Gravity Stack Goes All In On Generative AI

Gravity Stack, the standalone legal tech and consulting group connected to Reed Smith, is now going all in on generative AI, Managing Director, Bryon Bratcher, told this site.

The group began in 2018 and developed several legal tech tools for sale to clients, which it still maintains. Supporting those projects involved consulting work and this element has steadily grown. Now, the emphasis is very much on tech advisory work in relation to AI.

As Bratcher told Artificial Lawyer: ‘We see this as a paradigm shift. AI is doing now what we hoped AI could do before.’

Bryon Bratcher

That is to say, generative AI has delivered (albeit with some persistent challenges with accuracy and hallucinations) the kind of productivity and adaptability that many lawyers hoped the first wave of NLP / ML legal AI tools would be able to achieve out of the box.

Now that we have this technology on tap, Gravity Stack is embracing the opportunity to help clients with all their legal genAI needs.

‘GenAI is a game-changer and we need to integrate it within all we do. For example, our own team uses genAI tools, we use Perplexity AI instead of Google, and we are also testing out OpenAI’s new SearchGPT capability,’ he says.

While for their client work they are using Vellum AI, which is a tool that helps assess prompts and genAI models.

Bratcher notes that this evaluation work is all part of what they call their ‘AI Lab’, where they will study a client’s needs, explore the use cases, look at what sort of tools and solutions will work best for them and then advise on implementation.

‘We do the whole thing for them. We test in our labs, we look at open source and commercial LLMs, we analyse costs and security. It all goes into the funnel,’ says Bratcher.

Clients can even make use of Gravity Stack’s AI Lab to test out applications on their own.

‘We can give them a test area, or we can co-test together,’ he adds. ‘And there is dummy info they can test on if they’d like to use that.’

Plus, Reed Smith’s own tech team also makes use of the AI Lab. (And while we are on the subject of the mothership law firm, Bratcher mentions that Reed Smith is currently using Harvey. When asked what impact it has had on the firm, he replies that ‘it’s early days’.)

Bratcher explains that they recommend both legal tech companies which have their own genAI product, and also that clients can use an LLM suited to their needs, with some level of ‘wrapper’ around it for their use case.

Also, sometimes a client will have a legacy platform that they want to stick with, but Gravity Stack can come in and help see how that can be extended, such as developing a custom LLM solution for them. Or as Bratcher puts it ‘a layer to drop on top’ of what’s there.

In terms of genAI companies that they use most, it’s OpenAI and Anthropic, with sometimes Gemini. On the open source side it’s Llama.

From what this site has seen, this selection appears to be the most frequent among law firms and tech vendors. It also suggests that although the LLM market is evolving fast OpenAI and Anthropic are clearly still the providers of choice….for now.

All well and good. What about other things, such as accuracy?

Bratcher is not as focused on this as some other legal tech teams and says: ‘It’s important to look at, but we don’t see it as critical. The work product of a lawyer is the main factor and that will be reviewed.’

I.e. AI will get you to X point in a legal task, whether it’s very accurate or not, after that the lawyer’s overall work will be reviewed by more senior lawyers so the focus on immediate accuracy of the genAI tool is not so vital.

Of course, one could argue that not having to spend extra time reviewing the lawyer’s work product because the genAI outputs have shaped it so well would be a good thing. Or even better, not having to spend hardly any time at all reviewing it….because the genAI that delivered a large part of the work product was so on the ball.

But, moving on. Does he see a change in business models for law firms? I.e. efficiency leads to greater speed of delivery, such speed means selling time is pointless, so you need a new business model.

Bratcher notes that on some areas where clients specifically ask about efficiency, such as with M&A due diligence, speed does come into it. But, generally, he’s not sensing any major changes within law firms across the board due to genAI.

Although, one could then say, if the business model is staying the same, what impact is genAI having? The answer to that could be, as Bratcher noted above: it’s still early days.

So, going back to the beginning, Bratcher underlines that where they are seeing real change is in the level of demand for helping clients to implement genAI applications. This has its own challenges, he adds, such as sometimes they are just not ready for it.

‘You can’t go from 0 to 100 in one go, you need other layers before you use genAI. For example, some clients want to use genAI on their own knowledge base, but they have not structured their data, perhaps there is not even an index for it,’ he observes.

We then explored one emerging point: that as LLMs get better and have larger and larger context windows, the need to really structure data bases could reduce. I.e. if an LLM can turn 1000s of pages of unstructured data into something very structured through just a series of prompts will we need KM systems in the future?

We’ll need some form of storage system for all of a law firm’s or client’s documents, but will that information need to be finely tagged, super-organised and structured as we try to do today…?

We are not there yet, and clearly there are major benefits to detailed data architecture today when it comes to genAI, as it improves accuracy. But will that always be the case? Will the well-known KM systems out there be redundant in 10 years when LLMs can handle 100,000s of documents at once – and with great accuracy? Are KM systems for law going to go the way of physical law libraries?

Bratcher notes that LLMs are clearly getting better very fast, and the chips that support them are becoming more powerful very rapidly. But, we leave the question open and one to ponder on.

Overall then, lots of work for Gravity Stack to do and so their focus on genAI makes total sense. This area is only going to grow and grow. Bratcher is going to be very busy.