TLDR; Generative AI is the new(-ish) hot fad in VC (read: similar to the rush into Crypto / Web3). But, the technological advances are substantial enough that this fad is here to stay and offers a substantive risk to big tech.
So what is it?
Generative AI has been the buzzword of 2022 (whether we knew it or not).
But what makes generative AI so different from the AI we’ve been hearing about for years?
Generative AI is coined simply from the idea that it is able to create something new whereas prior AI technology was used primarily for analysis. This is a massive breakthrough.
To take a step back, AI is short for artificial intelligence. Although it may feel that AI is a new phenomenon, AI has been an area of interest since the 1950s (Harvard).
Over the last decade, AI has continued to expand with the advancement of technology. Within AI, the sub-area machine learning (ML) has exploded as more computing power has become available (for context, the average iPhone is exponentially more powerful than the computers NASA used for the Apollo 11 mission in 1969). Machine learning is the process by which algorithms learn (i.e. are trained) on data, allowing them to ultimately make decisions (which is how it connects back to AI).
Ultimately, the goal of AI is to recreate human intelligence through machines. It is debatable whether that goal is truly achievable but generative AI is a big step.
Why now?
Computing and processing power has improved drastically over the last decade, and as more variables are able to be processed, AI has improved (Crunchbase).
These models are trained by analyzing vast amounts of information from the internet (which in and of itself is a limitation given not everything on the internet is true to begin with). But, as the models are able to obtain and process more data, they can become more accuarate and can be further refined over time (NYT Code Red).
Wait... but show me the $$
VC’s have been pouring money into anything and everything that gets close to being considered “generative” (for context via Dealroom, GenAI startups raised $1.5B in 2022 as compared to $909M in 2021 and $213M in 2020; 600% increase from 2020!! 🤯)
For context, the largest investment to date was a $1B investment in OpenAI in 2019 led by Microsoft and Khosla Ventures (Dealroom).
Some of the biggest rounds of funding this year include:
Inflection AI - $225M financing with the goal of building products to reduce friction between human-machine interactions (TechCrunch). It’s worth noting the leadership team of Inflection is stacked - the three co-founders are Mustafa Suleyman, co-founder of DeepMind, Reid Hoffman, co-founder of LinkedIn, and Karén Simonyan, a leader in deep learning research.
Cohere - $125M series B financing to expand access to NLP via easy-to-deploy large language models that can be used across systems including enterprise services. Example use cases include scraping Reddit to understand and produce recommendations based on company sentiment or even writing blog posts or creating creative ads. Much of the senior leadership team came from Google Brain. Cohere also has a non-profit research lab, Cohere for AI aimed at solving complex ML problems as well as expanding the accessibility of these research areas.
Jasper AI - $125 series A financing to build out their AI Content Generator which can generate custom content across social media, advertising, blogs, etc. simply through a Chrome extension. Co-Founders Dave Rogenmoser and Chris Hull worked together previously to found Proof prior to moving on to Jasper. Jasper’s $1B valuation seems high given the bearish market conditions and that the funding round was completed in Oct., but Jasper brought in $35M in revenue in their first year, so they have an established track record (WSJ).
And one more for kicks, they didn’t make the largest round list but I’m personally pretty excited about what the company has to offer.
Mem - $23.5M financing round led by OpenAI to build a product that produces and organizes notes using AI (TechCrunch). What I like about Mem is it differentiates by allowing users to take notes across mediums, from SMS and other messaging apps to their native app to calendar integrations. Over time, they plan to include notes for long-form documents, e.g. summarizing long research papers, and surfacing those notes to users when relevant to save time and improve productivity (internal file disorganization is currently estimated to cost employers $3,900 per employee per year!).
It would be remiss to glaze over the big tech players in the field as well, Microsoft, Google, Meta, and Amazon have all noted the importance of this area and have been doubling down on research and investment in generative AI.
Microsoft is a lead investor in OpenAI, investing $1B in 2019
Google has a subsidiary DeepMind Technologies that just recently released a new, improved giant language model called Chinchilla
Meta also has large language models available for download and research called OPT-175B
Amazon has released a generative AI model for music, DeepComposer
Even smaller behemoths like Salesforce are focusing resources, ultimately with the goal of improving worker productivity (WSJ).
And an example that almost broke the internet, OpenAI
OpenAI is leading the way in terms of buzzy, user generating AI. First with DALL-E, a tool that allows users to type in image specifics and outputs an image, and next with ChatGPT, a chat-based tool that can do anything from answering a specific question to writing a college essay.
ChatGPT gained ~1M users within 5 days of launching (talk about viral growth!), and after playing around with the tool, it’s understandable why.
For example, typing the following into DALL-E produces the unique images below
3D render of a small pink balloon dog in a light pink room
And the images can get even crazier…
3D render of a small pink balloon dog carrying a rainbow unicorn on its back in a light pink room with a shiba inu dog observing them both
As you can see, the description doesn’t exactly match, but definitely not a bad start.
Or, in ChatGPT,
what is the meaning of life
The meaning of life is a question that has puzzled people for centuries and is a topic that has been explored by philosophers, theologians, and scientists. There is no one answer to this question, as the meaning of life is likely to be different for each individual and may change throughout a person's lifetime. Some people believe that the meaning of life is to achieve happiness and fulfillment, while others believe that it is to help others and make the world a better place. Still, others believe that the meaning of life is simply to exist and experience all that the world has to offer. Ultimately, the meaning of life is a personal and subjective concept that may be influenced by a person's cultural, spiritual, and philosophical beliefs.
write a haiku describing if Santa exists
Santa's existence,
debated Legend or truth,
it's hard to say
But love and joy he brings,
on Christmas day
Definitely try it out for yourself (DALL-E, ChatGPT)!
Whoa, this is cool! But… what’s the catch?
ChatGPT is able to provide nuanced answers to some complicated questions, but the answers can also be wrong.
This is a known issue to the point that OpenAI released the following disclaimer:
ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers.
Trying to decipher what is true and what is not becomes increasingly complex as AI expands, especially given that ChatGPT doesn’t cite sources.
ChatGPT tends to break down in the three following scenarios, although there are likely many more that will be discovered as more people interact with it (FastCompany)
Recent information - questions like, what's today’s weather in Boston?, how many times has McCarthy lost in his attempt to become the new Speaker of the House?, what is Google's revenue? are returned with the same response: I'm sorry, but I am unable to provide current information on <insert question> as my knowledge is based on data that was available up until 2021, and I am not able to browse the web or access current information. Is there anything else I may be able to help you with?
Niche topics - topics that are insufficiently available in the training data won’t be answerable
Bias-free response - although there are filters applied to prevent bias from being displayed for more direct questions, simple changes enable users to bypass these filters and display bias, as this professor identified
What’s next?
Google and other big tech companies are worried about this technology despite having invested in it. Google just last week called a “Code Red” indicating that this is a substantiated threat to their business (among others that we’ll discuss in the not-so-distant future 😉).
And to tide you over until next week, a few of my favorite recent articles.
The irresistible voyeurism of “day in my life” videos – an interesting take on what makes it so difficult to stop watching these Tiktoks (spoiler: watching someone’s seemingly perfect life can be addicting!)
How a Nepo Baby is Born – comprehensive deep dive on nepotism in Hollywood that leaves no actor / actress unscathed.
An A.I. Pioneer on What We Should Really Fear – an interview with Yejin Choi, 2022 MacArthur genius grant recipient, on the misconceptions and fears of AI. Generally, Choi’s view is although AI is great at complex, pattern-based tasks (e.g. Chess), its ultimate weakness is it lacks common sense which often acts as the basis for further logical thinking (e.g. which birds fly).