Zima Red ep 149: Alex Ker - Investor, Coder, Thinker: Our AI Empowered Future
Primer: What inspired Alex Ker to dive deep into AI? How has the sector changed since 2017? What are his thoughts on AI companionship? Let’s find out in this episode of the Zima Red Podcast.
Getting Started In AI
Started in high school
Send a cold email to Josh Tenenbaum, a professor at the Computational Cognitive Sciences lab at MIT
Did data collection research for one of his AI projects
What Made Him Dive Deeper Into AI?
Getting rid of the tedium in our work days:
There are a lot of tasks during work that are not appealing — like document entry or replying to the same type of emails
Think that humans are meant to do more creative things with their lives
Using AI to decode how the brain functions (e.g. intelligence, consciousness)
Change In The Sector Since 2017
A lot of the hype back then was on computer vision
There’s a transition to natural language processing and large language models
His initial motivation to start up P-ai, an incubator that executes AI projects and ideas, is because of the built-up interest in this field
In 2019, there were not enough people working in AI or enough resources to support people who were interested in getting into it
“A lot of the behavior of these models, because they're getting so large, is hard to understand. So it's sort of like the scientist has built up this Frankenstein-ish creature. And you're trying to retroactively understand how the inner functions work.”
- Alex Ker
Early Projects/Ideas People Were Working On
At first, there were computer vision projects
There was a brain-computer interface project where they were using brain signals to decode hand gestures
There was a reinforcement learning project where they were training a simulated drone with Unity, the game engine
The early innings of large language models were people trying to generate jazz music with GPT-2
People were generating websites with GPT
People were trying to see whether GPT-2 can write a basic HTML code for a Shopify storefront
AI’s Impact On The Industry
Most impact will occur in industries that are orthogonal to software or AI
Traditional industries such as agriculture or supply chain/logistics
AI will create a 100x improvement for those industries that have no software or minimal software usage
A lot of people focused on the sexy industries like B2B, SaaS, etc., but this is oversaturated
Wrappers Around GPT Or Novel AI Models?
Wrappers could be as niche or fine-tuned as possible
If companies are using generalized wrappers, it is not interesting
Harvey is an AI for the legal sector. It’s a specialized model that is fine-tuned on their dataset. This is a better play for GPT wrappers in a specific domain
Startups Doing More With Fewer Resources
“With AI, Startups Need To Raise Less Money And Do More With Fewer People”: True Or False?
Some startups will be able to leverage AI to reduce the human labour needed
Those roles could be writing marketing copies, doing SEO, or generating content
You still need humans to prove that type of content and knowhow to prompt engineer the best type of content that fits the company
Engineers should be using code interpreters or GitHub Copilot to help them write code and debug
The cost of starting a new side project or testing out a company idea has lowered
How Does He See It Playing Out?
The behavioural shift will come slowly
Sees AI taking on 3 roles:
A peer/colleague who can do a set of tasks that you delegate to
AI as a teacher/tutor
AI as an intern where you give low-level tasks
These 3 roles will play a huge role in everyone’s lives
Big Tech And AI
AI will become the next AWS
Large language model providers and APIs will be the play for bigger companies
There’s still a lot of opportunity for smaller companies (operating in a specific domain that you are very knowledgeable about, and have a specialized dataset that you know how to fine-tune)
AI Agents
More familiar with BabyAGI than ChatDev
These projects are very demo-oriented and have gone viral on Twitter
These AI agents are very brittle. If one component fails, the errors cascade down the line
Difficult problem for AI agents to recognize what went wrong, backtrack, and redo a task
Humans need to put in a non-trivial amount of work to fix AI agents when they stumble
Tasks That AI Agents Could Do Very Well Today
Helping you debug and generate skeleton code for you to build on
LLMs tend to be very verbose if you use them for marketing content. It requires a lot of human editing afterward
Thoughts On AI As Our Friends
It is a contentious topic
Two sides of the argument
The negative side is that AI companies want user engagement. Hence, they are incentivized to get you to stay on their app, driving everyone towards more isolation
The positive side is that when you don’t have access to social environments (e.g. living in a remote area), these AI agents can help people improve their social skills
Most AI companion users are actually females because of the psychological appeal
The AI girlfriend/companion space is very competitive
There’s an OnlyFans model who created an AI bot, Caryn AI, that functions as a chatbot which she charges on a per-minute basis
When people engage with a creator and the creator does not respond to them, they experience a sense of loneliness from the one-way interaction
In the future, smart businesses will be able to fine-tune chatbots from a creator’s past content and scale the level of interaction with their fans
Is It An Issue If It’s Not The Actual Person?
If the person knows it’s an AI, to begin with, it’s okay
What’s dangerous is people not knowing it’s an AI and they think it’s an actual human
Companies have to be as transparent as possible
Did AI Companions Catch Him By Surprise?
It caught him by surprise
He likes companies that understand the psychological needs of humans
The loneliness market is huge
“The loneliness market is huge. A lot of people are lonely and there's this need for intimacy, for friendship, for relationships and AI companions definitely fill that void for some folks.”
- Alex Ker
AI Use Cases That People Are Not Thinking About
The elderly market
As the baby boomers age, the market for the elderly is going to expand a lot
They have a lot of spending power
AI companionship could be something like an AI son/daughter who would talk to the elderly and listen to their stories. It’s an underrated market
Anything formulaic or tedious (e.g. accounting or consulting) could be replaced by AI
Moats In The AI Era
Moats are going to be network effects and the brand
The degree to which the company dominates user psychology and is ingrained in their habit loop is the best moat
Large language model providers will still have a strong technical moat
If He Had To Launch A Company Today
The Bootstrapped Version
Like media companies in general
You create content and get feedback from your audience, deeply understanding them
The bootstrapped version is an AI media company that launches software products around it
A Billion Dollar Capital
Antiquated industries that need a lot of change in software
Which Sector Got Him Most Excited Recently?
Media businesses and content creation
Thoughts On Brain-Computer Interfaces (BCIs)
He is a bit outdated in the industry despite working at Neurable from 2020 to 2021
Hard to read high-fidelity signals from the brain, especially if it is noninvasive
Apple filed a patent to put EEG sensors into AirPods
Which Takes Off First? Low-Stakes Or High-Stakes Use Cases?
There are companies tackling both
Neuralink and Kernel are going towards the clinical setting or tools for research
Others are going into the consumer biometric space, targeting things that can improve your life right now
A few years ago, people did not know what an Oura ring was. Now, a lot of people have it
Thoughts On Augmented Reality (AR)
AR is more interesting than VR because people want to see reality as it is while still being in the presence of other humans
AR could be used to train people on specific tasks (e.g. how to fix a bike, training to become a machine operator, etc.)
Thoughts On Web3
A lot of entrepreneurs who have started in Web3 have pivoted to AI
Not able to differentiate what is real and a scam in Web3
Think that the technology is very interesting
The whole premise of owning a JPEG struck him as funny
Thoughts On The Metaverse
A more immersive internet where you can merge VR and AR
The Current Situation In Early-Stage AI Investment
People are getting into AI deals that are not so good
Wrapper deals are priced at insane valuations
People are worried that if they do not get these deals, they might miss the next biggest thing
However, he thinks that wrappers are the first innings of the AI revolution
Bullish And Bearish On
Bearish
GPT wrappers
Think that they are transient behaviour that will be phased out or consolidated in the future
Bullish
Anything that distribution or network effects that become part of the customer’s habit
Is He An AGI Doomer Or Optimist?
An Optimist
Best Piece Of Advice He Has Received
Cold DM and cold outreach to more people, because you will be surprised how accessible people are
Start sharing your thoughts and ideas broadly and widely as soon as possible
What Motivates Him
Is motivated by meta problems — a class of problems where once the meta-problem gets solved, a bunch of problems becomes easier to solve or solved
Building schools is a meta-problem because students get smarter and they go on to do great things themselves
Intelligence or AI in general is a meta-problem because you are unlocking productivity for the future of work across different industries
All information presented above is for educational purposes only and should not be taken as investment advice. Summaries are prepared by The Reading Ape. While reasonable efforts are made to provide accurate content, any errors in interpreting and summarizing the source material are ours alone. We disclaim any liability associated with the use of our content.