How AI Has Turned My Work Into Flow - Part I

Raw Text

Exploring the Benefits of AI: A VC's Take on Leveraging New Technology

Gianfranco Filice March 28, 2023

Lately, I've enjoyed counting how many investors in my network have changed their Twitter and LinkedIn bios from "Web3 investors" to "NLP/AI/GPT-3 investors.” I don’t blame them; following whichever trend is “hot” and cutting edge makes it easy to answer limited partners’ eventual questions on what an early-stage investment strategy is for that space. I’ve been on that side for both Web3 and AI.

However, given that most investment funds operate on a 10-year life, you’d think there’d be less reactionary frameworks to early-stage investing, Web3, AI, or otherwise. Paying top-dollar valuations for what will become cheaper deals when there's no longer a spotlight has always been a unique phenomenon to assess from afar.

That being said, the excitement around AI has been more enjoyable than the excitement around Web3. Why? Largely because I can use AI and experiment with it in my day to day, which I couldn’t do as much with Web3.

Web3 represents a back-end revolution at its core, but since I don't spend my days coding, most innovations I invest in aren't meant for me. In contrast, the innovations made through transformers (the neural networks that power large language models) have made AI authentically useful for my workflows.

AI has turned my arduous affliction of work into the creative elucidation of craft.

My Work Flow

As a venture capitalist, my work is centered around a few key activities:

Writing essays

Assessing market trends

Writing Essays

What this looks like:

As a venture capitalist, part of my role is to prepare and write content that conveys a perspective on a trend or sector. It’s one of the few creative aspects of my job. However, writing can be a painful way to find a sense of artful fulfillment, despite its immense value. Many have taken pen to paper, or fingers to a keyboard, to explain some of the key benefits to writing well . So instead of focusing on that, I’ll explain why writing is so awful and how AI has helped the process be less so.

To start, writing is so f’in hard.

It’s hard because there’s no place to hide. If you write about something you don’t understand, it’s clear to the reader. If you try to gloss over background knowledge, the reader walks away with less value than you intended. If you write something too short, people won’t take it seriously (tweets), but if you write something too long, no one will take the time to read it even if it’s fantastic (War and Peace by Leo Tolstoy).

Writing forces you to say something, unclothed and unshielded. Your thinking receives a spotlight through words, making you look thoughtful or foolish, with almost no chance for an in between.

"Writing is thinking on paper. Anyone who thinks clearly can write clearly, about anything at all”

But on the other side of this pain is a sense of pride and reward. Good writing is craft. And the best part is, most people won’t care how much effort you put in taking a complex topic and making it simple, or turning a computer manual into a poetic cacophony of musings. I say this because it essentially means writers write for the sake of writing. In some respect, it can be freeing to realize that you are only writing for yourself rather than for ego.

And believe me, I’ve learned this first hand. I spent almost forty hours on a piece that I felt was an absolute banger, and much to my surprise, few outside of my friend group even read it.

As a VC, you just put yourself out there for the sake of the process. So yes, the process sucks.

But thanks to GPT-3 and the products that have integrated its API, writing has allowed me to focus more on the thinking and less on the writing, rewriting, editing, head-banging, thoughts of despair, second-guessing, and array of emotions that plague my practice.

Let’s take a look at some of the tools.

AI Tools + Writing

Starting with just an idea in my head rarely turns into beautiful written prose the first, second, or nth time. There’s an incredible amount of iteration and unease through the process. But the art of writing is really the art of just writing: getting ideas onto paper, whether they are refined or not, is the most important aspect. 10% of writing is your first draft and 90% is editing.

I haven’t found a workaround for the aforementioned percentages, which is why I spend as little time as I can on the first draft. So what makes that easier? My tools for writing with AI.

Take Lex for example, an AI meets Google Docs project from Every. To help me craft an outline for this post, I listed out all the ways GPT-3 helps me in my daily workflows, and with a simple command, I can see it spin out an outline for what I should cover. This is my first draft activity. The only thing I should focus on is editing, and AI tools like this make it easy.

Using Lex, I started out with a few prompts and information on what I planned in my head as ideas:

With just these prompts, Lex was able to spit out an outline for me to begin shaping into my own voice:

Is what Lex wrote good? No, in fact, most of it is horse-sh*t. But, it was the first draft, which allowed me to then spend most of my time editing. This, in of itself, has helped me overcome the “cold-start” problem with my own writing.

Let’s go in a little deeper: where AI can help with the editing aspect if that’s the crux of good writing?

Sometimes, I don’t know where a particular paragraph or section is going - I’m too in my head to stitch the pieces together. That’s where I can “chunk” my draft and feed it into ChatGPT to help me summarize and revisit the same ideas but with different words, potentially unlocking a new insight, connection, or thread for which I otherwise would have been unexposed.

For example, I fed two paragraphs that started this section into ChatGPT, and I asked it break down the paragraph into an outline, with the goal of seeing how well my ideas flowed together when pulled apart and isolated:

ChatGPT proved to be an invaluable resource. Its feedback helped me to gain a better understanding of the core messages that I wanted to communicate. More impressively, ChatGPT helped me to identify how each idea in each section connected to my overarching thesis: work flow challenges as a VC were made palpable with the assistance of AI writing tools.

The key thing to remember is that most of what these AI tools produce are outputs I’d never want to sign my name too, but it’s a starting point - a block of marble that I can chisel to an unappreciated masterpiece of my own creation.

Tool: Lex

Link: https://lex.page

Utility: 3.0/10

Drawbacks: There is a UX/UI cost to using this product, and it’s not super simple to use. The outputs lack the same fidelity as ChatGPT, but the benefit is a word processor integrated with GPT-3.

Tool: ChatGPT

Link: https://chat.openai.com/chat

Utility: 9.0 / 10.0

Drawbacks: There are size limitations, you are beholden to how superb your prompt engineering is, and the system can hallucinate if you try to ask for facts and figures.

Assessing Market Trends

One of the most important workflows in assessing market trends is parsing fad from value. While spinning up AI services has never been easier with simple, clean API calls to others’ foundational models, it takes a deeper, more critical look to see what are the eventual upstarts that can capture value in the medium and long-term.

What this looks like:

To make investment decisions, it’s critical to my work to stay on top of emerging technologies. My role is to take those technologies and assess the posited implications of both innovation and dollars flowing into a space, then work backward from a 10-year vision on how business durability can be derived from the former two. Pretty neat, huh?

To help me get there, I read a LOT of content. From paid Substacks that cover a niche corner of the internet to more broad base news articles and industry reports, I spend anywhere from three to four hours a day consuming and synthesizing my coverage themes for existing and potential portfolio companies.

The biggest constraint is both the time and attention dedicated to keeping up with the bleeding edge. Some articles I encounter can be well-written initially, but evolve into a bait-and-switch a third of the way through. The writing becomes less succinct and clear, and the messaging becomes less actionable and insightful.

I find myself wishing I could take away the main points and walk away from the post, without having to slog through the writing.

AI Tools + Assessing Market Trends

Fortunately, with AI, there’s been a way for me to streamline my reading and comprehension without the time burden.

Available as both a standalone website and chrome extension , SumUp takes a link of whatever article you are looking to read and gives you the following:

Quick bullet form summary

Q&A on the article (a new feature that lets you ask your own questions, relying on the text to answer the question in accordance with the frameworks and positions made in the article)

AI-bot’s assessment on the pros and cons of the article

Summary of any calls to actions from the article

Recommendations for similar or related articles

It’s pretty beneficial. For example, I can pop in an article and prime myself for what the main ideas are, and determine if it’s worth going deeper into the details and context.

As good practice, I try to add my favorite pieces to a private repository, but rarely, if ever, do I need to revisit a piece in its entirety. Instead, I can send myself an email in the future via Superhuman and revisit the SumUp summary to retrace some of the learnings and insights I had walked away with initially. I can feel more confident that I won’t forget nor need to revisit the same text to get the insights that made me appreciate the piece initially.

For saving time and accelerating my learning and retention, SumUp’s capabilities have impacted my work to be more about experiencing discovery and exploration rather than frustration and fatigue.

But that’s not the only tool. Sometimes you wish there was a way someone could take a dense, technical paper and break it down for you as if you were five years old. With all the demand for attention, you’d think more people would try to write in a more accommodating manner, but I digress.

As a Web3 investor, I am often tasked with reading a whitepaper for a project, which outlines its technical details and goals. Depending on if it is a consensus layer or decentralized application layer, the information and implications of various design choices can play an immense role in the return profile of an investment. The whitepaper often is the crux to these questions.

But they are written like academic papers. And I hate reading academic papers.

Thanks to ChatPDF, I can turn any PDF sub-50 pages into a question-and-answer interface. And I can prompt it to describe concepts like I am five 🥲

I probably don’t need to outline how powerful this is, but with just a few questions, I can get to the root of my main questions without having to parse, search, or synthesize any of the information.

For example, I’m doing research on Arweave, a blockchain-based storage platform that gives people the ability to store data permanently for free. Arweave’s vision is to create a permanent archive of humanity’s knowledge. The whitepaper is a brief, 22-page prose that’s pretty easy to navigate. So whether I am getting exposed to a whitepaper for the first time, or just looking for a Q&A-based refresher, ChatPDF has been godsend.

The output is simple, and starts out with a high level summary of Arweave:

It then provides me with some context-based example questions:

Naturally, I read the questions above and wanted to learn more about this blockweave technology, so I prompted the following:

Within a few seconds, I received the following response from the interface:

You can see how this tool can be leveraged for cutting through dense essays quickly or even act as a study guide. This tool has accelerated my ability to get through content and achieve better retention, given that I am “interacting” with the text through Q&A, just as I would do with an electronic tutor.

Tool: SumUp

Link - https://sumup.page/

Utility: 7.0 / 10.0

Drawbacks: Will not synthesize any article that surpasses 4,000 tokens (a token is 3/4th of a word)

Tool: ChatPDF

Link: https://www.chatpdf.com/

Utility: 8.5 / 10.0

Drawbacks: The Q&A needs to be contextualized to the PDF, though still can act like ChatGPT if you prompt it to give answers outside the subject matter of the PDF.

Summary

AI has give me a step-function improvement my work as a VC. With the help of AI tools like Lex and ChatGPT, I can focus more on the thinking and less on the writing, editing, and rewriting. Streamlining my writing process, AI has allowed me to focus on iterating and experimenting with my ideas more efficiently.

Additionally, tools like SumUp and ChatPDF have helped me to synthesize and comprehend essays and papers faster than ever, accelerating my learning and retention.

AI as it stands today is imperfect, but it has given me permission to focus on thoughtful intentionality and craft-oriented approach to my work.

Special thanks to Zoe Enright, Sam Wheeler, and Julian from Apeswap for their commentary, guidance, and feedback on this post.

Single Line Text

Exploring the Benefits of AI: A VC's Take on Leveraging New Technology. Gianfranco Filice March 28, 2023. Lately, I've enjoyed counting how many investors in my network have changed their Twitter and LinkedIn bios from "Web3 investors" to "NLP/AI/GPT-3 investors.” I don’t blame them; following whichever trend is “hot” and cutting edge makes it easy to answer limited partners’ eventual questions on what an early-stage investment strategy is for that space. I’ve been on that side for both Web3 and AI. However, given that most investment funds operate on a 10-year life, you’d think there’d be less reactionary frameworks to early-stage investing, Web3, AI, or otherwise. Paying top-dollar valuations for what will become cheaper deals when there's no longer a spotlight has always been a unique phenomenon to assess from afar. That being said, the excitement around AI has been more enjoyable than the excitement around Web3. Why? Largely because I can use AI and experiment with it in my day to day, which I couldn’t do as much with Web3. Web3 represents a back-end revolution at its core, but since I don't spend my days coding, most innovations I invest in aren't meant for me. In contrast, the innovations made through transformers (the neural networks that power large language models) have made AI authentically useful for my workflows. AI has turned my arduous affliction of work into the creative elucidation of craft. My Work Flow. As a venture capitalist, my work is centered around a few key activities: Writing essays. Assessing market trends. Writing Essays. What this looks like: As a venture capitalist, part of my role is to prepare and write content that conveys a perspective on a trend or sector. It’s one of the few creative aspects of my job. However, writing can be a painful way to find a sense of artful fulfillment, despite its immense value. Many have taken pen to paper, or fingers to a keyboard, to explain some of the key benefits to writing well . So instead of focusing on that, I’ll explain why writing is so awful and how AI has helped the process be less so. To start, writing is so f’in hard. It’s hard because there’s no place to hide. If you write about something you don’t understand, it’s clear to the reader. If you try to gloss over background knowledge, the reader walks away with less value than you intended. If you write something too short, people won’t take it seriously (tweets), but if you write something too long, no one will take the time to read it even if it’s fantastic (War and Peace by Leo Tolstoy). Writing forces you to say something, unclothed and unshielded. Your thinking receives a spotlight through words, making you look thoughtful or foolish, with almost no chance for an in between. "Writing is thinking on paper. Anyone who thinks clearly can write clearly, about anything at all” But on the other side of this pain is a sense of pride and reward. Good writing is craft. And the best part is, most people won’t care how much effort you put in taking a complex topic and making it simple, or turning a computer manual into a poetic cacophony of musings. I say this because it essentially means writers write for the sake of writing. In some respect, it can be freeing to realize that you are only writing for yourself rather than for ego. And believe me, I’ve learned this first hand. I spent almost forty hours on a piece that I felt was an absolute banger, and much to my surprise, few outside of my friend group even read it. As a VC, you just put yourself out there for the sake of the process. So yes, the process sucks. But thanks to GPT-3 and the products that have integrated its API, writing has allowed me to focus more on the thinking and less on the writing, rewriting, editing, head-banging, thoughts of despair, second-guessing, and array of emotions that plague my practice. Let’s take a look at some of the tools. AI Tools + Writing. Starting with just an idea in my head rarely turns into beautiful written prose the first, second, or nth time. There’s an incredible amount of iteration and unease through the process. But the art of writing is really the art of just writing: getting ideas onto paper, whether they are refined or not, is the most important aspect. 10% of writing is your first draft and 90% is editing. I haven’t found a workaround for the aforementioned percentages, which is why I spend as little time as I can on the first draft. So what makes that easier? My tools for writing with AI. Take Lex for example, an AI meets Google Docs project from Every. To help me craft an outline for this post, I listed out all the ways GPT-3 helps me in my daily workflows, and with a simple command, I can see it spin out an outline for what I should cover. This is my first draft activity. The only thing I should focus on is editing, and AI tools like this make it easy. Using Lex, I started out with a few prompts and information on what I planned in my head as ideas: With just these prompts, Lex was able to spit out an outline for me to begin shaping into my own voice: Is what Lex wrote good? No, in fact, most of it is horse-sh*t. But, it was the first draft, which allowed me to then spend most of my time editing. This, in of itself, has helped me overcome the “cold-start” problem with my own writing. Let’s go in a little deeper: where AI can help with the editing aspect if that’s the crux of good writing? Sometimes, I don’t know where a particular paragraph or section is going - I’m too in my head to stitch the pieces together. That’s where I can “chunk” my draft and feed it into ChatGPT to help me summarize and revisit the same ideas but with different words, potentially unlocking a new insight, connection, or thread for which I otherwise would have been unexposed. For example, I fed two paragraphs that started this section into ChatGPT, and I asked it break down the paragraph into an outline, with the goal of seeing how well my ideas flowed together when pulled apart and isolated: ChatGPT proved to be an invaluable resource. Its feedback helped me to gain a better understanding of the core messages that I wanted to communicate. More impressively, ChatGPT helped me to identify how each idea in each section connected to my overarching thesis: work flow challenges as a VC were made palpable with the assistance of AI writing tools. The key thing to remember is that most of what these AI tools produce are outputs I’d never want to sign my name too, but it’s a starting point - a block of marble that I can chisel to an unappreciated masterpiece of my own creation. Tool: Lex. Link: https://lex.page. Utility: 3.0/10. Drawbacks: There is a UX/UI cost to using this product, and it’s not super simple to use. The outputs lack the same fidelity as ChatGPT, but the benefit is a word processor integrated with GPT-3. Tool: ChatGPT. Link: https://chat.openai.com/chat. Utility: 9.0 / 10.0. Drawbacks: There are size limitations, you are beholden to how superb your prompt engineering is, and the system can hallucinate if you try to ask for facts and figures. Assessing Market Trends. One of the most important workflows in assessing market trends is parsing fad from value. While spinning up AI services has never been easier with simple, clean API calls to others’ foundational models, it takes a deeper, more critical look to see what are the eventual upstarts that can capture value in the medium and long-term. What this looks like: To make investment decisions, it’s critical to my work to stay on top of emerging technologies. My role is to take those technologies and assess the posited implications of both innovation and dollars flowing into a space, then work backward from a 10-year vision on how business durability can be derived from the former two. Pretty neat, huh? To help me get there, I read a LOT of content. From paid Substacks that cover a niche corner of the internet to more broad base news articles and industry reports, I spend anywhere from three to four hours a day consuming and synthesizing my coverage themes for existing and potential portfolio companies. The biggest constraint is both the time and attention dedicated to keeping up with the bleeding edge. Some articles I encounter can be well-written initially, but evolve into a bait-and-switch a third of the way through. The writing becomes less succinct and clear, and the messaging becomes less actionable and insightful. I find myself wishing I could take away the main points and walk away from the post, without having to slog through the writing. AI Tools + Assessing Market Trends. Fortunately, with AI, there’s been a way for me to streamline my reading and comprehension without the time burden. Available as both a standalone website and chrome extension , SumUp takes a link of whatever article you are looking to read and gives you the following: Quick bullet form summary. Q&A on the article (a new feature that lets you ask your own questions, relying on the text to answer the question in accordance with the frameworks and positions made in the article) AI-bot’s assessment on the pros and cons of the article. Summary of any calls to actions from the article. Recommendations for similar or related articles. It’s pretty beneficial. For example, I can pop in an article and prime myself for what the main ideas are, and determine if it’s worth going deeper into the details and context. As good practice, I try to add my favorite pieces to a private repository, but rarely, if ever, do I need to revisit a piece in its entirety. Instead, I can send myself an email in the future via Superhuman and revisit the SumUp summary to retrace some of the learnings and insights I had walked away with initially. I can feel more confident that I won’t forget nor need to revisit the same text to get the insights that made me appreciate the piece initially. For saving time and accelerating my learning and retention, SumUp’s capabilities have impacted my work to be more about experiencing discovery and exploration rather than frustration and fatigue. But that’s not the only tool. Sometimes you wish there was a way someone could take a dense, technical paper and break it down for you as if you were five years old. With all the demand for attention, you’d think more people would try to write in a more accommodating manner, but I digress. As a Web3 investor, I am often tasked with reading a whitepaper for a project, which outlines its technical details and goals. Depending on if it is a consensus layer or decentralized application layer, the information and implications of various design choices can play an immense role in the return profile of an investment. The whitepaper often is the crux to these questions. But they are written like academic papers. And I hate reading academic papers. Thanks to ChatPDF, I can turn any PDF sub-50 pages into a question-and-answer interface. And I can prompt it to describe concepts like I am five 🥲. I probably don’t need to outline how powerful this is, but with just a few questions, I can get to the root of my main questions without having to parse, search, or synthesize any of the information. For example, I’m doing research on Arweave, a blockchain-based storage platform that gives people the ability to store data permanently for free. Arweave’s vision is to create a permanent archive of humanity’s knowledge. The whitepaper is a brief, 22-page prose that’s pretty easy to navigate. So whether I am getting exposed to a whitepaper for the first time, or just looking for a Q&A-based refresher, ChatPDF has been godsend. The output is simple, and starts out with a high level summary of Arweave: It then provides me with some context-based example questions: Naturally, I read the questions above and wanted to learn more about this blockweave technology, so I prompted the following: Within a few seconds, I received the following response from the interface: You can see how this tool can be leveraged for cutting through dense essays quickly or even act as a study guide. This tool has accelerated my ability to get through content and achieve better retention, given that I am “interacting” with the text through Q&A, just as I would do with an electronic tutor. Tool: SumUp. Link - https://sumup.page/ Utility: 7.0 / 10.0. Drawbacks: Will not synthesize any article that surpasses 4,000 tokens (a token is 3/4th of a word) Tool: ChatPDF. Link: https://www.chatpdf.com/ Utility: 8.5 / 10.0. Drawbacks: The Q&A needs to be contextualized to the PDF, though still can act like ChatGPT if you prompt it to give answers outside the subject matter of the PDF. Summary. AI has give me a step-function improvement my work as a VC. With the help of AI tools like Lex and ChatGPT, I can focus more on the thinking and less on the writing, editing, and rewriting. Streamlining my writing process, AI has allowed me to focus on iterating and experimenting with my ideas more efficiently. Additionally, tools like SumUp and ChatPDF have helped me to synthesize and comprehend essays and papers faster than ever, accelerating my learning and retention. AI as it stands today is imperfect, but it has given me permission to focus on thoughtful intentionality and craft-oriented approach to my work. Special thanks to Zoe Enright, Sam Wheeler, and Julian from Apeswap for their commentary, guidance, and feedback on this post.