273 Ventures

273 Ventures

Technology, Information and Internet

Chicagoland, IL 833 followers

A new OS for legal data. Designed for your past, built for the future. Order from chaos.

About us

273 Ventures is developing a Legal Data Operating System (OS) to organize and connect data from various sources within legal organizations, including documents, dockets, and other sources. It includes legal-specific capabilities such as OCR engines and legal dictionaries, as well as connectors for language models like GPT-4. The OS is intended to be a purpose-built platform specifically designed for the legal industry, with an emphasis on compliance with information security standards and data protection laws. It is expected to be available in 2023 and is intended to help legal organizations increase productivity and better leverage data. LEARN MORE -- 273ventures.com kelvin.legal

Industry
Technology, Information and Internet
Company size
2-10 employees
Headquarters
Chicagoland, IL
Type
Privately Held
Founded
2022
Specialties
legal tech, data science, nlp, artificial intelligence, and legal data

Locations

Employees at 273 Ventures

Updates

  • View organization page for 273 Ventures, graphic

    833 followers

    Interested in how to engage in responsible AI projects in legal? Spoiler alert: you don't have to sacrifice your ethics to utilize AI! Our own Jillian B. talked to Marco Mendola and Hadassah Drukarch about this in their latest The Law of Tech podcast episode.

    View organization page for The Law of Tech, graphic

    2,808 followers

    Are you fed up with all the noise in the Legal AI space? To orient yourself better within the jungle of Legal AI, Marco Mendola and Hadassah Drukarch have interviewed a keynote expert in the field: Jillian B., Chief Risk Officer at 273 Ventures. A background in accounting and many years of experience in due diligence, compliance, and risk management across various highly regulated industries. Titled “At the intersection of AI and ethics in LegalTech: a deep dive into KL3M”, in this discussion we went beyond our comfort zone and expertise discussing with Jillian how KL3M is become the world’s first-ever ethically trained legal LLM, unpacking the practical applications of KL3M in the legal context, and stressing the need to foster of ‘Responsible AI’ projects. At the core of our chat, we also explored business purposes, and how to deal with assumptions and general misconceptions when running and piloting tech solutions in legal. Without further ado, take a listen to our latest episode on your podcast platform of choice ➡️ https://lnkd.in/emDPCgrd. #legalai #largelanguagemodels #legalinnovation

    • No alternative text description for this image
  • 273 Ventures reposted this

    View profile for Michael Bommarito, graphic

    Adviser, entrepreneur, educator, investor.

    Thanks to 48 hours on 64 H100s, kl3m-1.7b now has an 8K context window. You might ask, can an SLM like that really "use" a window like that...? Well, we'll have more results on that soon, especially as it relates to using activations from decoder-only models like this as "free" embedding models too.

    • No alternative text description for this image
  • 273 Ventures reposted this

    View profile for Jillian B., graphic

    AI + tech risk management for the legal industry | Co-Founder + Chief Risk Officer

    Big day for AI and regulation, as the FTC just announced an inquiry into Alphabet, Amazon, Anthropic, Microsoft, and OpenAI “requiring them to provide information regarding recent investments and partnerships involving generative AI companies and major cloud service providers.” Many of us were also attending today’s FTC’s Tech Summit while digesting this announcement.  A number of the speakers, including Lina Khan and other FTC leaders, addressed the importance - and absence - of competition in this space. The first section of the Summit covered hardware, highlighting NVidia’s incredible concentration (as we can attest to with our 48 A100/H100s). A few other points made during the event that particularly resonated with me: 🔨 Lina Khan reiterated that model disgorgement is going to be a significant method of enforcement against companies that have engaged in anticompetitive, deceptive, and unfair business practices to train those models. 🤫 One of the reasons that companies are being so secretive about their model training data is because they are scared about the regulatory/judicial implications that are coming down the line. From my perspective, this means that companies are aware that they have trained on questionable data. I acknowledge that there are very valid reasons to keep this information private. No one would fault Coca Cola for not being transparent with their recipe, but at the same time, they would expect them to represent that the recipe doesn’t contain illegal ingredients. 🤷♀️ Another panelist made the point that you’re not excused from the law just because you’re training a model. 🏅 Based on what we heard during this Tech Summit, I’m feeling pretty gosh darn good about the squeaky clean large language models that we’re building from scratch at 273 Ventures.  We’re already testing our smaller, fully-baked models against Llama2 7B and gpt-3.5-instruct, and let’s just say that we think the conversation is about to change… Next time someone tells you that it’s impossible to build a model without stealing data, send them to us 😉

    • No alternative text description for this image
  • View organization page for 273 Ventures, graphic

    833 followers

    Jillian Bommarito, CPA, CIPP/US/E is spot on here - got to be LLM Agnostic as a business continuity risk, etc.

    View profile for Jillian B., graphic

    AI + tech risk management for the legal industry | Co-Founder + Chief Risk Officer

    Well, that was a dramatic weekend.  Though it seems unwise to make any claims before the dust finishes settling, there is one clear learning so far - never underestimate the importance of supply chain risk management.  Back in February, I warned that “if interruption or discontinuance of [services] wasn't already on your risk register, it better be now. It's time to start thinking about 1) what your backup plan is if a particular vendor (OpenAI in this case) is no longer available, and 2) what your overall AI strategy entails (e.g., building your own LLM, managing regulatory/statutory risk).” AI-enabled products and services rely on models. Whether they are open source models that are used locally or proprietary commercial models that can only be accessed through a web interface or API, each method carries its own set of risks. When we built Kelvin, we worked hard to make it model agnostic, so that our customers could choose the model and access methods that best suited their business and needs. We’re also working harder than ever to help customers use and build models they control on their own hardware.  Unlike other competitors in the market, a seismic shift in OpenAI’s (or any particular other model provider’s) availability, access method, cost, or contractual terms isn’t a game changer for us. Though this wasn't on our bingo card, we did expect the game to change and built our product to withstand it. 🎯 CC Jessica Katz Daniel Katz Michael Bommarito

    • No alternative text description for this image
  • View organization page for 273 Ventures, graphic

    833 followers

    View profile for Michael Bommarito, graphic

    Adviser, entrepreneur, educator, investor.

    Happy Halloween from our Headless AI to you and yours! If you look closely, you'll see that this WordPerfect-themed command line interface shows the table that Daniel Katz posted yesterday (including Lucille's bonus). Would you really run an M&A diligence checklist in the command line? Probably not (unless you're me). Would you really design a modern UI to look like Halloween WordPerfect? Sadly, no. So, what's the point of this screenshot? It proves that behind our chat interface, the Agent has no Head - or more accurately, that you can easily swap user interfaces, should you so desire. The beauty of Kelvin Agent is that you can embed its capabilities into any product or system you already have, from vanilla O365 to whichever DMS, ELM, or CRM you live in day-to-day. Stay tuned for more content from Jessica Katz and Jillian B. on agents soon, including what they are, how to manage them, and how they work.

    • No alternative text description for this image

Similar pages