Jeremy Schlatter
@JGSchlatter
Followers
148
Following
15K
Media
6
Statuses
110
The thing I regretted about this was that it included a non-disparagement agreement. But, to my surprise, OpenAI has now released me from it. I am relieved and grateful for that!
0
0
5
I'm not sure who has and has not been released so far, but: https://t.co/cSg6LDDvTW
@JGSchlatter Anyone in this position can email hr@openai.com asking to be released from the non-disparagement provisions!
2
0
10
The thing I regretted about this was that it included a non-disparagement agreement. But, to my surprise, OpenAI has now released me from it. I am relieved and grateful for that!
I signed a severance agreement when I left OpenAI in 2017. In retrospect, I wish I had not signed it.
2
1
82
I don't mean to imply that my situation is the same as those in recent coverage. For example, I worked at OpenAI while it was still exclusively a non-profit, so I had no equity to lose.
2
1
69
I'm posting this because there has been coverage of OpenAI severance agreements recently, and I wanted to add my perspective. https://t.co/Sg0bYw0CVP
https://t.co/XY6ECae9uM
You can read some email exchanges between OpenAI and ex-employees over at https://t.co/p0ual66AZ0. There are a lot of forms of courage, but this sure is one of them.
2
0
46
I signed a severance agreement when I left OpenAI in 2017. In retrospect, I wish I had not signed it.
3
6
173
Anyway, aside from this point about PoW and PoS, I did really like the article. I actually clapped at the end š
0
0
1
One virtue of precision here: it matters exactly how high that cost is to revert blocks. On smaller chains, sometimes people actually pay the cost and revert old blocks!
1
0
1
It is also true that most validators probably care about ETH at least in the near term, and that they would be worse off if ETH lost value. But that's not the main reason that they don't "misbehave". The main reason for that is that it is just too expensive to do so.
1
0
1
As a rough ballpark, there is currently 14.4M ETH staked by ethereum validators. ETH is currently trading at $1500, so that's $21.6B worth of stake. To revert a finalized block would cost (_very_ roughly) 2/3rds of the stake, or $14B.
1
0
1
Cryptography prevents block producers from forging transactions, but it does not prevent them from _reverting_ transactions. The thing that keeps them from _reverting_ transactions is economic incentives: it would be _prohibitively expensive_ to do so.
1
0
1
The point of Proof of Work and Proof of Stake is not to prove that you care. The point is to make it expensive to revert transactions!
1
0
1
>[In Proof of Work] you buy a bunch of computer hardware and pay for a lot of electricity and do a bunch of calculations to prove you really care about Bitcoin.. [In Proof of Stake] you just invest a lot of money in Ethereum and post it as a bond, which proves you care.
1
0
1
There was a part of his exposition that I think is a bit vague to the point of being misleading, though. It shows up a few times, but most crisply in this passage:
1
0
1
I really enjoyed reading this. Matt is a good writer and I like seeing his perspective as someone who likes finance but isn't a crypto true believer.
what is crypto
1
0
5
I agree with this.
I think it's generally bad to work on AI capabilities. Not always (since alignment tech is often dual-use), and not forever (since never building TAI seems awful). But broadly, I think we should not build extremely powerful AI systems before we figure out how to make them safe.
0
0
5
And filtermap: (def filtermap (f xs (o p idfn)) (filter p (map f xs))) This one ended up being so short it didn't seem worth giving it its own definition in my little standard library. Which is great!
0
0
2
Same with filter: (def filter (p xs) (foldr (fn (x rest) (if p.x (cons x rest) rest)) nil xs))
1
0
1
Today I wanted to do a deep copy in Bel. Turns out itās pretty simple to implement: (def deep-copy (x) (if atom.x x (cons (deep-copy car.x) (deep-copy cdr.x))))
1
0
2
āThe Hammer and the Danceā This is a very good summary of the best analysis and strategy I've seen so far on #COVID19. Here's the strategy, in brief:
New Article! Summary: Strong #coronavirus measures today should only take a few weeks, there wouldnāt be a peak of infections afterwards, and it can all be done for a reasonable cost to society, saving millions of lives along the way. https://t.co/ceZHwTZmHM
2
12
32