Opinion #Investment
Read time: 03'23''
31 August 2023
Schrodinger’s Generative AI: both game-changing and overhyped at the same time

Schrodinger’s Generative AI: both game-changing and overhyped at the same time

Generative AI is the hot new thing in VC and just about the only technology sector that is still seeing plenty of FOMO from companies investing in startups. That’s no surprise; seeing a large language model (LLM) create an image or write an essay for the first time is almost magical.

However, the real game-changing potential lies less in the ability to generate new content and more in the fact that in LLMs we now have the ability to understand and analyse content in a much more accurate, generalisable and scalable way than has ever been possible, which means many tasks that could only be done by humans can cost effectively be automated.

Over the next few years, we’ll see more and more processes automated using LLMs which will likely improve efficiency across a whole range of verticals, particularly those like insurance, law and medicine where, due to the volume of unstructured data, automation has been difficult. The potential to boost productivity in huge sectors such as these justifies investors’ excitement.

That said, the ease with which LLMs can be leveraged to build simple automations, often in minutes or hours, has led to lots of hype with an explosion of startups leveraging LLMs to deliver solutions such as copywriting (JasperCopy.aiRytr), chatbots (Character.aiHeydayChai) or sales assistants (OutplayRegie.aiExceed.ai). Here, the ease of building with LLMs, often through providers such as OpenAI or Anthropic, means that products can be built and launched very quickly, which will lead to a plethora of startups and lots of noise. We’re already seeing that in some spaces like sales assistants, where there are a lot of companies building similar solutions and it’s difficult to choose between them.

There are three problems with this wave of startups:

  1. While the analysis and automation provided by LLMs might be key functionality, it likely won’t be sufficient on its own without workflow, UX, integrations etc. and building that functionality will be harder than integrating output from someone else’s LLM. This means that incumbents, who have already built those things, will be better placed than new entrants.
  2. Just incorporating an LLM does nothing to provide the defensibility that is needed to build a venture scale business. If you can quickly automate a process using an off the shelf LLM, so can your competitors.
  3. The technology is developing so rapidly that there is a risk of building capabilities that are quickly superseded or rendered obsolete as technology improves.

This means that although generative AI will likely significantly impact the way many jobs are done with big productivity gains, many of the first-generation generative AI startups will fail. The challenge for VCs is to find companies that are leveraging generative AI in a way that allows them to build a defensible business.

We think that the answer to this challenge is to focus on the user and their pain point. The capabilities of LLMs mean that many previously intractable problems can be solved and entrepreneurs who deeply understand these problems are best placed to see the opportunities. While incumbents will add generative AI to their products, it’s hard for them to re-architect them with intelligence at the core, which can give startups building from the ground up an advantage, so long as they are building products that solve a problem for customers rather than looking for somewhere to use exciting new technology.

These winners will emerge in industries that have large volumes of text and image data that is too variable to automate with previous technologies, meaning that lots of poorly paid staff have to manually work through repetitive processes. This is expensive, slow and error prone and I’d expect to see a new wave of tools that can leverage the power of LLMs to massively boost productivity and make use of data that was previously unusable.

In healthcare, we could see the dawn of scalable proactive care with the insight that was previously trapped in doctors’ notes used to identify patients who would benefit from care before their health suffers, while insurance could move towards automated claim settlement, allowing much faster payments to those who have suffered loss. Law is already being disrupted by firms like our portfolio company Robin.ai, which reduces the time for contact review by 85%. However, use cases like these will require deep insight into the current ways of working and the pain points of users rather than just adding a thin layer of AI to existing approaches.

The rise of generative AI is a generational shift with innovative new companies being built and many previously invincible incumbents becoming vulnerable, but it will also see a raft of startups come and go with the hype. We’re looking to back to entrepreneurs who are leveraging these technologies to solve real customer pain points and build huge businesses and we think there has never been a better time to build solutions to some of the hardest problems people are facing.

Luke Smith is Investment Partner at Forward Partners.