←
Ben Sovocool639 posts
Ben Sovocool avatar

Ben Sovocool

@BenSovocool

I like to read things and sometimes think about them too. Noted idiot lawyer. Opinions solely my own.

◷ Joined April 2020▧ 23 Photos and videos
Ben Sovocool avatar
Ben Sovocool@BenSovocool·3h

You can go to a used clothing store (in NYC, Gentleman’s Resale has a lot) or EBay and walk out with a stack of gorgeous ties — Drake’s, Charvet, Kiton, etc. — for less than the cost of a decent dress shirt. Skip the Hermes (although they are nice) and it’s open season.

Ben Sovocool avatar
Ben Sovocool@BenSovocool·4h

One of the underrated pleasures of ties is also that very nice used ties are now dirt cheap. You can swag yourself out on a price/performance ratio that is basically impossible anywhere else. And people are scared of nice fabrics now, which enhances their power…

Been wearing neckwear lately. Ties. In San Francisco! People definitely notice. They comment. Nobody comments if you wear a scarf. Same concept, you knot a piece of fabric around your neck, totally different social connotations. One is like my "neck is cold," or "I want to accessorize". The other is like "I'm going to court." Nobody wears a tie on the West Coast. Even back east. Even in Europe! Around the world, everybody has a collar and no tie. If someone wears a tie, they are one of a few things: rich, from the government, going to a party, selling you something, killing you, or late to a funeral. We're almost to an equilibrium where ties will die out. The second day that nobody in the world wears one, you know, The first could be a fluke. But it's been a campaign against them for 50 years. Since the boomers at least, every generation of Americans has thought of themselves as rebellious upstarts, and ties represent The Man. And people who want to be The Man take them up reactionarily. But that's not what I mean by it. I just like the way it looks! My neck is cold. I want a fun accessory. It's kind of funny. It confuses people. It's a latent meme still in the culture, and I want to wear one, and see what happens. You can do it too. There is still a subtle expectation, a fashion gradient, pulling on you. There are gestures you don't even know that you know how to make, learned from old movies. You know it looks cool. It would make your clothes make sense. It comes from wartime, right? Croatian mercenaries wearing different colors on their necks to show their allegiances. And then Louis the 14th was like "I fw what those Cravats are doing. Everyone has to wear that now". Are you serious about your life? Do you think you're better than Le Roi? You can wear a tie and program computers. You can wear a tie and dig a hole. You can wear a tie and sell drugs. You can wear a tie and write tweets. You can wear a tie and be a soldier, or a missionary, or a king. You can learn the language of knots, and fabrics and collar. It's way easier than video game lore. You can look better than people in New York because they If someone comments, and they will, you can just say "Yeah I noticed you're not wearing one of these" and they will remember that everybody is supposed to wear one. And then they will apologize to you, for not wearing one. And then they have permission to get a fun accessory. And that is how it spreads.

Ben Sovocool avatar
Ben Sovocool@BenSovocool·Nov 17

Harvey now does this really neat thing where it delegates all or nearly all drafting to subagents, then when you ask the main agent to explain “its drafting choices” it (correctly) identifies a bunch of extraneous or shitty drafting…

Ben Sovocool avatar
Ben Sovocool@BenSovocool·Nov 17

More broadly people need to remember why we have different roles and different people in organizations. The model does not need to put on five different little hats because there are five different roles in your org that do the things the model will do. Different substrate.

Please don’t replicate the organizational structure of your company in my SaaS tool. Thanks

Ben Sovocool avatar
Ben Sovocool@BenSovocool·Nov 17

Please don’t replicate the organizational structure of your company in my SaaS tool. Thanks

This by Anthropic seems both right & a condemnation of many “agentic solutions” & harnesses being built at companies. Rather than a sprawl of specialized subagents built by different teams, you often get better & cheaper results using one agent to do it all claude.com/resources/article…

Post attachmentPost attachment
Ben Sovocool avatar
Ben Sovocool@BenSovocool·Nov 12

“Leibniz spoke of this…”

Since our pitch deck was uninformative, let me try to explain our whole thesis. Sorry if any part of it sounds obvious. I argue that any reasonable definition of AGI must imply a system capable of learning as efficiently as a human. LLMs obviously don't pass this bar. When you talk to them, they don't learn anything, since the weights don't change. So, unless true infinite context lands, or GPUs become 10000x cheaper, the current methods won't scale to AGI, at least in that sense. (Note that this has nothing to do with raw capabilities. Perhaps Fable 10 will be building Dyson spheres, yet still be unable to internalize new knowledge as efficiently as humans do. So, one might ask if we even *need* AGI in the sense above. I think we do, if only because a drastic drop in pre-training costs would make AI more democratic, and prevent one of the worst outcomes in my mind: over-concentration of power.) As you know, modern AI is based on neural networks. What not everyone knows is that NNs are just very, very slow programs. Nobody in their right mind would sort a list using NNs, because that'd be millions of times slower than just using C. So, why do we implement *intelligence* with NNs? As a thought experiment, imagine if we could take Fable's weights and distill them, not into a smaller model, but into a raw C program that was observationally identical to it. Literally, a C file with astronomically complex, tangled procedures that, when compiled and executed, behaved just like Fable 5.2. If such a thing were possible, then we could have Fable at a million tokens per second. And here's the cool part: such a thing *is* possible. At least, as far as computer science is concerned, nothing says it isn't. So, why don't we do just that? Because we don't know how to *train* a raw program. Neural networks are slow, but they're differentiable, so we know how to train them. Step by step, slowly, but surely. Raw programs are fast, but we have no methods capable of training them. So we don't. We just stick to NNs, and accept that it takes $50 billion to train a model. We take that as "normal". But is it? In theory, it shouldn't be like that. As the old saying goes: first you make it work, then you make it fast. We know AI works. Now, we will make it fast. That's the premise behind Symbolic AI. HOC's first milestone is to train a model comparable to GPT-2, with no NNs. If that works, we might have the fastest inference engine in the world. Not by 2x, but by 1000x. So, in a way, HOC is like DeepSeek or Jev, in the sense that we're all about efficiency, but in a much more extreme way. People are aiming for hardware, but we believe there are still 1000x gains to extract in the software layers, if we're willing to rethink the whole stack from literal scratch. Many laugh and say this is impossible, but they say so with no basis. They just observed that many failed to build symbolic AI models in the past, and concluded it is impossible, or very hard. That's not a formal proof. In fact, there are many reasons people in the past could've failed, yet the actual problem may end up being easier than most anticipate. Symbolic AI flourished half a *century* ago. Back then, they didn't have: - As much knowledge as we do - Massive GPU clusters - Methods to scale Symbolic AI (Bend3, HVM5!) So, just like everything that once seemed impossible, there's one core prerequisite to making it possible: *trying*. That's what we'll do! I don't think I'm smarter than all the brilliant researchers who worked on this problem in the past, and failed. I just think I'm equipped with tools, compute, and knowledge that they didn't have. So, standing on the shoulders of giants, and with enough dedication, I don't think it is so unlikely we stumble upon some symbolic algorithm from which, given enough compute, unexpected capabilities emerge. If it does work, I do think business is straightforward. At the very least, we would be able to sell incredibly high quality datasets to big labs, but, at that point, just selling inference might work as well. Business is obviously not the wall, it is all on whether such massive software-based speedups exist, and whether we're able to discover them.

Ben Sovocool avatar
Ben Sovocool@BenSovocool·Nov 10

Simulated lawyer conversation re: benchmarks: >lawyer 1: did you see that [insert mystery meat model of the weak] beat [claude/GPT] on [legal benchmark]? >lawyer 2: nope. that’s weird >l1: yeah >l2: … ok. … >l2: I wonder if they’ll give us [claude/GPT] soon. >l1: idk man

I mean, at the end of the day benchmarks have value if and because they’re good proxies for actual economically valuable work tasks. In this case (a) the benchmark has issues and (b) legal procurement ppl DGAF about benchmarks to begin with so benchmaxxing would be -EV

Ben Sovocool avatar
Ben Sovocool@BenSovocool·Nov 6

Really looking forward to this discussion and would encourage everyone to save some time next Thursday at 5pm PT for this discussion. If (like me) you have strong opinions on the nature of “legal judgment” this should be right up your alley.

Next Thursday (Nov 12), @zebanderson, @abraham_jmb, @aaronklaw, @BenSovocool, and I will host a live X Space on legal AI: How do we measure the quality of AI-generated legal analysis and judgment? How do we build that quality into AI systems? And how do we train for it? Join us for the conversation!

Ben Sovocool avatar
Ben Sovocool@BenSovocool·Oct 31

These do appear to be legitimate, although I defer to German lawyers on whether the citation to (1a) versus (1) is material. Finding issues like this (and ideally finding ways to monetize them as an outsider) has been an interest of mine recently, so here are a few thoughts. 1. The German legal system is distinct in being particularly rigid and therefore particularly vulnerable to exploitation. The more flexible the system is, the less vulnerable to exploitation it is. 2. However, there are many areas in the U.S. law where consequences are more mechanical. For example, one could automatically identify short-swing trades and locate stockholders to file a claim for disgorgement. 3. Generally speaking, the limiting step for exploitation is standing rather than discovery. The examples provided in the QT presume that an aggrieved individual is aware of the flaw, but this is quite distinct from cybersecurity in that a random third party generally cannot directly exploit the vulnerability. We see in mooting disclosures that "vulnerabilities" in proxy filings are widespread, but that the limiting step for a commercial business is actually finding shareholders. 4. A more pernicious example is ADA suits, which are a serious drag on legitimate businesses. I think the bigger attack vector here is an explosion in semi-legitimate claims (welfare, ADA, similar) rather than "exploits" but they are of concern nonetheless. Overall, I am very glad to see more attention being paid here. I would love to see more attention paid on preemptive clean-up efforts to work through existing regulation.

We have discovered a substantial number of vulnerabilities in the German legal system using LLMs. These may enable access to confidential documents, travel without an otherwise required visa, enforceable leave for soldiers at NATO’s JSEC, and more. Similar to computer systems, the written law can contain vulnerabilities that invite exploitation. In our blog post, we share a selection of vulnerabilities, and present potential exploits for each. To our knowledge, none of these have been previously recorded. We share only those vulnerabilities whose exploitation we believe would not cause serious harm. blog post: felixringe.com/legal-exploit…

Post attachment
Ben Sovocool avatar
Ben Sovocool@BenSovocool·Oct 29

The creative pressure of 20-odd new followers is pushing me to the brink... In lieu of new thoughts please see my github webp: bsovocool16.github.io/index.… (I wrote on github b/c my prior employer's webfilter blocks github). cheers and DM for free rants on the SEC Plain English Rule

Ben Sovocool avatar
Ben Sovocool@BenSovocool·Oct 29

Like so...

Post attachment

I don’t think Weil can straightforwardly sell its own data (which mostly belongs to the clients), but could it drop a SPV (DataCo) that manages the custodianship and return/destruction in exchange for an upfront payment by a data buyer (& license)?

Ben Sovocool avatar
Ben Sovocool@BenSovocool·Oct 28

Weil has to pay for custody of documents in any case, so this would defray those costs and provide some return to clients who consent to redaction+license of their data.

Ben Sovocool avatar
Ben Sovocool@BenSovocool·Oct 28

I don’t think Weil can straightforwardly sell its own data (which mostly belongs to the clients), but could it drop a SPV (DataCo) that manages the custodianship and return/destruction in exchange for an upfront payment by a data buyer (& license)?