Read on Twitter
Idea from machine learning applied to startups
Repeat founders are like linear models, first-time founders are like deep learning
Repeat founders are like linear models, first-time founders are like deep learning
In ML, inductive bias is what lets a model generalize beyond training data.
Models that make strong assumptions (e.g., linearity) can learn quickly from data
Models with weaker inductive bias (neural networks) are more flexible but need more data to learn
Models that make strong assumptions (e.g., linearity) can learn quickly from data
Models with weaker inductive bias (neural networks) are more flexible but need more data to learn
Repeat founders are awesome, they work right out of the box lol
They often have domain expertise, and already have experience running a startup
They don't need that many reps (see that much data) to know how to do the right things
They often have domain expertise, and already have experience running a startup
They don't need that many reps (see that much data) to know how to do the right things
First-time founders start out doing lots of things wrong (or maybe that's just us...)
Like deep learning models, you can't really tell for a while if they're going to converge to anything good
Just keep tweaking params and hope for the best hahah
Like deep learning models, you can't really tell for a while if they're going to converge to anything good
Just keep tweaking params and hope for the best hahah
As Justin Kan says, we're bad at everything 
https://twitter.com/justinkan/status/1066859732244193280
https://twitter.com/justinkan/status/1066859732244193280
To make this framework useful...
First-time founders, do what deep learning researchers do:
1. be deliberate about how you collect data
2. collect a lot of data
3. carefully design the processes (learning algorithms) to turn that data into better decisions and better product
First-time founders, do what deep learning researchers do:
1. be deliberate about how you collect data
2. collect a lot of data
3. carefully design the processes (learning algorithms) to turn that data into better decisions and better product
Talking to users is the main way that you'll collect data to train your "model" â here's a great thread on how that looks as you grow 
https://twitter.com/spencerfry/status/1293566829022126080
https://twitter.com/spencerfry/status/1293566829022126080
Tactically there's a lot that you can do here:
- build feedback into your product like Stripe
- automate interview scheduling with users
- watch session recordings (Fullstory)
- send out surveys
Just make sure the channel matches the complexity of the information you need
- build feedback into your product like Stripe
- automate interview scheduling with users
- watch session recordings (Fullstory)
- send out surveys
Just make sure the channel matches the complexity of the information you need
To make sure that you're squeezing as much learning as possible out of the data you collect, I highly recommend @AskHerald
They'll help you track and organize your data in one place so that you can actually use it to make decisions and prioritize
They'll help you track and organize your data in one place so that you can actually use it to make decisions and prioritize
What other processes, tactics, and tools are startups using to increase their learning rate?
I think about this all the time but as a first-time founder, this framework says that we're probably getting a lot of this wrong
I think about this all the time but as a first-time founder, this framework says that we're probably getting a lot of this wrong
No shade to repeat founders btw, figured more people would be familiar with linear models but you're probably more like xgboost ;)
