Fringe Legal #89: Speed Matters

Is speed the ultimate differentiator? Let's explore why it might be, examining user experience, product development, and experimentation. I believe that speed is not just an advantage but a necessity.

User Experience: The Need for Speed

When OpenAI announced GPT-4o (“o” for “omni”), one of the standouts was the voice model's human-like responses (we’ll put aside the controversy). Was this because of the voice's human-likeness? Not quite.

It was the speed with which it responded, alongside the ability to interrupt.

According to OpenAI, “Prior to GPT-4o, you could use Voice Mode to talk to ChatGPT with latencies of 2.8 seconds (GPT-3.5) and 5.4 seconds (GPT-4) on average.”

With GPT-4o, the model can now respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds. That’s a significant improvement.

No doubt, a lot of clever engineering is done to make this possible. I’m relatively certain some of this is achieved by quickly generating the first few words of the responses, which in many of their demos start with “sure,” “of course,” “sounds amazing,” “hmm,” etc. This buys time to load the rest of the response while providing a great experience and a sense of magic.

At its simplest, its respond immediately and then load progressively. For complex data, return the first chunk of results to the user immediately, even if incomplete. Then, progressively load and append the remaining data. This feels much faster than waiting for the entire dataset. This isn’t a new concept—websites, apps, and even OSs have been doing this for some time.

Development: Rapid Iteration and Tight Feedback Loops

What sets startups apart from incumbents? Speed. For example, at Lupl, we have one of the tightest feedback loops in the industry (in my view) between our customer-facing and product teams. Feedback and feature requests from client interactions are relayed to our product teams within the hour - we include the unfiltered version in the clients’ own words and relevant commentary from the customer-facing team who might have additional context. This approach lets us act on requests at supersonic speed, consistently delighting our customers.

Listening closely to our users and iterating rapidly is our secret sauce. It ensures we continuously improve and adapt to our users' needs, delivering results faster.

Experimentation: The Value of Quick Trials

How do you effectively run experiments to evaluate legal tech? There are two main approaches:

1. extensive evaluation; and,
2. rapid iteration.

The first approach involves meticulous planning and evaluating every potential outcome before implementation. This method reduces risk but can be slow and resource-intensive.

The second approach, favored by many innovation teams, is rapid iteration. Quick trials allow for immediate feedback and fast learning. These are also more efficient and cost-effective compared to extensive upfront planning. They require fewer resources and allow legal tech companies to test ideas before making significant investments. Failed experiments can be discarded early without major sunk costs.

The key takeaway? Embrace quick trials. They enable legal tech firms to stay agile, innovate continuously, and meet evolving user needs.

🔗 In other news

• 💸Lexis is set to acquire Henchman to deliver personalized AI solutions to its customers. Last ILTA I spoke with LexisNexis CPO Jeff Pfeifer on moving past the AI hype.
• 💰Jigsaw and Structure Flow both raise Series A. Jigsaw raised $15m and Structure Flow raised $6m. Wondering what the big deal is - I covered this emerging trend back in 2021.
• 💸 Mitratech acquires HotDocs from CARET.
• 🤖 Maybe we can understand how LLMs think? The Anthropic "Scaling Monosemanticity" paper represents a breakthrough in AI interpretability, demonstrating the ability to identify and manipulate specific conceptual "features" within a large language model like Claude, enabling unprecedented insights into its inner workings and paving the way for potential improvements, including safety.
• 🕧 GenAI’s impact on billing models: will we see more AFAs? The Thomson Reuters Insustitue’s 2024 Generative AI in Professional Services report is worth a read.

Funner finds

• 🎧 Listen to the Acquired episode covering Starbucks with Howard Schultz providing commentary is outstanding
• 🚴 Netflix released Season 2 of its docuseries - Tour de France Unchained - giving a behind-the-scenes look at the hardest mainstream competitive sporting event (my favorite sporting event to watch each year!). It also served as a topic of inspiration for this issue.