20VC Newsletter - 8th Sept 2024
Here are the transcripts and top takeaways from 20VC episodes released this last week!
Wednesday’s episode with Zico Kolter, Board Member @ OpenAI:
Download the full transcript:
My 6 key takeaways:
The Myth of Data Scarcity: We're Just Scratching the Surface!
Public models are trained on ~30TB of data.
That is a tiny amount compared to all the multimodal data we have.
We are nowhere close to tapping into the data reserve internet companies have today.
Why AI Architectures are Irrelevant Now
We are in a post architecture phase.
We need to stop looking at models on their particular structure.
We need to think about them in terms of their data input & capabilities they produce downstream.
The Biggest Problem in Models Today
We have a buffer overflow problem & we don’t know how to fix it.
This is fine for chatbots but becomes a bigger problem in more complicated LLMs.
We need to figure this out or it will get worse downstream.
Why AI & Nukes are Not a Good Analogy
Nuclear weapons have one purpose which is to destroy things.
AI is more like nuclear technology, there are still plenty of benefits.
The nuclear weapon analogy is overstated.
AI is an Accelerant for Misinformation
People already don’t believe anything they read or see that doesn’t conform to their beliefs.
AI will 100% be an accelerant of this.
We are no longer in a world where we have a record of objective truth.
How Companies Should Use AI
The companies that succeed will not fire all their workers & replace them with AI.
They will leverage their workforce & make the best use of this technology.
We don’t have AI products that are maximally used by workforces.
Friday’s episode with Nicholas Chirls, Founder @ Asylum Ventures:
Download the full transcript:
My 7 key takeaways:
The Only Way to Make Money at Seed
Invest in something no one cares about.
Those are not $25M post.
It is one single most important lesson I’ve learned over the last 13 years.
Why Most VCs Are Sheep and Most AI Investments Will Go to Zero
VCs are all playing catch up.
Every single AI company is going to be $25 - $30M post.
They should have been making those investments years ago when no one cared.
The Best From 2015 Have Not Returned Cash and Why the Last Vintage Will Not Return Cash
The top quartile of funds in 2015 have not fully returned cash.
A 1x fund will be the top quartile in the 2021 vintage.
75% of funds raised in 2020-2021 will not return money.
Why Investing in Competitive Markets Is for Losers
If an idea is good & understood by the market, we assume there are at least 10-15 competitors.
There is too much uncertainty: We don’t know what they’re doing & how good the teams are.
We love investing in good ideas that the markets don't care about yet.
Why You Should Always Sell When the Founder Sells
We want to be aligned.
We only sell when the founder is selling & we never sell all the shares.
There are many investors who are afraid to have that conversation.
Why Big Firms Want to Fund Capital Inefficient Businesses and Why LLMs Are Perfect Then
Their business model is deployment, so they’re raising as much money as possible & getting it out the door ASAP.
Foundation models require billions of dollars for Nvidia GPUs.
They’re a big VC firm’s dream.
Why This Investor Wants to Buy Common Not Preferred Shares
We want to be better aligned with the founder.
If buying common shares gets us in the right companies, it is worth it.
I’ve been thinking we should buy common, not preference shares from now on.
Let us know what your big takeaways from this week’s shows were in the comments below!
Thank you for reading, and don’t miss the great guests we have next week:
Sean Rad, Founder @ Tinder
David Schneider, GP @ Coatue
Mehdi Ghissassi, Director & Head of Product @ DeepMind
Thank you for reading 20VC.
This post is public so feel free to share it.
Thanks for reading 20VC! Subscribe for free to receive new posts and support my work.