Roy Geva Glasberg (right) and Lee Moser.

Ethics of LLMs in investments: dos and don’ts

Roy Geva Glasberg and Lee Moser, Managing Partners of AnD Ventures, discuss the rise of large language models (LLMs) in the investment world - the benefits, limitations, and ethical considerations

In recent years, large language models (LLMs), programs that use AI to process language and generate new content, have exploded in popularity. They have revolutionized how people interact and gain knowledge, and have allowed information to become more accessible than at any point in history. At the same time, however, they have introduced important ethical dilemmas as they have become more frequently utilized within the investment world.
It is important to remember that as language models, LLMs are not meant to replace human functions, but to instead provide basic frameworks from which people can access information for themselves. At our VC, we understand that their main purpose is to facilitate objective processes, and not to replace critical human interactions or to replicate the “human touch.” It is a widespread misunderstanding that these technologies are able to produce entirely accurate information on all topics, including those related to investment. This is not true, and as such, we actively encourage our team, as well as all others operating within the investment world, to take heed of certain considerations when using LLMs to produce their work.
1 View gallery
Roy Geva Glasberg and Lee Moser
Roy Geva Glasberg and Lee Moser
Roy Geva Glasberg (right) and Lee Moser.
(Eyal Toueg)
LLMs can be useful to model basic market research and competitive analysis, given that historical metrics are widely available on the internet. Likewise, they can offer generic trends – projections, P&L, and otherwise – for a given company, given that publicly sourced information is generally more preferable to on-a-whim human predictions.
We know this firsthand. Our analysts frequently utilize LLMs to conduct competitive analysis. These technologies have allowed us to find companies that fit the same genre as our prospective startups – that offer similar products – which is difficult to discover via a simple Google search. It has helped our market research efforts considerably, allowing us to expedite the process in ways not possible before.
LLMs have the added benefit of doing the reverse as well; along with sourcing information from links, companies can also use these technologies to authenticate information and find the internet links associated with them. In its early stages, one of our prospective startups offered our team information about its traction and competitive analysis. Our analysts used LLMs to source back their data and verify it, allowing us to move along in the process and eventually bring them on board as one of our portfolio companies. LLMs can assemble basic background information on founders and industry experts, and, owing to their advanced technological capabilities, wordsmith and articulate this information in a more cohesive and engaging manner than previously possible.

LLMs also serve as an important middle ground when modulating between different softwares, or when accounting for holes and errors present when attempting to source or produce information. Members of the Team have repeatedly employed LLMs to account for discrepancies when transferring information from one program to another, such as from to Google Excel, in ways that such forms of software cannot account for on their own. As well, when cross-referencing information between said platforms, LLMs have proven useful in troubleshooting and accounting for potential gaps apparent across them. In this sense, LLMs inhabit a vital “middle ground,” bridging the divide between start and finish when human funcion is unable to do so.
It’s important to note that there are areas where these technologies should not be used. Any aspect of the investment process that requires full accuracy or original human perspectives should not be conducted through LLMs alone. Customer reviews, reference calls, and any other element that fits this description should be done primarily through human interaction and individual research. As well, to protect the privacy of individual investors, their personal information should not be inserted into or searched for in any LLM program.
While these technologies may not be a silver bullet, they do have their purpose. As we see for ourselves, they can expedite the investment process in ways unseen before and provide a myriad of benefits for any company navigating the investment process.
Roy Geva Glasberg and Lee Moser are Managing Partners of AnD Ventures