Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Can you reverse engineer our neural network?

blog.janestreet.com

Post reply on HN
Print view View original

191–200 of 216 posts

Re: Can you reverse engineer our neural network?

#191

Post by robotpepi »

js8 wrote: post #178
robotpepi wrote: post #159

Earlier quoted context omitted.

Yes, it was clear that you wanted to refocus this as a moral problem of people. But that's irrelevant. The point of the guy above is that there is a system (the market) that creates certain incentives, and as a result, we have what we have. That's why I ask: what's your point? We still have all these problems.

So are you asking how to change it? I think that's pretty obvious once people understand it's a collective social choice - organize to change it. The point is "the market" is not some mysterious unreachable force. For example, this comment https://news.ycombinator.com/item?id=47181837 is wrong; even if you had large amount of people acting like that person does, you would still likely have a system that doesn't work…

I'm no asking how to change it. And I don't think anyone has suggested that the market is some mysterious unreachable force. To be honest, at this point it's clear that you're being condescending and assuming people believe something foolish instead of trying to understand what they're actually saying.

Re: Can you reverse engineer our neural network?

#192

Post by expensive_news »

aizk wrote: post #183
expensive_news wrote: post #139

Earlier quoted context omitted.

I was one of the solvers. It took me about a week to figure out. This is what I wrote out in my submission with the answer: > After looking at the final two layers I was somewhat quick to intuit that this was some sort of password check, but wasn’t entirely sure where to go from there. I tried to reverse it, but it was proving to be difficult, and the model was far too deep. I started evaluating the structure and saw…

Did you get an interview with Jane Street?

Yeah, they put me on the quant researcher track. Made it to the final round onsite, but they did not extend an offer T_T

Re: Can you reverse engineer our neural network?

#193

Post by sublinear »

bethekind wrote: post #35

Model interpretability is going to be the final frontier of software. You used to need to debug the code. Now you'll need to debug the AI.

Why? The vast majority of software doesn't need to be written by AI and moves at the speed of the humans making the decisions, not the speed of writing the code.

They make a shit ton of money because of this. If you're working at a place where the code matters more than the decisions that went into it, you're basically working at a sweatshop for people who are desperate for a win and will throw away you and all your code once the MVP stage is over, and that's the only way this "works".

Generative probabilistic AI is not equivalent to a compiler and never will be until we can do this kind of thing completely deterministically. No matter how much you reduce the error in the "model", it's still more error than the error rate of the logic gates. It's completely futile considering the sheer depth of indirection at play, and that indirection is the whole point of software.

Re: Can you reverse engineer our neural network?

#194

Post by blackqueeriroh »

sublinear wrote: post #193
bethekind wrote: post #35

Model interpretability is going to be the final frontier of software. You used to need to debug the code. Now you'll need to debug the AI.

Why? The vast majority of software doesn't need to be written by AI and moves at the speed of the humans making the decisions, not the speed of writing the code. They make a shit ton of money because of this . If you're working at a place where the code matters more than the decisions that went into it, you're basically working at a sweatshop for people who are desperate for a win and will throw away you and all your…

Lmao bro compilers aren’t deterministic either

Re: Can you reverse engineer our neural network?

#195

Post by vsgherzi »

blackqueeriroh wrote: post #194
sublinear wrote: post #193

Earlier quoted context omitted.

Why? The vast majority of software doesn't need to be written by AI and moves at the speed of the humans making the decisions, not the speed of writing the code. They make a shit ton of money because of this . If you're working at a place where the code matters more than the decisions that went into it, you're basically working at a sweatshop for people who are desperate for a win and will throw away you and all your…

Lmao bro compilers aren’t deterministic either

could you expand on this a little? As far as I know it's a bug in rustc gcc and llvm if the outputs are not deterministic

Re: Can you reverse engineer our neural network?

#196

Post by FreakLegion »

ses1984 wrote: post #150
lupire wrote: post #125

Earlier quoted context omitted.

By "pumping and dumping" do you mean "buying and selling" to outsmart other people trying to outsmart them? Did they lie about the financial health of the securities they traded?

You could just read the article. This is outside my domain, and I don’t know the details, but in many cases Jane street functions as a market maker, market makers have access to information they can exploit to skim from anyone that trades through them, especially retail investors who place market orders. Pump and dump is a strategy that whales can use to bully smaller traders, not unlike how in poker the smaller your…

What you're describing isn't a pump and dump, but in any case what Jane Street did wasn't a pump and dump or what you're describing. It also had nothing to do with market making.

India's market trades options much, much more than the underlying stocks. This means that on one hand you can trade a lot of options without moving the market, and on the other you can move the market by trading comparatively few shares. Since options prices tend to be bounded by the price of the underlying, this is...a problem. For example you could buy shares to move the price up, sell calls, buy puts (aka a collar), then sell the shares to move the price back down so both calls and puts make money.

But it doesn't necessarily look like this is what Jane Street was doing. Instead they seem to have realized that stock and option prices already regularly diverged, and put the collars on to profit from corrections. In other words: arbitrage. Which, fair, can be functionally indistinguishable from market manipulation. But on paper it looks like they made prices better for everyday folks at the expense of the market makers and other institutions.

Matt Levine wrote a long Money Stuff column about this around the middle of last year.

Re: Can you reverse engineer our neural network?

#197

Post by alternatex »

vsgherzi wrote: post #195
blackqueeriroh wrote: post #194

Earlier quoted context omitted.

Lmao bro compilers aren’t deterministic either

could you expand on this a little? As far as I know it's a bug in rustc gcc and llvm if the outputs are not deterministic

Don't bother. They're going to claim that cosmic radiation can alter bits so compiler determinism is comparable to LLM. I'd rather AI evangelists claim determinism is not required to build decent software than perform the mental gymnastics required to make these comments.

Re: Can you reverse engineer our neural network?

#198

Post by js8 »

janalsncm wrote: post #177
js8 wrote: post #33

Earlier quoted context omitted.

"The market" here is just a convenient substitute for "people weighted by the disposable income", which today roughly approximates to "rich people".

Businesses and governments spend money as well. For example, the US government is pretty interested in having really good weapons. So the market responds by developing weapons for the government.

These people, who make purchasing decisons, also make them on behalf of someone's preferences, mixed with their own, to the extent they have a say. Like the market, government represents certain weighted subset of people. With the market, the people in the government are under influence just as much as anybody else. You can't really say that preferences flow a certain way, from government to business.

Re: Can you reverse engineer our neural network?

#199

Post by mettamage »

vsgherzi wrote: post #195
blackqueeriroh wrote: post #194

Earlier quoted context omitted.

Lmao bro compilers aren’t deterministic either

could you expand on this a little? As far as I know it's a bug in rustc gcc and llvm if the outputs are not deterministic

I was having dinner with a friend yesterday who told me that lock free data structures aren’t lock free since they lock on the hardware level. He also had a thing to say about compilers in C++.

He said something like: it’s crazy that you have the C++ standard but then also how a compiler implements that standard. And then it still matters how that compiler is implemented on a particular architecture! It’s always different and the annoying thing is that this is permutative. This is why we use X86-64 Intel CPUs at and (I forgot the name). That simplifies it but you still need to know 3 things. He also was frustrated how lock free data structures and memory reordering are 2 completely different concepts in practice but not in C++ but I didn’t fully follow him. I wrote 100 hours in C++ so I was already happy I understood everything else.

But based on his rant I can sort of see why compilers feel not deterministic.

Post reply on HN