One commenter already clarified how bounty amount is related to black market value. Now a lot of others might seem how Google doesn’t value security enough. (Or other companies).

But one has to understand that for security purposes they SHOULD pay as little as possible. If they pay out more there is more incentive in finding bugs and then there unfortunately you’ll also raise more black market.

So GTO strat is to just cut off black market with as little money as possible.

A physical coin is biased?

Empirically, human-flipped coins have about a 1% bias toward the same side they started on: https://www.ams.org/publicoutreach/math-history/hap7-fifty-o...

But why it says "favors heads ~51%"

What sort of uncivilized mook dares flip a coin with tails facing up? Preposterous

Updated to "favors one side 51%".

I feel like when I flip a physical coin it pretty much always lands on exactly the same side unless I flip it real wierd.

Yes this works good for me too rather take your time and do the first prompt right

Yes it’s usually worth it to try to write a really good first prompt

More than once I've found myself going down this 'little maze of twisty passages, all alike'. At some point I stop, collect up the chain of prompts in the conversation, and curate them into a net new prompt that should be a bit better. Usually I make better progress - at least for a while.

This becomes second nature after a while. I've developed an intuition about when a model loses the plot and when to start a new thread. I have a base prompt I keep for the current project I'm working on, and then I ask the model to summarize what we've done in the thread and combine them to start anew.

I can't wait until this is a solved problem because it does slow me down.

Yes when new models come out it feels like breaking up.

Why is it so hard to share/find prompts or distill my own damn prompts? There must be good solutions for this —

What do you find difficult about distilling your own prompts?

After any back and forth session I have reasonably good results asking something like "Given this workflow, how could I have prompted this better from the start to get the same results?"

Analysis of past chats in bulk.

Don’t outsource the only thing left for our brains to do themselves :/

Thank you!

At these stakes you can expect anyone to do what they believe is making them the most amount of money.

I’m not sure that doesn’t sit well with me.

Rule 1 should be: Reproduce with most minimal setup.

99% you’ll already have found the bug.

1% for me was a font that couldn’t do a combination of letters in a row. life ft, just didn’t work and thats why it made mistakes in the PDF.

No way I could’ve ever known that if I wouldn’t have reproduced it down to the letter.

Just split code in half till you find what’s the exact part that goes wrong.

Related: decrease your iterating time as much as possible. If you can test your fix in 30 seconds vs 5 minutes, you’ll fix it in hours instead of days.

Rule 4 is divide and conquer, which is the 'splitting code in half' you reference.

I'd argue that you can't effectively split something in half unless you first understand the system.

The book itself really is wonderful - the author is quite approachable and anything but dogmatic.

It’s not just opensource. It’s also Claude, Meta and Google, of which the latter have real estate (social media and browser)

Yes and Anthropic, Google, Amazon are also facing commoditization pressure from open-source

Well one key difference is that Google and Amazon are cloud operators, they will still benefit from selling the compute that open source models run on.

Wait how did you buy it. I’m just getting forwarded to Team Plan I already have. Sitting in Germany, tried US VPN as well.

The endpoint for upgrading for the normal web interface was returning 500s for me. Upgrading through the iOS app worked though.

I do start at $300/hr

I didn’t just set that, I need to set that to best serve.

Debugging must be fun