MediumD

Of course most startups fail, and most equity is worth nothing.

I guess I didn’t think “routinely” implied a specific percentage, just that it isn’t uncommon for options to be worth a lot.

If even 5–10% of VC startups succeed, then it’s still worth considering the expected value of the equity when comparing job offers.

While I think it’s good advice to live as if the equity is worth zero, treating all equity as if its worth nothing, seems a bit over-reductionist when equity packages can routinely be worth millions of dollars.

Obviously it’s a crapshoot and should never be seen as a guarantee, I think treating it as zero is bit too far on the opposite extreme.

A lot of people saying the business model doesn't justify a $1bn valuation (rightfully so), but I'm guessing the valuation wasn't for their current business, but on the possibility that Cameo became the new way for booking talent in the age of the internet.

They could have become a $1bn business if they had "revolutionized talent management" (or something like that). Not saying it was a good investment, or one I would have made, but I'm guessing they pitched a larger vision than simply a buttload of cameos from washed-up/reality TV stars.

To be fair, I also didn’t include the session layer!

My writing isn’t a strength of mine, so I appreciate the criticism. My writing going from “bad” -> “is it AI?” is progress.

I struggled with where to “cutoff” the explanation and public key cryptography seemed like a good boundary and better explained elsewhere, as did various OSI layers.

I probably should have gone over the cert and potentially the full chain of trust, I’ll give you that.

While I agree no one is rewriting history, it is potentially a big deal because it speaks to the biases present when training/RLHF-ing. Considering this will be used by millions (if not tens of millions), calling it a “silly toy” feels off.

Bias in the model can lead to bad outcomes in certain situations (hint: we have an election coming up)

Yes this is innocuous, but it does hint at the possibility of more damaging bias being a possibility.

> We provide evidence for the Reversal Curse by finetuning GPT-3 and Llama-1 on fictitious statements such as "Uriah Hawthorne is the composer of 'Abyssal Melodies'" and showing that they fail to correctly answer "Who composed 'Abyssal Melodies?'". The Reversal Curse is robust across model sizes and model families and is not alleviated by data augmentation.

This just proves that the LLMs available to them, with the training and augmentation methods they employed, aren't able to generalize. This doesn't prove that it is impossible for future LLMs or novel training and augmentation techniques will be unable to generalize.

I no longer work at Brex, but I believe they did this using Kotlin rather than Elixir.

Which headphone did you try? That sounds amazing

This looks cool! What would be the biggest motivation for using this over something like slateJS?

I think this is one of those things where people overestimated how much things would change in the short-term, but will grossly underestimate how much they will change in the longterm.

5 years is a pretty short window in the grand scheme of things when talking about the adoption of technologies.