ip26
ip26 on Hacker News
Stories are like thought experiments. I agree that blindly trusting you should “be like this” is silly, but the story works through the preconditions and consequences with you.
steer the direction a bit... and after a while, you have a long-term view, are making the decisions, but there's someone a few nodes downstream who can identify some problems with your idea?
such as helping you create a network that's good at playing goThis is really what’s happening right now… LLMs are excellent computer users. Which makes one of the frontiers “amazing things computers always could have done, but we lacked the programmers to do”.
If basically anyone can train a high quality CNN for just about any purpose now, what new opportunities appear that computers can help with?
Agreed, I keep thinking to myself that there's a huge mathematical question right in front of us today that is in exactly this vein - all the various nuances of why LLM's work so well is a mathematical question. As far as I know, it's not really understood beyond "
we do this basic thing (that makes sense) to predict that a noun is followed by a verb, and then we scale it up a bazillion fold and it can contribute to mathematics research".
As a comparison, classical computing has been scaled up a bazillion fold too, and can do things which are absolutely miraculous, but every layer of abstraction is discretely understandable.
Take garbage hauling. You have five haulers running the exact same routes through town, stopping at different houses. Government is unhappy with the tremendous added wear on the roads from the redundant trucks, and the extra traffic, so it strikes a deal and grants exclusivity to one hauler.
This is a granted monopoly. It has real positives, such as the same service at 5x less road wear. It should also be obvious that to be positive overall the deal needs to prevent abuse of the public.
Regulation is usually what you accept in
exchange for a monopoly. I would argue a granted monopoly without any regulation is evidence of regulatory capture.
If you want to argue from first principles, and we accept for a moment that granted monopoly is the system we are working in (whether or not you feel it's the optimal regime) then I'd argue there's a clear gap in regulation, as flagrant abuse of the consumer has not been prevented.
I question whether that's the dominant assumption. I know there's hoarders, but there's also
loads of "stuff" associated with day-to-day life. Live in one place and you'll have boxes of bandages and aspirin, tools for fixing the lawn mower, a small compressor for blowing out your sprinklers, board games, extra coffee mugs, shoes that are ok but you don't like very much, on and on. None of that "junk", even if you might use it rarely and certainly won't perish without it.
There's little nobility in throwing out your fourth set of perfectly serviceable bed sheets or the nicked pair of nail clippers in the guest bathroom. Which leads to accumulation, like moss on a tree.
We can dream that every human should be self reliant enough to personally inspect a used car, test knife steel for ductility and carbon content, inspect the stitching in a pair of pants, on and on.
But if you need vast expertise and extensive product testing just to make a trip to the grocery store, that’s not a fluid market. You are requiring a system with a heavy reliance on unpaid labors by all parties.
There’s always a place for expertise, but trusted brands support market efficiency. The simplest is “I bought one of these before and I liked it; I trust that the brand & vendor will sell me an identical copy” without needing to interrogate their supply chain to confirm it’s the same good. Another is “if I buy this and the product has a defect, they will help set it straight”.
Trust in the seller is not far away in importance from transparent pricing, and arguably if the product hides a flaw (like a knockoff) the pricing is ipso facto not transparent.
The dark part is you, as a consumer who is willing to pay to not watch ads, are worth considerably more than a few cents to advertise to. In fact, the more you are willing to pay to have no ads, the more valuable showing ads to you is…
A large fraction of the innovation in CPUs is driven by working around the memory wall. I anticipate AI inference will follow the same trend, and innovations that work around the autoregressive nature will be enormously impactful.
Speculative decoding is an example. An accurate draft model can reduce the number of times you stream through memory by a factor of 4x.
I've got bad news for you jsrozner, John down in accounting at <your employer> did something very unethical last week. So you, jsrozner, a first level manager in customer support who has never even met John, are going to jail for a decade and all lifetime compensation will be clawed back.
You can't give all citizens effective veto power and then expect it will only be used for good in cases where the law is behind.
It sounds more like a joke or a lampoon of always-connected culture.
you need to be able to experimentWith what? It’s not prima facie bad to have limits or bounds. I’d agree it’s good to be able to be goofy, but there’s plenty of actually-illegal activities teenagers used to “experiment” with (and skate by under the radar) that doesn't feel like a loss if it’s gone.
Clearly the point of this paper was not to flag a specific lake as being a specific problem. It seems pretty clear it's a case study meant to provide generalizable information and principals for remediation, and while they didn't predict the exact glacial lake that would collapse first, they were flagging danger in exactly the right region.
It’s a cop out, but depends on the company and it’s somewhere in between. Even with
success-of-the-company as your compass, other stuff always matters. Good work done invisibly can’t be rewarded. Working on the lowest-profile projects is often interchangeable with working on the least important projects to the company. And if you are building credit with the wrong people, maybe you aren’t working on the right problems.
There’s obviously the dysfunctional form of all this, but it’s also rare that there’s one simple and obvious “do this thing and it’s best for the company and everyone will automatically understand what you did”
I'm always suspicious we read far too much into most works, as if they were all live examples of The DaVinci Code. It feels inevitable that one day, centuries from now, someone will "realize" that JK Rowling embedded a map of hyperspace into her phonemes or that Banksy was actually issuing incisive commentary on the contrast between ancient Sumerian government and veganism all along.
An intelligent person can convince themselves of anythingNot to mention via shortcuts, reductionism, and other (ironically) lazy thinking. After all, they're too smart to take a shortcuts that lead them to the wrong answer, so if they took a shortcut it saved them effort and was also right!
Seriously, aside from the folks with the “
2hrs of sleep is enough” gene, it’s not. It’s an actively harmful myth that is sold hard. Start gathering your own evidence, with a critical fact-finding eye. Don’t just accept the narrative. It’s one of those legitimizing stories that serves the teller.
Try watching The Dawn Wall or Free Solo, keep notes, and fill in the blanks on how long they were gone from family. It’s a canned example of how it gets glossed over.
The same attitude as towards gear in weightlifting is appropriate. They tell you it’s natty all day long, but you know delts don’t get that round without gear.
I've started peering behind the curtains when I get the chance, and I've not yet found someone wildly successful who is also a model parent. Time after time, their spouse is quietly shouldering an enormous load, or their children spin their wheels and wind up rudderless, etc.
You can certainly be successful and be a good parent, but one of the first steps is recognizing and picking out the false prophets. You can't hold yourself to a superhuman standard, which is what a parent maximizing career while letting family malinger will present. They will say, Look at me, a captain of industry with vaults of gold, who has the physique of a twenty-five year old and is also an all-star parent. And you have to figure out that it's a lie.
I had roughly the same realization about the pillars of math & science years ago. Many terrible, terrible parents whose children suffered mightily. While I had aspired to intellectual impact all my life, I had an unexpected realization. I could not bear the thought of being famous and intellectually successful yet having totally failed my children. It reversed my longstanding priorities, and I realized I wanted to be a parent more than I had thought.
I still believe that having passionate and driven parents is good for kids.
Exactly, it feels dishonest that this point is so rarely brought up in essays on formal methods. You can do things like prove that all possible faults are always caught, or any memory that is accessed has first been malloc’d, or that the API endpoint will always respond (liveness). These are often both easy to specify and difficult to guarantee with conventional testing.
What exactly does a bespoke ISA buy you? What will justify hundreds of engineers writing a new good one & working through the compiler support?
I bet it would be pretty easy for a user to interoperate between 67 different drafts of the HTTP spec once they were all written, validated, and library implementations coded. But what would that Cambrian explosion of API revs actually get you?
Have done a lot of “parallel invention” where I invent something genuinely novel & new-to-me, only to later discover it was first developed eighty years ago.
The part that’s really disheartening is there’s so much knowledge out there, I wonder if this is actually efficient - maybe you can’t reasonably search the state of the art across all of human knowledge and so you have to invent it again. But that’s such a sad outcome.
They even loudly declared how they were going to clean up all the messy corners and warts.
The underlying truth seems to be that a “clean” ISA doesn’t by nature make the beer taste better, and many of the warts probably have a reason for existing.
It's usually a pretty clear case of reaching for statements that both sound impressive while also being broad and vague enough they are less likely to be factually wrong. In this regime, being difficult to parse is actually part of the performance, because it prevents the user from being able to spot a clear error.
It’s getting everyone to understand the problems, working together (etc)…There are manuals for this too, and interestingly enough this has been studied since the Romans at least! Is it then really the hard part?
There’s an ego element though. Being mission driven and choosing to work for Anthropic is implicitly validating of the “peril” and of the leadership team as the ones to steer through it, so they are indirectly asking for this validation.
Certainly room for such discussion. However if the parent waits to first save a large sum before moving there (instead of moving there straight out of high school), their revealed preferences suggest even they do not feel the tiny village is great for young people.
If you have dependents that are children, I feel there is a moral consideration. If you have earned $1M and then take the family to live who-knows-where slowly drawing on that $1M.. will your children be able to replicate your success? Will they be able to earn their own $1M? Or will leading this new life significantly impair their prospects?
To me this is the most critical problem with the "move somewhere crazy cheap" plan. I feel a parent owes their children at least somewhat similar opportunities as they themselves had. If you got to work as "high flying white collar tech worker", but the only option your children have is "subsistence fisherman", you have wronged them.