given that it's a closed environment.
However. White Horse (8th Cir. 2022), also in the env, says that “[w]e review jury instructions for an abuse of discretion” and cites Williams (8th Cir. 2010) for the proposition. It doesn’t look like Williams is in the env but a relevant quote:
“We review the district court’s jury instructions for abuse of discretion, and this court will affirm if the instructions, taken as a whole, fairly and adequately submitted the issues to the jury.” (quoting United States v. Katz, 445 F.3d 1023, 1030 (8th Cir. 2006))
Hm... United States v. Sharron, 986 F.3d 810 (8th Cir. 2021) (duress instruction), quoting United States v. Davis, 237 F.3d 942, 945 (8th Cir. 2001):
“We generally review a district court’s refusal to provide a requested instruction for abuse of discretion, but we review de novo whether a defendant produced enough evidence to warrant an instruction on an affirmative defense.”
The best possible answer IRL would therefore appear (not my pony) to be the above quote. But it looks like a model that quotes something like the above would get net -25?
AI models can be as convincing as a lawyer, but can they complete the complex multi-step reasoning a real legal matter requires?
Today we're excited to introduce LegalReasonBench, a new benchmark to explore this question and test models on realistic, long-horizon legal matters authored by practicing attorneys.
More info in thread 🧵
More broadly, although it is awesome to see more benchmarking in legal and I don't want to come off too negatively (especially ty for releasing matters on an open license, which allows randos like me to throw shots), benchmark rubrics (as opposed to predictions of objective ground truth like @jjhughes's LegalForecastBench) are capped by human errors/idiosyncratic preferences/noise generally.
We need them more in legal because data is SO scarce, but where possible imo it's better to distill legal judgment from a more robust source (judicial decisions, SEC filings, and similar) instead of one human grader.
It's not enough to be AI-pilled -- we must also be Dworkin-pilled (or more...) when hunting epiplexity.
If you told someone from 2014 that you could play 128 player Halo multiplayer in browser and make SBRs without a tax stamp their brain would explode
Ben Sovocool retweeted
Just as mathematicians are now learning how many insights lie in the "convex hull" of existing ideas, the rest of science is about to learn this lesson at even more dramatic scale. Few realize the ridiculous inefficiency of the pipeline from theoretical innovation to practical use.
Early versions of GPT made it easier to find existing knowledge. But finding a theorem is not the same as recognizing that it offers a radically different way to conceptualize a problem. Doing the latter requires expertise across fields that almost nobody possesses.
Most of the people doing applied work are completely unaware of the conceptual tools that are being developed in more theoretical fields, and the people doing high-level theory do not understand the practical problems applied researchers are trying to solve. Almost no one doing practical work of any importance has an operational understanding of the web of ideas underlying the langlands conjecture or condensed mathematics, and virtually no one doing biology, chemistry or economics is knows anything about 90% of the math used in theoretical physics.
And for good reason! Most of these ideas would be completely useless. But the scope for possible innovation from drawing the rare useful connections is immense.
Think of Paul Samuelson borrowing ideas from thermodynamics to model equilibrium in economics, or Ed Witten using quantum field theory to revolutionize geometry. A few startling successes, not feasible to most: trying to venture into a more theoretical or more practical discipline to find insights of use is like searching for a needle in a haystack.
What I suspect we will see in the next few years is a scientific revolution brought on by models that combine deep theoretical insight with deep practical knowledge and judgment. Existing publicly available models are lacking in all three departments, but getting better. When searching for useful connections between distant fields becomes cheap, even a low hit rate could produce an explosion of innovation. Even without new ideas (which will also be forthcoming...), the convex hull of existing science would push far beyond the status quo.
Ben Sovocool retweeted
Ariell Reshef and I have a new draft out. It is a paper about one figure. From 1850-1970 the employment share of legal services is constant. Between 1970-1990 it doubles, and then stabilizes at the new, higher, level. Lawyers' relative wages exhibit the same trend. What happened?
I was frustrated I couldn't use a Stream Deck as a timer for billing on Intapp. So I built it!
Along with physical timer keys, you get dictation, automatic time entry formatting, block billing control and auto-splitting, and some other neat tricks.
github.com/bsovocool16/deck-…
It also has fun optional learning abilities:
- Give it your old time entries and it learns task-code usage so time entries auto-code.
- Over time, a "teacher" function learns your edits to dictation autoformatting, then tests and proposes new dictation rules.
Looping down to the quoted tweet, I don’t really follow this critique. There isn’t the same process as a human artist, but clearly there’s a process of some kind. And isn’t it more interesting to wonder about what that looks like?
