ChatGPT is an advanced model that marks an inflection point in artificial intelligence and machine learning capabilities. Given any text prompt, it can generate responses varying from a paragraph to an article or even computer code! Created by OpenAI and launched in November 2022, it is a generative model that can create original content rather than simply analyzing or summarizing existing text. This enables ChatGPT to have a more natural and flexible conversational style. In 2019, Microsoft invested $1bn in OpenAI. There’s ongoing talks that Microsoft is getting ready to put another $10bn, giving OpenAI a valuation of $29bn.
ChatGPT is incredibly easy to use - it takes less than a minute to sign up and has a no-code interface accessible to anyone with an internet connection. ChatGPT has found many use cases including writing ads, Twitter threads, inventing product names, writing blogs and many more. Today, we look at how it can be applied by a professional investor.
As a VC or Angel investor, there are many manual, repetitive and structured tasks that take up a large chunk of time each day. Some examples include: data extraction and processing, competitor or similarity analysis, identifying experts in a field, reaching out to founders and writing rejection emails. The great news is these are all tasks that can be automated using ChatGPT. It might take a few attempts to fine tune the prompts but the below can be used as building blocks for your own purposes.
Below we tested out ChatGPT’s capabilities to extract information from the website of a supply chain start-up (Project44) and group the information in a logical manner.
Prompt: summarize the website of Project44 into a tabular format comprising company name, location, industry and description.
Going a step further, we asked ChatGPT to identify companies similar to Project44. The results are mostly spot on. However, a discerning reader will note that for #3, ChatGPT refers to a “cold chain industry” which doesn’t make any sense. We talk about some of ChatGPT’s limits below but for now this type of misinformation is the by-product of statistical methods used by ChatGPT to predict the next sequence of words in a given context.
Prompt: name 5 companies most similar to Project44
ChatGPT can answer this on a numeric scale and share the rationale behind the number.
Prompt: How similar are the companies Project44 and FourKites on a scale of 0 = no similarity and 10 = equal
It is important to note that the above examples work with specific company names that are unique. If there are similar sounding companies, it is better to identify them using their website or another unique identifier.
Prompt: Who are the experts in supply chain resilience I should consult for investment advice in that space?
While the response was directionally correct, ChatGPT leaned towards identifying supply chain experts (e.g. #1 and #2) rather than those specifically in supply chain resilience (e.g. #4).
With the ability to generate unique responses based on slight tweaks to the prompt, this is a powerful use case to create tailored emails en-masse.
Prompt: Draft a cold email to the founder of Project44, introduce myself as a partner at a VC firm and mention that I'm interested in investing in their next round of funding because I am excited about the potential of supply chain analytics and their value proposition.
As above, concrete feedback can be structured into an effective email that’s personalized for each response.
Prompt: Draft a rejection email to the founder of a start up with the reasons for rejection being no clear market, complex go to market strategy and lack of clarity in value proposition
ChatGPT can be used to generate written text on a variety of media, such as news articles, stories and product descriptions. Below we look at compiling research outlines, creating hooks for blogs and writing the blog itself!
ChatGPT can provide comprehensive outlines that can be further fleshed out through targeted prompts.
Prompt: Give me the outline for a research article on why today’s supply chains are not resilient
Multiple hooks can be generated if the first one doesn’t quite hit the mark.
Prompt: Give me a hook for a blog with this opening paragraph:[....]
When combined with the research outline above, it can be a powerful combination.
Prompt: Write a blog on the importance of resilient supply chains
It can be used to process large amounts of text to complete investor sentiment analysis, provide company and industry overviews and analyze the impact of macro factors on an industry.
Prompt: Classify the sentiment between 0 = very bad to 10 = very good in the following article: [paste the article here]
Prompt: what challenges are facing global supply chains today?
Prompt: How are increasing interest rates affecting supply chains?
As above, it is a quick reference point to get a high level understanding on a topic or book.
Prompt: Explain internal rate of return like I'm a 5 year old
Prompt: Summarize the book you can be a “Stock Market Genius”
Prompt: Tell me a joke about investing in supply chains
Prompt: Give me a limerick on supply chains
In addition to using ChatGPT for individual prompts, it can be taken a step further and integrated directly into existing applications and enriched with proprietary datasets. Through integration with existing apps, users can send prompts and receive responses from ChatGPT without leaving that app. Likewise, the publicly available information used by ChatGPT can be enriched using proprietary datasets to harness the intelligence of ChatGPT across relevant data.
In order to scale the workflows, ChatGPT needs to integrate with other applications (e.g. Slack, Google Sheets). ChatGPT conveniently provides the API key required to connect the pre-trained models with these applications. The API key can be accessed from here.
Unfortunately, APIs are often not standardised and require a bit of coding to get them connected. However, there are some other automation tools that can build these connectors and make them more accessible. One such tool is Zapier that has integrations with more than 5k apps, including Slack, Gmail and Sheets. As the below example shows, using Zapier, information can be sent between OpenAI and Slack automatically without any code.
Through the use of APIs, ChatGPT can connect to proprietary databases and thereby extend the knowledge of the pre-trained OpenAI models. By doing this, we can leverage the intelligence of ChatGPT to tap into not just the publicly available information available till June 2021 but also any proprietary data, allowing investment team members to naturally interact with the combined dataset via the chatbot.
Prior to using ChatGPT, users are presented with a warning that the information might be false or misleading. Misinformation arises due to the processes by which language model powering ChatGPT learns to represent language. At its core is a massive text dataset that is used to predict what the next sequence of words will be in a given context. ChatGPT looks for statistical patterns and relationships between words to generate coherent text. As a result, while the constructed paragraphs look legitimate, the response may be factually incorrect.
The data powering ChatGPT is current as of June 2021. Accordingly, asking it to share the latest news on a recent start-up or make a prediction on time sensitive data will not yield many useful results.
Each time a prompt is sent to ChatGPT, it goes straight to the servers of OpenAI for processing before the results are shared with the user. Given data is still being collected for research purposes, it is likely that this data will be stored for future reference.
ChatGPT has exploded in popularity over the last 2-3 months and as a result, its servers are often at capacity. Users are required to wait till the load drops before logging on. That said, this is more a temporary issue that will be solved in the near future.
Implementing ChatGPT has the potential to significantly improve the productivity of a VC or Angel Investor. We saw how it can be used to automate common tasks, generate content, supplement research efforts, accelerate learning and integrate with existing apps and datasets. The biggest risk with using ChatGPT is misinformation arising from how the language model processes data - a risk that will need to be monitored closely as the barriers to creating content are eliminated and the volume of available content rises exponentially.