So, what happens when you test it on 11 digit numbers? I don’t mean that as a gotcha or “LOL dumb transformer” snark. More like, does the accuracy start to drop as you add digits? Or instead, maybe it’s the transformer equivalent of a stack overflow and it outputs a picture of a burning spoon or something?
And for that matter, what’s it do with 9 digit numbers? Like, is it more accurate with them, or are these little guys mainly good at adding numbers with exactly 10 digits?
Basically, are the failures modes a gentle increase in inaccuracy, or spectacle failure outside their parameters?
One major limitation of the LLM architecture is that even the failure mode varies unpredictably between inputs.
The set of 11-digit numbers with any given failure mode (or even successful output) has no discernable pattern, merely whatever randomness the training process baked into the model.
You can't predict ahead of time when they will fail spectacularly, nor draw a clear boundary around the failure cases. And early major example of this were the "glitch tokens" introduced into most LLMs by training on reddit data.
But there is an "in general"/"average failure rate across all inputs of a given size" answer: LLMs performance drops off a cliff once the input reaches too much complexity. (A "┐" shaped curve) In contrast to humans, where you can ask a child to add two N-digit numbers and the error rate will be approximately linear to N.
Most humans struggle to compute 10 digit stuff. They use tools instead. Can LLM learn to use calculator? Sorry if that is a stupid question. Maybe brains are not well suited for calculations natively.
Yes. LLMs use calculators to great effect. More often, Python as a calculator.
Also, there exist autistic savants who prove that a human brain can be used to perform rote calculations on large numbers much faster than a human with a calculator can.
Depends on how the transformer has been trained. If it has seen 11 digit examples while training it might work, else the input will be out of distribution and it will respond with a nonsensical number.
For instance the current high score model (311 params [0]), when given 12345678900 + 1, responds with 96913456789.
An interesting experiment would be: what's the minimum number of parameters required to handle unbounded addition (without offloading it to tool calls).
Of course memory constraints would preclude such an experiment. And so a sensible proxy would be: what kind of neural-net architecture and training would allow a model to handle numbers lengths it hasn't been trained on. I suspect, this may be not be possible.
"This deployment is temporarily paused" is the crappiest "this account has gone over its budget" error page ever. Does that mean the site's down, or is that some meta-joke like "you've reached the end of the Internet"? Quick, explain to my non-technical friends what a "deployment" is. They're not trying to go on a deployment; they're trying to look at a website.
In this case, I'm not sure it matters what it says or how your non-technical friends interpret it. The site is down. Why it is down doesn't change the next thing casual viewers will do (close the tab).
But it does matter to their opinion of the site: is it down because the author took it down, is it down due to a technical problem, or is it is it down because the hosting provider took it down?
"This deployment is temporarily paused", if anything, sounds like the people who put the site up took it down again. That sends the wrong message.
Personally, if my hosting provider took my post down, I'd want them to make that obvious to my visitors. Or at the very least make it look like a technical issue. Not make it look like I took it down.
Is it? The title is "The Robotic Dexterity Deadlock". For all I know, it's a joke about what deadlock looks like for robots, showing what could be interpreted as a deadlock in a webserver. At a glance, I can't tell if the site is down, or if it's up and correctly showing its very short message.
So, yeah, in reality, I'm 99% sure it really is an error message. That's only because I've seen similar error messages in the past and can infer how to interpret it.
We don’t teach creationism in school for the same reason we don’t teach the earth is flat: it’s a factually wrong, non-scientific idea. I don’t want someone telling my kids that the moon is made of cheese, nor do I want them lying that the earth is only 6,000 years old. That’s not censorship. That’s keeping science class scientific.
That begs the question by assuming that a belief system must be religious. Much of philosophy is about building a belief system that doesn’t depend on a religion at all.
For starters, “treat others well because everyone deserves our empathy” is completely compatible with atheism.
I took a continuing ed course on this recently, and it was horrid. You can do some pretty despicable stuff if you convince others that morality is fake and you can do whatever sounds fun. It’s like a cult of sociopathy.
Their general MO was to push a kid on the fringes to do something illegal, then blackmail them into upping the risk and illegality. They collected videos of all those “for lulz”. Various crimes included producing CSAM, torturing animals, and assaulting or killing random people.
It’s rare that I sit my kids down and have them watch training videos with me. With this, I did, to show them what’s out there and help them identify it if they ever came across it.
Oh if you’re talking about O9A and the like, those groups are absolutely real and also terrifying. They are like a worse version of Kiwi Farms. KF are actually saints compared to them, as KF seems to have a vague goal of bullying bad people (in their view) off the internet rather than pushing them to terrorism.
My problem with the NVE label is the scope creep, some would like to apply that label to anyone who has a strong opinion and doesn’t fit in a neat box.
(I also don’t think “nihilist” is the right label for these groups. They have beliefs, they just believe in evil.)
I could see that. It would be awfully easy to paint any group you disagree with that way. “These people don’t believe in my obviously correct and true moral framework? Nihilists. They marched in a protest? Violent extremists.”
> KF seems to have a vague goal of bullying bad people (in their view) off the internet
Has KF changed since I last checked? They had an explicit policy of "don't touch the poo"; that is, users mock and ridicule on the forum, but don't try to disturb the lolcows, unless they're committing crimes. That's what I saw; I lurked occasionally in the past. There's been a lot of misinformation about KiwiFarms since it became a minor news story, round about that time of the big DDoS attacks.
I think it’s more like you registered the car in their name. Now they’re allowed to use it, and also responsible for the thing which they didn’t want.
Consider that the “imposter” starts uploading child porn or something, and it’s on an account registered to your address. I think it’s perfectly A-OK to tell the service that it’s not me using the thing and I want them to close the account someone created in my name.
I’m a different person, but this happens to me, too. I have the kstrauser@yahoo.com email address because I signed up for it like 25 years ago. I log in every 6 months to see what the few other kstrausers in the world have signed me up for.
Not jsmith, but kstrauser. Not Gmail, but Yahoo. And I still get banking docs, and HOA meeting minutes, and birthday party invitations, and Facebook logins, and other bizarre random stuff.
I have so many questions. I’ve typoed my address before and had to correct it. That’s understandable. But to wholly invent one and say, yep, that looks good even though I’ve never used it before, I’m sure it’ll be fine! I just don’t get it.
I have a catch-all on a .com.au domain where there exists a later 1000+ people organisation with the equivalent .gov.au. I get what you described but from many, many people - divorce proceedings, legal discussions, financial documents, health things, etc.
Yeah I have josephg@gmail. The amount of spam that account gets is wild - about 50-100 emails hit the inbox per day. I got soft-locked out of google docs a few months ago because my google account's 25gb quota was exhausted.
Some of the emails are really unfortunate stuff. "Your account was added as a backup address." - Then inevitably, a few weeks later, dozens of password reset emails. Sorry bud. I've received pay stubs. Orders and invoices. I get phone bills every month for someone in India. Its chaos.
Early on I'd sometimes reply to these random emails telling people they've got the wrong address. The most astonishing reply I ever got was from HSBC bank telling me I needed to come into the branch to change my email address. Over the course of a week, I explained about 3 times that that was impossible. That I live in Australia. That I'm not their customer, and its not my account. Eventually they told me they were disabling online banking on my account. Now I've given up replying at all.
Send emails into that pit of PII misery if you want. I don't read them.
Some of these banks are ridiculous. HDFC bank insists that I send them my photo id, address, phone number, and my Indian id number to prove that I'm not their customer. I tried explaining that I don't have an Indian id number because I don't live in India but they insisted they can't help me unless I provide all of this. Then they sent me legal notices threatening me for not paying "my" bills. I send all their stuff to spam now.
I had one that person seemed to think their @twitter name was the same thing as my gmail address. Haven't seen it in a while, maybe they figured it out after I told their kid's teacher they had the wrong person...
I think you’re misreading this. OP has an email account. Someone else signed up for some website that doesn’t verify that you own the address before allowing you to log in and use the service. If the site did verify it, the user wouldn’t have been able to log in because OP would have been getting the verification emails, and not the user.
Later, after OP told the user and they failed to change their address, OP logged into the site and changed their password, putting an end to the spam they were receiving from the user’s actions.
I don’t have an ethical qualm with this. He didn’t want to sign up for the service. Someone else signed his email address up for it. Legally, I can’t imagine that being prosecutable.
One thing I've found, occasionally the hard way, is that helpful bystanders are always offering advice based on "ethical", "intuitive", "logical" and "common sense", usually without any aspect of "legal".
I got divorced a decade ago, and every well-wishing person in my life was strongly urging me to do things which were shockingly counter-productive / dangerous / wrong, based on their confident understanding (assumption, really) of the law which was completely and dangerously inaccurate.
Hacker News audience is global. People start accounts for various purposes. Yet people still freely share the notion that logging in to some unknown website run by an unknown company from a hard to spell country and then touching things is universally safe.
I miss the old "IANAL" tag which at least provided basic warning and self-awareness :-).
While true, I think that's implicit in all online conversations. I'm certain my thinking is 100% wrong in some jurisdictions elsewhere. Anything I say is wrong somewhere.
"It's OK: you can curse on the Internet." "Not when you're typing from Iran!" "Well, OK, if you're in Iran, don't take this American's advice for dealing with a government."
Part of our obligation as a reader is to consider what others are saying in the context of our own circumstances and experiences before trying to apply it. If you don't, and things end badly, that's on you.
But I stand on my words: I think it's ethically OK. You may not. That's alright. We're not required to have the same ethics or morals. And I don't think that's prosecutable. That's my opinion, based on my circumstances, not a statement of fact that applies in all jurisdictions around the world.
Above all else, I got tired of giving disclaimers about every single thing I say lest someone jump in with a "gotcha! scenario" I hadn't considered because it's not relevant to the context of the discussion.
IANYL, though! Offering legal advice with the disclaimer “I am not a lawyer” could be prosecuted as practicing law if a reasonably party could still infer a potential lawyer-client relationship from your message and/or intent. Instead, “I am not your lawyer” explicitly denies the lawyer-client relationship, which closes the door on both being accused of practicing law illegally and on being found as party to a lawyer-client relationship whether or not you have the appropriate certifications.
> closes the door on [...] being accused of practicing law illegally
Does it? So I can say, "I'm not your lawyer, but I'm happy to go ahead and give you specific legal advice on your case." and I can't be accused of illegally practicing law? I was under the impression that this could still get you into hot water. But not being your lawyer, due to the fact that I am not a lawyer at all, I don't know if it is true or not.
As with all things, who are you going to get in trouble with? And what's so magical about legal practice as opposed to, say, giving shitty medical advice or telling someone how to build porch? Asking genuinely. No one falls all over themselves to say "I am not a doctor, but...", even though their next words could kill someone. The implication is that they don't have formal training but they saw something on Facebook that you should try. What happens next is on you, not on them.
> No on falls all over themselves to say “I am not a doctor, but”
This is precisely why I’m pointing this out: IANAL is a very curious case of people self-labeling their statements as “not trustworthy for the topic”. I can think of perhaps no other cases where it is so popular to claim to not be a professional in the relevant field, which suggests that IANAL is a ‘badge of honor’ rather than a proper legal disclaimer. Certainly few (if any) claim IANAD before writing about their experiences with medical issues, body things, or nutritional supplements here, even though those topics are (as you correctly indicate) potentially lethal.
Thus, IANYL: if your goal is to ensure that the recipient of your advice / opinion / whatever does not have grounds to claim that you provided legal advice, and therefore are their lawyer, then you can either do so weakly with TINLA (“this is not legal advice”), which still leaves the door open for awkward claims by some desperate grifter-rando to reach a bench, or you can do so strongly with IANYL (“I am not your lawyer”), which closes that vulnerability in full.
Not once in years of using IANYL have I seen anyone else properly protect themselves from this vulnerability; meanwhile, “IANAL but” remains in use as a badge of honor. So, yeah, I don’t think anyone considers the particular avenue of vulnerability a serious threat, and yeah, the general context of IANAL here is prideful rather than protective. But after twenty years of dealing with a stalker who was adept at internet and tried to fuck with my job at one point, I do now tend to value closing off legal vulnerabilities with certainty, and as a bonus it doesn’t imply insult to the professions of law.
We already have mass surveillance, and yet we still have major crimes. It's not working, and I see no reason to believe that removing more freedom will lead to having safer streets. Why are we giving up liberty and getting nothing in return? That's an excellent reason to protest against adding more surveillance.
Our public surveillance is actually limited relative to other developed countries because it makes people here uncomfortable for cultural reasons. You’ll also note that our crime rates are pretty high, especially relative to the surveillance happy countries in East Asia.
Regardless, I’m happy to take a results oriented approach here. Does tracking license plates make it easier to catch criminals? Does it make it easier to track stolen vehicles? I suspect cities wouldn’t be signing these expensive contracts if they didn’t see any benefits.
And finally, surveillance of public spaces is not inherently at odds with personal freedoms. Your mobility is not restricted at all, your core rights have not been touched. And you are always welcome to go live in the woods off the grid.
I firmly believe that living in dense urban areas with millions of others requires a reasonably limited expectation of privacy in public spaces.
And for that matter, what’s it do with 9 digit numbers? Like, is it more accurate with them, or are these little guys mainly good at adding numbers with exactly 10 digits?
Basically, are the failures modes a gentle increase in inaccuracy, or spectacle failure outside their parameters?
reply