No problem, all of you are welcome! Very enjoyable article.
595 karma · joined June 27, 2023
It's the syntax for sed substitution for someone who mistypes, is dyslexic or does not have an excellent memory.
Yes, and even worse:
> Even more damning, on 6 Oct 2022, we see the following: one commit at 21:53:09 +0300 followed by another at 17:00:38 +0800. This is only a difference in a matter of minutes!
No, this is a difference of almost 10 hours...
Maybe the author inadvertently switched the two analyses? Or maybe the author is just not very good with timezone math... :)
Why would the password be sent in plaintext instead of, say, sending a hash of the password calculated with a salt that is unique per SSH server? Or something even more cryptographically sound.
In fact, passwords in /etc/shadow already do have random salts, so why aren't these sent over to the SSH client so it can send a proper hash instead of the plaintext password?
I would imagine seeing the ground above you to be much worse for passengers...
You could create a spec that would specify exactly what a program does, but generally speaking that would have no useful purpose.
For every program that does something, there are multiple ways to implement it. Often, different ways of implementing a program have different performance profiles, which is why many programs become more complex than the most naive way of implementing it.
A spec almost never specifies anything other than the simplest possible way that a program should be implemented (because when creating a spec, performance is irrelevant). This is usually way easier to make sure that it is correct than a spec that would specify exactly how a program should be implemented, including things done for performance.
Furthermore, you don't have to necessarily trust a specification. Specs can also be proven to have certain properties, i.e. they can be formally verified and proven to have absence of (perhaps certain classes of) bugs, and in some cases, can even be proved to have full correctness (yes, the spec itself, not just the program that implements it).
So I don't think it's fair to equate a program to a "sufficiently detailed spec", since most of those details are actually irrelevant (or at least, they should be, and can be proven to be) with regards to the correctness of a program.
One cool thing about formal verification is that you can not only prove things about your code, but you can prove things about the specification itself (with some approaches, at least). This includes proving arbitrary properties, proving the presence of bugs, proving the absence of bugs and in some cases, even proving full correctness of the specification.
> I'm bullish on more sophisticated type systems [...] but not formal verification.
I don't know what kind of formal verification framework you used that left you with this conclusion, but the more sophisticated is your type system, the closer you are to doing formal verification.
Yes, even today, for example, Costa Rica has no military yet they have their own currency. Its Central Bank manipulates its value by buying and selling the currency in the free market, which is what many countries do nowadays (although not always by the same mechanisms).
> Has there been a dramatic military failure that hasn't been followed by currency decline?
Yes, unless by "dramatic military failure" you mean having their economy destroyed, at which point of course the currency fails, but that's because economy destruction leads to currency failure regardless of any military defeats.
However, the better question is: has any country with a strong military ever had its currency fail without any military defeat?
And the answer is yes, plenty, which proves that having a strong military has nothing to do with having a stable currency.
Let's say your currency starts to fail (say, due to hyperinflation)... what is the government going to do with the military to prevent the collapse? Order them to intimidate the market?! That's a completely ridiculous proposition...
Finally, someone who understands money. As opposed to the typical "but my military!", as if no country with a strong military had ever had its currency fail...
I would only emphasize the importance of the scarcity aspect of money, which seems to be a (purposely) forgotten lesson nowadays. It reminds me of the historical importance of the Spanish Price Revolution, which with its 1.2% yearly inflation, nowadays it would be considered laughable.
Tell that to sleepwalkers!
Hopefully without having to induce those brainwaves, otherwise you might be able to do it during the day but you wouldn't exactly be conscious. But there's also the question of whether those brainwaves are doing more than just cleaning the brain.
I think some types of learning and memory formation also happen during sleep, right?
I see some signs that you've never been in a drug deal gone awry...
> Why do we keep promoting tall people???
They have a better (over)view of the situation.
But more seriously, there is also an argument to be made that free will is an illusion, at which point you have to wonder what "freedom" really means, if anything.
If an entity can predict the correct answer to a question (with a sufficiently low margin of error), then it knows the answer.
However, if the prediction contains too much uncertainty, then the entity should not act like they know the answer.
The above is valid for humans and LLMs.
So we "just" need to model and train LLMs to take uncertainty into account when generating outputs. Easy, right? :)
I assume you meant asking rather than answering?
> An LLM doesn't have the other human motivations a person does when asked questions, pretty much at this point with LLMs there are only one or two 'voices' it hears (system prompt and user messages).
Why would LLMs need any motivation besides how they are trained to be helpful and the given prompts? In my experience with ChatGPT 4, it seems to be pretty good at discerning what and how to answer based on the prompts and context alone.
> Whereas a human will commonly lie and say I don't know, it's somewhat questionable if we want LLMs intentionally lying.
Why did you jump to the conclusion that an LLM answering "I don't know" is lying?
I want LLMs to answer "I don't know" when they don't have enough information to provide a true answer. That's not lying, in fact it's the opposite, because the alternative is to hallucinate an answer. Hallucinations are the "lies" in this scenario.
> In addition human information is quite often compartmentalized to keep secrets which is currently not in vogue with LLMs as we are attempting to make oracles that know everything with them.
I'd rather have an oracle that can discriminate when it doesn't have enough information to provide a true answer and replies "I don't know" in such cases (or sometimes answer like "If I were to guess, then bla bla bla, but I'm not sure about this"), than one which always gives confident but sometimes wrong answers.
Are you saying that LLMs can't learn to discriminate between which questions they should answer "I don't know" vs which questions they should try to provide an accurate answer?
Sure, there will be an error rate, but surely you can train an LLM to minimize it?
Answering "I don't know" or "I can't answer that" is a perfectly plausible response to a difficult logical problem/question. And it would not be a hallucination.
I haven't read the paper, but that sounds like it would only be true if the definition of "hallucinating" is giving a wrong answer, but that's not how it's commonly understood.
When people refer to LLMs hallucinating, they are indeed referring to an LLM giving a wrong (and confident) answer. However, not all wrong answers are hallucinations.
An LLM could answer "I don't know" when asked whether a certain program halts and yet you wouldn't call that hallucinating. However, it sounds like the paper authors would consider "I don't know" to be a hallucinating answer, if their argument is that LLMs can't always correctly solve an NP-complete problem. But again, I haven't read the paper.
It's not him, no double spaces in his writings.
The problem is the transition into that new world.
Although unsolicited calls are not a problem here, fortunately.
I know of at least 2 countries where police will be immediately dispatched to a robbery in progress if a person is caught on camera by the alarm system and the owner confirms it's a robbery (e.g. on a phone call). By law, police won't be dispatched unless a person is caught on video, due to false alarms.
I know this because that's what happened when my home was robbed and I was out of town. Fortunately, since the police response was quick the thief didn't have enough time to take anything (!), as he must have been in and out in 2 minutes at most.
It also helped that there was no jewelry or cash inside the house, of course.
Regenerative breaking is not a benefit compared to steady driving, it's a waste of energy due to the second law of thermodynamics.
It's only a benefit when you compare it to non-regenerative breaking.
Edit: source (45 seconds): https://www.youtube.com/watch?v=fvFixaZdvCg