Nima Keivan
@nkeivan.bsky.social
Here for the discussion. Dad x 3 . Building autonomous SW dev
@ durable.ai. Writing durable.substack.com. 2x founder (prev exit to AMZN)
@ durable.ai. Writing durable.substack.com. 2x founder (prev exit to AMZN)
> It was like, created somehow
It was created because people were incentivized to publish the product of their thinking and writing online free of charge. I imagine getting credit for it was at least part of the incentive structure.
It was created because people were incentivized to publish the product of their thinking and writing online free of charge. I imagine getting credit for it was at least part of the incentive structure.
December 16, 2024 at 2:09 AM
> It was like, created somehow
It was created because people were incentivized to publish the product of their thinking and writing online free of charge. I imagine getting credit for it was at least part of the incentive structure.
It was created because people were incentivized to publish the product of their thinking and writing online free of charge. I imagine getting credit for it was at least part of the incentive structure.
What's the difference between a good coach and a good advisor? I've been thinking a lot about this after getting lucky and finding an amazing coach for my son.
A good advisor is someone with experience that you can reach out to when you need help. They're here to help, but you set the bar.
A good advisor is someone with experience that you can reach out to when you need help. They're here to help, but you set the bar.
December 16, 2024 at 2:05 AM
What's the difference between a good coach and a good advisor? I've been thinking a lot about this after getting lucky and finding an amazing coach for my son.
A good advisor is someone with experience that you can reach out to when you need help. They're here to help, but you set the bar.
A good advisor is someone with experience that you can reach out to when you need help. They're here to help, but you set the bar.
The problem with always speaking with certainty is being confidently wrong.
As a listener, the only rational choice is to be skeptical of all of it.
As a listener, the only rational choice is to be skeptical of all of it.
November 19, 2024 at 4:28 PM
The problem with always speaking with certainty is being confidently wrong.
As a listener, the only rational choice is to be skeptical of all of it.
As a listener, the only rational choice is to be skeptical of all of it.
I'm been thinking about what makes a good explanation. My current theory:
A good explanation is one that can't be trivially used to explain something else.
Take "how does an LLM predict the correct next token?"
🧵
A good explanation is one that can't be trivially used to explain something else.
Take "how does an LLM predict the correct next token?"
🧵
November 16, 2024 at 2:19 AM
I'm been thinking about what makes a good explanation. My current theory:
A good explanation is one that can't be trivially used to explain something else.
Take "how does an LLM predict the correct next token?"
🧵
A good explanation is one that can't be trivially used to explain something else.
Take "how does an LLM predict the correct next token?"
🧵
Neats v. scruffies was differentiating along the wrong axis (neatness). The real differentiation is about legibility. It should be something like:
designers v. growers
or
synthetics v. organics
designers v. growers
or
synthetics v. organics
November 16, 2024 at 2:18 AM
Neats v. scruffies was differentiating along the wrong axis (neatness). The real differentiation is about legibility. It should be something like:
designers v. growers
or
synthetics v. organics
designers v. growers
or
synthetics v. organics
Once the gains from pretraining plateau, the closed source model providers will likely need to build in the application layer themselves to meet growth expectations.
We’re already starting to see this with searchgpt.
We’re already starting to see this with searchgpt.
November 16, 2024 at 2:18 AM
Once the gains from pretraining plateau, the closed source model providers will likely need to build in the application layer themselves to meet growth expectations.
We’re already starting to see this with searchgpt.
We’re already starting to see this with searchgpt.