Some personal musings around the YC W23 Generative AI Landscape
There are 36 AI/ML-related startups in this YC Winter Batch out of the 201 startups accepted in total.
This represents ca 18% of the current cohort being some kind of Generative AI startups.
Is 18% a lot or not?
Let’s compare this with the Y Combinator Winter & Summer Batch from ’21 when we were at peak Crypto & Web3 hype:
- The ’21 YC Winter Batch had in total 333 startups with 9 Crypto / Web3 ones: representing ca 2.5% of the cohort
- The ’21 YC Summer Batch had in total 391 startups with 15 Crypto / Web3 ones: representing ca 4% of the cohort
📈 So ca 3% were Crypto & Web3 startups in the 2021 YC batches compared to ca 18% being Generative AI startups in the 2023 Winter Batch.
Should concerned conclusions be made?
I wouldn’t personally make too many conclusions from this data, albeit I believe that the success rate in Generative AI investing will in the short- to medium-term be quite low as many tourist VCs with no background in the space are jumping on the hype bandwagon, throwing too much money at some startups/founding teams (when they should still iterate on their approaches).
📚 I wrote a few weeks back a Medium article for Better Programming, where I among others discuss the concerns about whether one can build a defensible business on top of any of the foundation model platforms: https://lnkd.in/dKjTC8uA
The risk with any hype cycle is that things get overheated and we might in a worst-case scenario experience another AI Winter. A certain level of hype can’t be maintained — and at some point, the industry starts underdelivering. AI/ML turns out to be surprisingly fail-ridden. Companies and people that try using it to solve everyday problems discover it’s prone to errors, often quite mundane ones.
We’re not heading toward another AI Winter but we need to be cautious
Personally — I’ve been very bullish about the AI/ML transformation since 2015/16 and remember how jaw-dropped I was when testing out Transformer based text-models such as BERT in 2018.
However, we know that AI/ML has become truly big when it becomes a boring and standardized component of workflows & infrastructure, just like e.g. relational databases.
Gergely Orosz noted recently on Twitter how he never believed crypto would be anything meaningful innovation-wise because it was a unregulated finance pretending to be a technological breakthrough (it was not).
Gergely — among many others — sees AI/ML as a technology breakthrough that can (and most likely will) change a large variety of industries.
PS. 65 startups out of 201 are now AI/ML related if categories below are used for filtering YC W23 startups:
So suddenly, 32% of the startups in the YC W23 batch are AI/ML related across the following sub-categories:
→ Artificial Intelligence
→ Generative AI
→ Machine Learning
→ Conversational AI
→ Reinforcement Learning