V @_vwood

@JesterInABarrel But surely there’s some method to this madness..

We’ve rediscovered trains

Boden | China Views @BodenViews

2 months ago

China is testing a different way to automate trucking.Instead of replacing every truck at once,one driver can lead a convoy while autonomous trucks follow behind.This kind of deployment could make freight automation arrive much earlier in controlled logistics sites than on roads

227 805 5K 2.4M 1K

I used Counsel Stack to conduct a multi jurisdiction survey on TCPA and state analogs. In 2 turns it produced a full spreadsheet with links and a pivot table. The legal research took about 15 minutes using our new Claude connector and a basic prompt. It traced 57 different regimes, including state statutes from the post-Duguid "mini-TCPAs" (Florida, Oklahoma, Washington, Maryland, Arizona) through the older ADAD statutes (New York, Georgia, California, D.C., Minnesota) and conduct/DNC regimes (Texas, North Carolina, Virginia, New Jersey) to name a few. It also found modern convergence on a narrow reading of the TCPA definition of autodialer across circuits despite an older case's footnote claiming a circuit split on that issue.

keyboard_arrow_left Previous keyboard_arrow_right Next

Counsel Stack redlines to improve legal substance and accuracy with quotes from the underlying cases, statutes and regs. It builds a table of authorities linking to Counsel Stack's library. This combines the best of both worlds - state of the art reasoning in the best models and clean, curated legal data. In this example, we connected Counsel Stack's citator to Mike OSS. Props to @willchen500 and team for opensourcing a clean platform for lawyers. Below is an example response to the question "Can law enforcement obtain historical cell-site location data without a warrant?" It's a broad query that could be refined to focus on a specific jurisdiction or fact pattern. But the initial answer was quite good. In our evaluations, Mike's logic for managing a conversation and using tools was superior to base LLMs. We think this is because it has less competing instructions and is specifically designed for legal work. Sonnet 4.6 was surprisingly the most consistent model, performing better than Opus and Fable across runs.

keyboard_arrow_left Previous keyboard_arrow_right Next

We connected Counsel Stack's citator to Claude. It researches like a lawyer, based on our prompting, and provides answers with links. Counsel Stack helps Claude find relevant legal sources and crawl around our 200m edge citation graph. Below is a simple query. But Claude + Counsel Stack can also do more complex litigation workflows, edit / redline briefs grounded in curated legal data, and integrate with your other connected apps. Right now the MCP server is local, too. We host a 250gib dataset containing more than 50 million citations, including caselaw, federal statutes and regulations, state statutes and constitutions, and agency enforcement actions / adjudications. I have tested with 27b param local LLMs as well through vLLM with slightly worse results, but promising nonetheless. In the very near future you will be able to run a ~120 IQ LLM on your own machine, calling on local legal data that is updated daily from an oracle, and can answer most legal questions without connecting to a third party service. The main bottlenecks are pricy vRAM and low token throughput when using parallel sub agents. But it's coming. Stay tuned for a post about connecting Counsel Stack to the Mike OSS front end - I think its harness was superior to Claude's because Mike is purpose built for legal, and doesn't have as many competing instructions for connections. I'd like to see how this compares to other legal research tools.

keyboard_arrow_left Previous keyboard_arrow_right Next

@gooby_esq Regardless of how long the process takes, Does a 15 min expert review place the work product in the 99th percentile? That’s the main question i hear in various forms

Across 200 real legal research questions, Alexi, Counsel Stack, Midpage, and ChatGPT all outperformed practicing lawyers as rated by lawyers and law librarians. Overall, Counsel Stack came out on top. (2/5)

2025 was a great year for Counsel Stack We raised our first round of funding, got SOC2 compliance, onboarded more than 100 legal professionals, ran paid pilots, got an acquisition offer, and won VLAIR - the premier independent legal research benchmark - outcompeting teams with 100-1,000x our resources. And soon we'll announce a new partnership with one of the most well respected brands in legal tech. Here are a few things I learned along the way: First, obsession is the only way to get ahead. There is no shortcut. If there was, someone already would have arbitraged it. Everyone thinks they work hard. Work feels hard to everyone. And people like to glorify how hard they work. But there's a huge difference between a person who acts like they work hard and one who is obsessed. The fruits of obsession compound. And when you finally take off the blinders to look around, you'll be pleased to see how much farther ahead you are. You can cultivate obsession in yourself and others by being genuinely excited about solving a legitimate problem. Second, avoid talkers. People with high linguistic IQ but low execution initiative will waste your time. Instead invest in people that get things done. Third, form teams comprising people high in crystalized IQ and people high in fluid IQ. Crystalized IQ is knowledge and wisdom gained through experience, whereas fluid IQ is raw reasoning power and pattern matching ability. Elite teams pair members of both camps. Fourth, immediately write down your good ideas. Fifth, taking action in ambiguity is better than paralysis. Building a business is a series of bets. You'll rarely have a perfect dataset of information. Sometimes you just need to make a bet and see what happens. The result will at least reveal new information that you can learn from. Trust yourself to place good bets, but only if you're obsessed. If not, the house will probably win. Finally, confidence is key. Embrace responsibility and accountability. No one's coming to save you. Be the person you would have looked up to. Propel yourself towards success until it finds you. The world is indeed more malleable than one might expect. Sometimes you can make things happen through sheer willpower, and unshakeable confidence is a prerequisite for that. I certainly don't have everything figured out, but 2025 was a solid year. Stay tuned for more posts about our citation verification system.

@gooby_esq It’s more like purgatory 😂

@OrinKerr The degradation of REOP continues

I’ve been thinking a lot about the popularity of event contract markets and how they’ll impact the legal industry. Legal research helps you see where a case is probably heading. When you dig through enough opinions, you start to notice how judges handle recurring issues, which arguments survive contact with precedent, and where the real friction points sit. Patterns show up and trends emerge. Legal outcomes are not perfectly predictable, but with the right heuristics they are not a coin toss either. Attorneys understand this. But when prediction markets like @Kalshi and @Polymarket form around major cases as they have with other forecastable events, the advantage will go to whoever understands the facts and underlying political pressures, and arguably most importantly, deeper patterns in the doctrine. Then research becomes a tool for reducing uncertainty. It gives one a clearer sense of the range of outcomes and why one trajectory makes more sense than another.

@karpathy 2.0 and 2.5 family models are currently doing this, too. I sent caselaw citations from 2024 and 2025 and the models claimed the citations must be fake because they’re from the future.

@strictly_JC Neo go tell him he missed a spot

@ProfRobAnderson Seems this true of most peer reviewed journals

@SteelToeScribe @BoringBiz_ Nope

@FBGreatMoments 💪🏼💪🏼🫡 that’s flying

@nickcammarata @cxgonzalez Banger

2 federal judges sent me letters admitting their staff used AI to draft error ridden court orders Theyre now working 2ensure it doesnt happen again That’s a step fwd but we need to make sure this problem doesnt resurface in ANY federal court My oversight will continue

@jiratickets Pittsburgh

@ajhodls Deck1.png