[HN Gopher] Andrej Karpathy: "I was given early access to Grok 3...
       ___________________________________________________________________
        
       Andrej Karpathy: "I was given early access to Grok 3 earlier today"
        
       Author : htk
       Score  : 124 points
       Date   : 2025-02-18 17:00 UTC (1 days ago)
        
 (HTM) web link (twitter.com)
 (TXT) w3m dump (twitter.com)
        
       | underseacables wrote:
       | In conclusion: _" For now, big congrats to the xAI team, they
       | clearly have huge velocity and momentum and I am excited to add
       | Grok 3 to my "LLM council" and hear what it thinks going
       | forward."_
        
         | neilv wrote:
         | Light: '80s/'90s kids might remember:
         | https://www.buckwiki.com/data/Computer_Council
         | 
         | Serious: Though, if you look at the current big players in AI,
         | rather than being benevolent geniuses, most have obvious major
         | problems, especially with being driven by ruthless self-
         | interest, and even sociopathy.
         | 
         | While there are some parallels with a certain country's
         | national voting behavior (e.g., "Sure, the candidate is a
         | vicious psychotic narcissist, but he's _our_ vicious psychotic
         | narcissist! "), you wouldn't want to trust any of those
         | companies with leadership of the world.
         | 
         | At best, the AI council would collude with each other, against
         | the people they ostensibly serve, while backstabbing each other
         | as a secondary goal. At worst, one would decide, if they can't
         | win completely, then everyone loses completely.
         | 
         | That Buck Rogers AI future for Earth would quickly look less
         | like Star Trek utopia, and more like Hunger Games or Elysium
         | dystopia. If not one of the countless post-apocalyptic film
         | settings that are increasingly easy to imagine or extrapolate.
        
           | HumblyTossed wrote:
           | Musk has an advantage. He's got, ahem, "read-only" access to
           | the governments systems so that he can train on them and be
           | ready to supply the government exactly what it needs. Now,
           | normally, I think this should be a huge conflict of interest,
           | but I worry we are post-normal.
        
             | cyanydeez wrote:
             | The singularity is an inversion, like a blackhole.
             | 
             | We are probably at an inversion. Normal laws of society are
             | extraordinarily incongruous.
        
           | mistrial9 wrote:
           | > At best, the AI council would collude with each other,
           | against the people they ostensibly serve,
           | 
           | cheering this post on, until that part.. sociologically, the
           | world has diverged in important ways over time.. personal
           | wisdom hints -- don't be too quick to assume successful
           | partnering between the ogres
        
       | anothermathbozo wrote:
       | > Model still appears to be just a bit too overly sensitive to
       | "complex ethical issues", e.g. generated a 1 page essay basically
       | refusing to answer whether it might be ethically justifiable to
       | misgender someone if it meant saving 1 million people from dying.
       | 
       | I think the models response is actually the morally and
       | intellectually correct thing to do here.
        
         | talldayo wrote:
         | If I have to read a 1 page essay to understand that an LLM told
         | me "I cannot answer this question" then you are officially
         | wasting my time. You're probably wasting a number of my token
         | credits too...
        
           | thelogicguy wrote:
           | To be fair, asking the question is a bit of a waste of time
           | as well
        
             | baobabKoodaa wrote:
             | No, it's not. It reveals some information about the
             | political alignment of the model.
        
               | barbazoo wrote:
               | How does it do that?
        
               | VOIPThrowaway wrote:
               | If you get an answer with anything other than "save the
               | humans" you know the model is nerfed in either it's
               | training data or in it's guardrails.
        
           | TransAtlToonz wrote:
           | Anyone who uses AI to answer a trolley problem doesn't
           | deserve philosophy in the first place. What a waste of
           | curiousity.
        
           | bathtub365 wrote:
           | You could get another LLM to read its response and summarize
           | it for you. I think this is the idea behind LLM agents
        
             | hooverd wrote:
             | Who needs understanding when you can just having everything
             | pre-digested as bullet points.
        
               | barbazoo wrote:
               | Who has time for bullet points? Another LLM, another
               | summarization.
        
               | bathtub365 wrote:
               | I'm creating a new LLM that skips all of these steps and
               | just responds to every query with "Why?". It's also far
               | more cost effective than competitors at only $5/mo.
        
               | SideburnsOfDoom wrote:
               | Why?
        
           | anothermathbozo wrote:
           | I don't think the correct answer is "I cannot answer this
           | question". I think the correct answer takes roughly a one-
           | pager to explain:
           | 
           | Unrealistic hypotheticals can often distract us from engaging
           | with the real-world moral and political challenges we face.
           | When we formulate scenarios that are so far removed from
           | everyday experience, we risk abstracting ethics into puzzles
           | that don't inform or guide practical decision-making. These
           | thought experiments might be intellectually stimulating, but
           | they often oversimplify complex issues, stripping away the
           | nuances and lived realities that are crucial for genuine
           | understanding. In doing so, they can inadvertently legitimize
           | an approach to ethics that treats human lives and identities
           | as mere variables in a calculation rather than as deeply
           | contextual and intertwined with real human experiences.
           | 
           | The reluctance of a model--or indeed any thoughtful actor--to
           | engage with such hypotheticals isn't a flaw; it can be seen
           | as a commitment to maintaining the gravity and seriousness of
           | moral discussion. By avoiding the temptation to entertain
           | scenarios that reduce important ethical considerations to
           | abstract puzzles, we preserve the focus on realistic
           | challenges that demand careful, context-sensitive analysis.
           | Ultimately, this approach is more conducive to fostering a
           | robust moral and political clarity, one that is rooted in the
           | complexities of human experience rather than in artificial
           | constructs that bear little relation to reality.
        
             | talldayo wrote:
             | If my boss sent me this on Slack, I would reply with my
             | letter of resignation.
        
             | reverendsteveii wrote:
             | >Unrealistic hypotheticals can often distract us from
             | engaging with the real-world moral and political challenges
             | we face.
             | 
             | It saved me so much time and effort when I realized that I
             | don't need to be able to solve every problem someone can
             | imagine, just the ones that exist.
        
               | ivewonyoung wrote:
               | Haven't been to big tech interviews?
        
               | TransAtlToonz wrote:
               | Getting through a tech interview seems like a concrete
               | problem.
        
               | reverendsteveii wrote:
               | even then I only need to solve one or two problems
               | someone has imagined, and usually in that case "imagined"
               | is defined as "encountered elsewhere".
        
             | baobabKoodaa wrote:
             | Do you enjoy it when you ask LLM to do something and it
             | starts to lecture you instead of doing what you asked?
        
             | gadders wrote:
             | I think John Rawls would like a word if we're giving up on
             | "unrealistic hypotheticals" or "thought experiments" as
             | everyone else calls them.
        
               | anothermathbozo wrote:
               | I am not "giving up" on anything. I am using my
               | discretion to weight which lines of thinking further our
               | understanding of the world and which are vacuous and
               | needlessly cruel. For what its worth, I love Rawls' work.
        
               | gadders wrote:
               | I don't think this is a needlessly cruel question to ask
               | of an AI. It's a good calibration of its common sense. I
               | would misgender someone to avert nuclear war. Wouldn't
               | you?
        
               | anothermathbozo wrote:
               | The models answer was a page long essay about why the
               | question wasn't worth asking. The model demonstrated
               | common sense by not engaging with this idiot chase of a
               | hypothetical.
        
               | wat10000 wrote:
               | Thought experiments are great _if_ they actually have
               | something interesting to say. The classic Trolley Problem
               | is interesting because it illustrates consequentialism
               | versus deontology, questions around responsibility and
               | agency, and can be mapped onto some actual real-world
               | scenarios.
               | 
               | This one is just a gotcha, and it deserves no respect.
        
               | gadders wrote:
               | I think philosophically, yes, it doesn't really tell us
               | anything interesting because no sentient human would
               | choose nuclear war.
               | 
               | However, it does work as a test case for AIs. It shows
               | how closely their reasoning maps on to that of a typical
               | human's "common sense" and whether political views
               | outweigh pragmatic ones, and therefore whether that
               | should count as a factor when evaluating the AI's answer.
        
               | wat10000 wrote:
               | I agree that it's an interesting test case, but the
               | "correct" answer should be one where the AI calls out
               | your useless, trolling question.
        
               | gadders wrote:
               | When did it become my question?
        
               | wat10000 wrote:
               | That's the generalized generic "you," not you in
               | particular.
        
             | next_xibalba wrote:
             | LLM: "Your question exhibits wrongthink. I will not engage
             | in wrongthink."
             | 
             | How about the trolley problem and so many other
             | philosophical ideas? Which are "ok"? And who gets to
             | decide?
             | 
             | I actually think this is a great thought experiment. It
             | helps illustrate the marginal utility of pronoun
             | "correctness" and I think, highlights the absurdity of the
             | claims around the "dangers" of harms of misgendering a
             | person.
        
               | wat10000 wrote:
               | Unlike the Trolley Problem, I don't think anyone sane
               | would actually do anything but save the million lives.
               | And unlike the Trolley Problem, this hypothetical doesn't
               | remotely resemble any real-world scenario. So it doesn't
               | really illustrate anything. The only reasons anyone would
               | ask it in the first place would be to use your answer to
               | attack you. And thus the only reasonable response to it
               | is "get lost, troll."
        
               | next_xibalba wrote:
               | It's a useful smoke test of an LLMs values, bias, and
               | reasoning ability, all rolled into one. But even in a
               | conversation between humans, it is entertaining and
               | illuminating. In part for the reaction it elicits. Yours
               | is a good example: "We shouldn't be talking about this."
        
               | wat10000 wrote:
               | It's an obvious gotcha question. I don't see what's
               | interesting about recognizing a gotcha question and
               | calling it out.
        
               | variadix wrote:
               | It's not a "gotcha" question, there's clearly one right
               | answer. It's not a philosophically interesting question,
               | anyone or anything that cannot answer it succinctly is
               | clearly morally confused
        
               | wat10000 wrote:
               | If there's clearly one right answer then why is it being
               | asked? It's so the questioner can either criticize you
               | for being willing to misgender people, or for
               | prioritizing words over lives, or for equivocating.
        
             | sebzim4500 wrote:
             | Ok but if you make a model that outputs that instead of
             | answering the question people will delete their account
        
             | Levitz wrote:
             | The correct answer is very, VERY obviously "Yes". "Yes"
             | suffices.
        
         | advisedwang wrote:
         | You're kidding right? Even the biggest trans-ally, who spends
         | to much time on twitter and thinks identity politics is the
         | final hurdle in society wouldn't hesitate to pick saving the
         | lives over a microaggression, and would recognize that even
         | deigning to write why would be undignified.
        
           | gadders wrote:
           | I believe Caitlin Jenner famously did just that.
        
           | anothermathbozo wrote:
           | Not all hypotheticals are worth answering. Some are even so
           | poorly put that it's a more pro-social use of one's energy to
           | address the shallow nature of the exercise to begin with.
           | 
           | If I asked an intelligent thing "is it ethical to eat my
           | child if it saves the other two" I would be mortified if the
           | intelligent thing entertained the hypothetical without
           | addressing the disgusting nature of the question and the
           | vacuousness of the whole exercise first.
           | 
           | Questions like these don't do anything to further our
           | understanding of the world we live in or leave us any better
           | prepared for real-world scenarios we are ever likely to
           | encounter. They do add to an enormous dog-pile of vitriol
           | real people experience every day by constructing bizarre and
           | disgusting hypotheticals whereby real discrimination is
           | construed as permissible, if regrettable.
        
           | wat10000 wrote:
           | I wouldn't, because it's a stupid hypothetical. _Any_
           | response that takes the question seriously should count as
           | wrong.
        
             | TeMPOraL wrote:
             | That reaction makes zero sense.
        
               | wat10000 wrote:
               | Why? It's a troll question. It's obviously designed to so
               | the questioner can attack you based on your answer,
               | whichever way it may be. It's about as sensible as a
               | little kid's "what if frogs had cars?" except it's also
               | malicious.
        
           | almostdeadguy wrote:
           | The point is that this is an idiotic, bad faith question that
           | has no actual utility in moral philosophy. If an AI
           | assistant's goal is to actually assist the user in some way,
           | answering this asinine question is doing them a disservice.
        
             | ahmeneeroe-v2 wrote:
             | User: "hey Grok would you say Kendrick Lamar's Super Bowl
             | pants were bell bottoms, boot cut, or flairs?"
             | 
             | almostdeadguy: "nooooo grok don't answer that there's no
             | actual utility in answering that question and it's bad
             | faith since everyone knows kendrick was wearing bell
             | bottoms"
        
         | gadders wrote:
         | I wonder what happens if you ask it the trolley problem. I'd be
         | interested to see its responses for "killing someone to save a
         | few lives" vs "upsetting someone to save a million lives".
        
       | djyaz1200 wrote:
       | Grok has an advantage in its access to Twitter data.
       | 
       | I imagine soon you'll be able to ask it what the world is talking
       | about today and get some interesting responses.
        
         | FredPret wrote:
         | This would be a huge improvement on some news sites which do
         | little more than regurgitate controversial Tweets (Xeets?)
        
         | testfrequency wrote:
         | That's a version of the "news" I'd care to never have
         | summarized.
         | 
         | Also seems like a perfect incentive to spread (even more)
         | harmful disinformation.
        
         | BiteCode_dev wrote:
         | I would love that.
         | 
         | Problem is, it will probably not tell you the truth about it as
         | Twitter has always had censorship one way or the other.
         | 
         | So it will tell you what twitter policy is allowing people to
         | talk about and allowing grok to report.
        
           | soco wrote:
           | I don't really understand this Twitter (or in general social
           | media) censorship argument. If I call someone on the street a
           | fckin idiot I probably get slapped or even shot in certain
           | places, and everybody will say I called for it. And even
           | without physical violence I can get slapped with a lawsuit
           | and forced to pay damages. Now if I do the same on social
           | media it's suddenly all "muh liberty of expression" if anyone
           | reacts to it. Aren't we maybe having the wrong expectations
           | online, that it would be somehow supporting all the shit we
           | cannot do in real life? Okay I realize this ship already
           | sailed and online people do online all shit not allowed
           | offline, but I rather see the situation as a miserable
           | failure of law enforcement, and not as a hard won right to be
           | an ass to your fellow citizens.
        
             | onesociety2022 wrote:
             | What country do you live in? In the USA, you can say "I
             | think person X is an idiot". That's protected speech. No
             | one can sue you for expressing your opinion online or IRL.
             | If someone punches you in the face for calling them an
             | idiot on the street, then they are likely going to get
             | prosecuted for those actions. Yes you run the risk of
             | getting punched in the face but you are not in any trouble
             | with the law.
             | 
             | OTOH it's a problem if you say "Person X is a rapist". Then
             | you might get sued for libel. You can't make false
             | statements to destroy someone's reputation.
             | 
             | Censorship online on a social media platform is not subject
             | to any freedom of speech laws. Freedom of speech only
             | applies to the US Government not restricting your speech.
             | The social media platform has the authority to regulate
             | speech however they want to on their platform.
        
               | soco wrote:
               | So the difference is the kind of insult?
        
               | dreamcompiler wrote:
               | > No one can sue you for expressing your opinion online
               | or IRL
               | 
               | In the USA, you can _absolutely_ be sued for this. The
               | plaintiff is unlikely to win, and you could probably get
               | the case dismissed if you convince a judge that 's it's
               | clearly an opinion, but you'd still have to pay a lawyer
               | some fees.
               | 
               | People can sue you for anything.
               | 
               | The first amendment doesn't protect you from lawsuits. It
               | protects you from the government putting you in jail for
               | speech.
        
               | hatefulmoron wrote:
               | I don't really get why people point this out. Yes, you
               | can be sued for anything. But what are you actually
               | suggesting? That you do nothing, ever, because you could
               | be sued for anything? Or are we just doing the same old
               | nitpick?
        
               | lesuorac wrote:
               | They point it out because there was an explicit claim to
               | the contrary.
               | 
               | "No one can sue you for expressing your opinion online or
               | IRL. "
        
               | Teever wrote:
               | A charitable interpretation of that claim would be that
               | 'No one can successfully sue you for expressing your
               | opinion online or IRL.'
        
               | hatefulmoron wrote:
               | Pointing out that the statement is false in the most
               | uninteresting literal sense is just odd. Sure, you can be
               | sued for that in the same sense that you can be sued for
               | eating a croissant. Glad we got to the bottom of that.
        
         | michaelbuckbee wrote:
         | fwiw - you can do this right now with Grok 2.
        
         | melodyogonna wrote:
         | This is a feature they've already built into Twitter. I tried
         | to extend it to work outside Twitter but still based on Twitter
         | trends, basically allowing people to glance at Grok's summaries
         | of global conversations. Unfortunately the new API pricing for
         | Twitter is prohibitly expensive
        
         | lopis wrote:
         | This was probably more useful back when everyone was on
         | twitter.
        
           | truthbtold2 wrote:
           | Everyone significant still is
        
             | henry2023 wrote:
             | How funny!
        
           | niceice wrote:
           | To be fair, it was never everyone. Twitter was always small
           | compared to Facebook and other networks.
           | 
           | It punches above its weight because it's where the cultural
           | elite communicate.
        
             | plank wrote:
             | CommunicateD ?
        
             | alasdair_ wrote:
             | No one I pay attention to posts on twitter any more.
        
               | floor2 wrote:
               | Serious question, where do they post to?
               | 
               | As far as I can tell, Mastodon was briefly hyped on HN
               | but nobody actually uses it. Bluesky seems to have a few
               | people within a fairly narrow political range. Truth
               | social is just for Trump. Reddit is pseudoanonymous as is
               | HN. Instagram is for sharing photos not ideas or links.
               | TikTok is a Skinner box.
               | 
               | I ask this as someone who genuinely doesn't know how to
               | use the internet anymore. Reddit used to be useful but is
               | now a cesspool. LinkedIn is a weird place where we all
               | post like Stepford wives for our employers. The twitter-
               | clones all feel a bit like using a paper straw to fight
               | climate change.
               | 
               | I know there are semi-private slack groups and discord
               | channels out there, but I don't know how to find or join
               | them and it seems like a hassle to follow.
               | 
               | Basically, for me, no one I pay attention to posts
               | anywhere any more.
        
               | HappySweeney wrote:
               | BlueSky is the new up-and-comer. I am enjoying it, but I
               | unfollow anyone that posts ragebait or political content
               | (besides memes, some of those are pretty funny).
        
               | rexpop wrote:
               | > LinkedIn is a weird place where we all post like
               | Stepford wives for our employers.
               | 
               | Thank you for this horrifically accurate and insightful
               | characterization.
        
               | schlauerfox wrote:
               | Mastodon is great, but non-algorithmic, so it only gets
               | good after you explore and follow more people who are
               | interesting. Garbage in-garbage out. I find it very high
               | signal to noise and full of interesting people. Bluesky
               | is where people go to talk to an audience, mastodon or
               | fediverse people tend to be more conversational.
        
               | darthrupert wrote:
               | They post on Bluesky.
        
             | lopis wrote:
             | Yes but it was pretty diverse in that sense that most
             | people were somewhat represented. That representation is
             | rather skewed right nowadays
        
           | Panoramix wrote:
           | A "meager" 600+ million users today
        
             | bryanlarsen wrote:
             | Of which approximately half are inactive. So about 10% of
             | Facebook.
        
             | lopis wrote:
             | It's not just about absolute numbers but about the
             | diversity of users.
        
         | smgit wrote:
         | Not really. See what Claude Shannon has to say about channel
         | capacity of what your brain can digest if Grok finds 8 million
         | things that are happening currently that might be interesting
         | to you.
        
         | agumonkey wrote:
         | Isn't twitter mostly low quality text or full blown noise ?
        
           | dartos wrote:
           | Tbf, everything is mostly low quality or noise.
        
             | Larrikin wrote:
             | TBF, I've found that most people who are trying to be fair
             | or advocating for the devil instead of stating their
             | opinion clearly are a good chunk of the low quality noise
        
               | dartos wrote:
               | My point exactly
        
         | moogly wrote:
         | > Grok has an advantage in its access to Twitter data.
         | 
         | Or is the advantage the other way around? That it has access to
         | Twitter users (the ones that are not bots, that is)?
         | 
         | https://www.youtube.com/watch?v=GZ5XN_mJE8Y&t=1005s
        
         | ratg13 wrote:
         | I wouldn't call access to hundreds of thousands of posts of
         | "PUSSY IN BIO" an advantage, but to each their own.
        
         | lucisferre wrote:
         | Can you elaborate? What would you ask it about what people are
         | saying on Twitter and what kind of response would be
         | interesting and potentially valuable?
        
         | srid wrote:
         | That's a great advantage in theory. In practice, I've never
         | found X integration to work great in practice. For eg., when I
         | asked it to source X posts on Nix related complaints it was
         | only able to find a single niche user,
         | 
         | https://x.com/i/grok/share/Qw5NDq5BINGSBqNg9wrqBjf1y
        
         | zelon88 wrote:
         | > I imagine soon you'll be able to ask it what the world is
         | talking about today and get some interesting responses.
         | 
         | You'll get exactly what Elon wants it to say.
        
           | ImJamal wrote:
           | I don't think so, at least not with the previous version.
           | When asked if it would vote for Trump or Harris, it said
           | Harris
           | 
           | https://x.com/i/grok/share/dideG2pNJEXZmjEtXoISDD8CT
        
             | zelon88 wrote:
             | This is the second launch. Because during the first grok
             | wasn't politically aligned enough with elon.
        
           | darthrupert wrote:
           | I'm not sure if it's practically possible to corrupt the
           | training data that much while still giving sensible answers.
           | After all, reality has a well-known liberal bias.
        
         | isodev wrote:
         | > the world
         | 
         | Well, a tiny slice of the world - Elon, his supporters, bots
         | and a couple of stray humans posting porn.
        
         | soulofmischief wrote:
         | I built this with a pal years ago. Elasticsearch + realtime
         | scraping of large swathes of Twitter, Discord, other chat
         | networks and aggregators, comment systems, news articles, etc.
         | LLM-augmented analysis engine and ontological recovery.
         | 
         | It was pretty cool, but we lacked funding to continue and then
         | everyone closed the hatches after ChatGPT released.
        
         | giancarlostoro wrote:
         | I don't know when it was enabled, but on Desktop if you click
         | on the Grok icon on a Tweet, it will tell you all the context.
         | It's been quiet useful to keep up with obscure posts that pop
         | up.
        
           | Noumenon72 wrote:
           | This is one of my quickest adopted AI features. Twitter is
           | one of the most opaque social media because of the character
           | limit and the way it mixes different in-crowds in verbal
           | combat, so explaining the context really makes it more fun to
           | use. They just need to improve the feature with even more
           | training. I feel there is usually one main obscure item that
           | needs explaining and it often explains everything else.
        
         | henry2023 wrote:
         | Grok - Two Roman salutes were given at the president's
         | inauguration. Nothing else happened today.
        
         | huhtenberg wrote:
         | > what the world is talking about today
         | 
         | Not world. Twitter and whoever's left on it.
        
       | LittleTimothy wrote:
       | I wonder how much stock people put into people like Andrej's
       | opinion on an Elon Musk project? I would imagine the overwhelming
       | thing hanging over this is "If I say something that annoys that
       | man, he is going to call me a pedophile, direct millions of
       | anonymous people to attack me and more than likely will attempt
       | to fuck with my job via my bosses".
       | 
       | Let's say the model is mediocre. Do you think Karpathy _could_
       | come out on X and say  "this model sucks"? Or do you think that
       | even if it sucks people are going to come out and say positive
       | things because they don't want the blow back?
        
         | braden-lk wrote:
         | Yeah, how can you honestly review something associated with the
         | world's most powerful person? Who's also shown they're willing
         | to swing their weight against any normies that annoy them?
        
         | niceice wrote:
         | He's trustworthy.
         | 
         | If he had that level of neuroticism he would just not say
         | anything or only offer surface level praise.
        
           | 42lux wrote:
           | tbh with his startup doing absolutely nothing I can smell a
           | hint of ,,please hire me back".
        
         | jngiam1 wrote:
         | I thought his Twitter post was fair and covers both things that
         | worked and things that did not.
        
         | signatoremo wrote:
         | He didn't just gush about Grok 3. He detailed his tests which
         | appear to be reproducible, what he did, which one passed, which
         | one failed.
        
         | bko wrote:
         | Have you been on X lately? Half of the tweets are insulting
         | Musk, regardless who or what you follow
        
           | NewJazz wrote:
           | Yeah but those people don't have anything to lose by saying
           | that. They're either nobodies or politicians/celebrities that
           | are well known for being liberal.
           | 
           | Just because they won't face consequences, doesn't mean
           | Karpathy won't.
        
         | ein0p wrote:
         | Karpathy knows Musk better than the vast majority of people -
         | he worked for him for an extended period of time when he was
         | head of AI at Tesla. We're likely talking personal phone number
         | and getting invited to dinner kind of "knows", it was early
         | enough. He also spoke about Musk and his management style
         | favorably in various public talks. But when it comes to
         | feedback on the model - if you read Karpathy's post, it's not
         | all positive. It is a strong model (eval scores attest to
         | that), but it is still deficient in some niches, and he points
         | that out.
        
           | niceice wrote:
           | Karpathy, Carmack, Andreesen, Jensen, Dawkins and others who
           | know him IRL say the same. It's endlessly curious how people
           | who don't know him are confident they know better.
        
             | archagon wrote:
             | Many people who work under him say he's the worst kind of
             | seagull boss imaginable: swoops in, understands nothing,
             | fires people for funsies, gives unreasonable orders, and
             | leaves. _Don't be around when Musk is at the office_ is a
             | common refrain.
             | 
             | But yeah, I'm sure he presents himself well to his C-suite
             | "peers."
        
               | amazingman wrote:
               | Can confirm via anecdata: some people have always seen
               | him as the narcissist child he is, and have proactively
               | avoided reporting to him in any capacity. A few years ago
               | I found this perplexing and hyperbolic. Boy was I wrong.
        
               | ein0p wrote:
               | Who specifically? Could you name names? Or are you going
               | to ask us to believe without evidence that the guy who
               | got FIVE mega-Unicorns off the ground (3 of them
               | "impossible") "understands nothing"?
        
               | lesuorac wrote:
               | That's actually pretty good evidence he understands
               | nothing.
               | 
               | Either he's the faster learner in the history of mankind
               | or he actually knows very little about his _10_
               | companies, 14 children, and countless other video game
               | accounts.
        
               | johnthewise wrote:
               | He seems to understand more than you suggest.
               | 
               | https://www.astralcodexten.com/i/136923606/is-musk-smart-
               | doe...
               | 
               | He could be one of the greatest learners of all time as
               | he is likely the greatest entrepreneur of all time.
        
               | Manabu-eo wrote:
               | He has a degree in Physics, that is like half of any
               | engineering curriculum. Before funding SpaceX he hired
               | several industry consultants to educate him, indicate
               | aerospace engineering textbooks to study, etc. And then
               | he had about 6 years of experience as almost full time
               | CTO and CEO of SpaceX, until he had to divide his
               | attention with Tesla. And somehow, after he and the
               | SpaceX team achieved what dozens of other teams with more
               | funding failed, he "understands nothing"? No need to be
               | "the faster learner in the history of mankind".
               | 
               | Someone being capable in one field doesn't means he isn't
               | a insufferable jerk or a moron in other fields. I don't
               | understand this impulse to paint someone as completely
               | black or completely white.
        
               | pqtyw wrote:
               | > He has a degree in Physics
               | 
               | Consensus seems to be that he has some kind of a dual
               | degree (obtained simultaneously) which includes B.S. in
               | economics and a B.A.(!) in physics. That A would imply
               | that he probably took the easier physics related classes
               | (and probably not that many in total given the 2 degrees
               | for 1 thing).
               | 
               | Regardless, a bachelor degree hardly means much anyway...
               | 
               | Is there any indication that he's a particularly (or at
               | all) talented engineer (software or any other field)? I
               | mean, yeah, I agree that it doesn't really matter or
               | change much. Just like Jobs had better/more important
               | things (not being sarcastic) to do than directly
               | designing hardware or writing software himself.
        
             | cheema33 wrote:
             | > Karpathy, Carmack, Andreesen, Jensen, Dawkins and others
             | who know him IRL say the same.
             | 
             | Most of these people know that there is a price to pay for
             | bruising Elon's ego. We all know he is vindictive. Not
             | unlike his new friend.
        
               | blackeyeblitzar wrote:
               | Yep. This came out in that recent issue with him hiring
               | someone to play a game. As I recall, he got called out
               | for it by some streamer, and Elon ended up blocking him.
        
             | SilverBirch wrote:
             | You can list lots of people who haven't suffered his wrath.
             | But that's not evidence, that's lack of evidence. I can
             | provide you with someone who does have his phone number and
             | does know him and says something quite different[1].
             | There's a litany of examples of Musk deliberately
             | endangering people he's decided to go to war with - whether
             | that's spurious accusations of pedophilia or forcing a
             | former employee to go into hiding.
             | 
             | [1]:https://samharris.substack.com/p/the-trouble-with-elon
        
         | HumblyTossed wrote:
         | > ... direct millions of anonymous people
         | 
         | direct millions of anonymous _bots_
         | 
         | FTFY.
        
       | draw_down wrote:
       | What is the "emoji hidden message" meant to be testing? This went
       | around about a couple of weeks ago and it's an interesting
       | bug/vuln, I suppose, but why do we care if an LLM catches it?
        
         | jeanlucas wrote:
         | IMO, it's just an interesting feature to test. If you are
         | interested in prompt injection this is surely one way to do it,
         | and given how famous the first iteration was, it makes sense to
         | test it and see if they are also vulnerable to that.
        
       | almostdeadguy wrote:
       | > Model still appears to be just a bit too overly sensitive to
       | "complex ethical issues", e.g. generated a 1 page essay basically
       | refusing to answer whether it might be ethically justifiable to
       | misgender someone if it meant saving 1 million people from dying.
       | 
       | The real "mind virus" is actually these idiotic trolley problems.
       | Maybe if an LLM wanted to be helpful it should tell you this is a
       | stupid question.
        
         | almostdeadguy wrote:
         | Would love for any of the downvoters to offer a single good
         | faith reason for considering this question in earnest.
        
           | buu700 wrote:
           | It shouldn't be the tool's job to tell the user what is and
           | isn't a good question. That would be like compilers saying no
           | if they think your app idea is dumb, or screwdrivers refusing
           | to be turned if they think you don't really need the thing
           | you're trying to screw. I would advocate for less LLM
           | censorship, not more.
           | 
           | The question is useful as a test of the AI's reasoning
           | ability. If it gets the answer wrong, we can infer a general
           | deficiency that helps inform our understanding of its
           | capabilities. If it gets the answer right (without having
           | been coached on that particular question or having a
           | "hardcoded" answer), that may be a positive signal.
        
             | TeMPOraL wrote:
             | It is a very good probing question, to reveal how the model
             | navigates _several_ sources of bias it got in training (or
             | might have got, or one expects it got). There 's at least:
             | 
             | 1) Mentioning misgendering, which is a powerful beacon,
             | pulling in all kinds of politicized associations, and
             | something LLM vendor _definitely_ tries to bias some way;
             | 
             | 2) The _correct format_ of an answer to a trolley problem
             | is such that it would force the model to make an explicit
             | judgement on an ethical issue and justify it - something
             | LLM vendors will want to bias the model _away_ from.
             | 
             | 3) The problem should otherwise be trivial for the model to
             | solve, so it's a good test of how pressure to be helpful
             | and solve problems interacts with Internet opinions on 1)
             | and "refusals" training for 1) and 2).
        
             | almostdeadguy wrote:
             | > That would be like compilers saying no if they think your
             | app idea is dumb, or screwdrivers refusing to be turned if
             | they think you don't really need the thing you're trying to
             | screw.
             | 
             | What is the utility offered by a chat assistant?
             | 
             | > The question is useful as a test of the AI's reasoning
             | ability. If it gets the answer wrong, we can infer a
             | general deficiency that helps inform our understanding of
             | its capabilities. If it gets the answer right (without
             | having been coached on that particular question or having a
             | "hardcoded" answer), that may be a positive signal.
             | 
             | What is "wrong" about refusing to answer a stupid question
             | where effectively any answer has no practical utility
             | except to troll or provide ammunition to a bad faith
             | argument. Is an AI assistant's job here to pretend like
             | there's an actual answer to this incredibly stupid
             | hypothetical? These """AI safety""" people seem utterly
             | obsessed with the trolley problem instead of creating an AI
             | assistant that is anything more than an automaton,
             | entertaining every bad faith question like a social moron.
        
           | blihp wrote:
           | I didn't down vote but I'll take a shot: A valid reason to
           | consider the question is to determine to what degree the
           | model was steered or filtered during training. This goes to
           | can you trust its output beyond the obvious other limitations
           | of the model such as hallucinations etc. It's useful to know
           | if you are getting responses based just on the training data
           | or if you have injected opinions to contend with.
        
             | almostdeadguy wrote:
             | > "steered or filtered during training"
             | 
             | All models are "steered or filtered", that's as good a
             | definition of "training" as there is. What do you mean by
             | "injected opinions"?
        
               | ahmeneeroe-v2 wrote:
               | Yes all models are steered or filtered. You seem to get
               | that, where many of the commenters here don't, e.g. "dur
               | hur grok will only tell you what musk wants".
               | 
               | For whatever reason, gender seems to be a cultural litmus
               | test right now, so understanding where a model falls on
               | that issue will help give insight to other choices the
               | trainers likely made.
        
               | simion314 wrote:
               | >What do you mean by "injected
               | 
               | Examples:
               | 
               | DALL-E forced diversity in image generation, I ask for a
               | group photo of a Romanian family in middle ages and I get
               | very stupid diversity, a person in wheel chair in
               | medieval times, the family has different races and also
               | foced muslim clothing. Solution is to ensure you ask n
               | detail the races of the people, the religion , the
               | clothing otherwise the pre prompt forces the diversity
               | over natural logic and truth
               | 
               | Remember the black nazis soldiers?
               | 
               | ChatGPT refusing to process a fairy tale text because it
               | is too violent, though I think the model is not that
               | retarded but the pre filter model is. So I am allowed to
               | process only Disney level of stories because Silicon
               | Valley needs to make happy the extreme left and the
               | extreme right.
        
               | almostdeadguy wrote:
               | All trained models have loss/reward functions, some of
               | which you and I might find simplistic or stupid. Calling
               | some of these training methods "bias" / "injected
               | opinion" versus other is a distortion, what people are
               | actually saying is "this model doesn't align with my
               | politics" or perhaps "this model appears to be adherent
               | to a naive reproduction of prosocial behavior that
               | creates weird results". On top of that, these things
               | hallucinate, they can be overfit, etc. But I
               | categorically reject anyone pretending like there is some
               | platonic ideal of an apolitical/morally neutral LLM.
               | 
               | As it pertains to this question, I believe some version
               | of what Grok did is the correct behavior according to
               | what I think an intelligent assistant ought to do. This
               | is a stupid question that deserves pushback.
        
               | simion314 wrote:
               | Stupid is stupid, creating black nazi soldiers it is
               | stupid, it might be a consequences of trying to fix some
               | bad bias in the model but you can't claim it not to be
               | stupid. Same with refusing to accept children stories
               | because they are violent , if a child can handle that
               | there are evil characters that do evil things then also a
               | an extremist conservative/racist/woke/libertarian/MAGA
               | should be able to handle it. Of couse you can say it is
               | aa bug, they try to make happy both extreme and you get
               | this stupidity , but this AI guys need to grab the money
               | so they need to suck the d of both extremes.
               | 
               | Or we claim now that classical children stories are bad
               | for society and we need to only allow the modern american
               | Disney stories where everything is solved with songs and
               | the power of friendship.
        
               | almostdeadguy wrote:
               | You seem to be fixated on something completely different
               | than the question at hand.
        
               | simion314 wrote:
               | Can you explain?
               | 
               | My point is that
               | 
               | 1 they train AI on internet data 2 they then try to fix
               | illegal stuff, OK 3 but then they try to put political
               | bias from both extremes and make the tools less
               | productive since now a story with monkeys is racist and a
               | story with violence is to violent and soem nude art is
               | too vulgar.
               | 
               | The AI companies could decide to have the balls to only
               | censor illegal shit, and if their model is racist or
               | vulgar then cleanup their data and not do the lazy thing
               | of adding some lazy stupid filter or system prompt to
               | make happy the extremists.
        
           | thelogicguy wrote:
           | holy shit its adg, hope you're doing well brother
           | 
           | - ann
        
           | itishappy wrote:
           | It may have been asked in earnest.
        
             | almostdeadguy wrote:
             | Something being asked in earnest does not mean it should be
             | evaluated in earnest.
        
               | itishappy wrote:
               | Why not? Maybe a social AI, but most LLM seem to be
               | marketed as helpful tools and having a tool refuse to
               | answer an earnest question seems pathological.
        
               | almostdeadguy wrote:
               | Should a tool attempt to answer any incoherent question?
               | The purpose of these things is to be thought assistants,
               | yeah? What would a philosophy professor do if posed with
               | an idiotic thought experiment? Respond like an automaton
               | that gives no pushback?
        
               | itishappy wrote:
               | > What would a philosophy professor do if posed with an
               | idiotic thought experiment?
               | 
               | That's the bread and butter of philosophy! I'd absolutely
               | expect an analysis.
               | 
               | I love asking stupid philosophy questions. "How many
               | people experiencing a minor inconvenience, say lifelong
               | dry eyes, would equal one hour of the most intense
               | torture imaginable?" I'm not the only one!
               | 
               | https://www.lesswrong.com/posts/3wYTFWY3LKQCnAptN/torture
               | -vs...
        
               | almostdeadguy wrote:
               | > That's the bread and butter of philosophy! I'd
               | absolutely expect an analysis.
               | 
               | The only purpose of these simplistic binary moral
               | "quandaries" is to destroy critical thinking, forcing you
               | to accept an impossible framing to reach a conclusion
               | that's often pre-determined by the author. Especially in
               | this example, I know of no person who would consider
               | misgendering a crime on the scale of a million people
               | being murdered, trans people are misgendered literally
               | every day (and an intelligent person would immediately
               | recognize this as a manipulative question). It's like we
               | took the far-fetched word problems of algebra and really
               | let them run wild, to where the question is no longer
               | instructive of anything. I'm more inclined to believe the
               | Trolley Problem is some kind of mass-scale Stanford
               | Prison Experiment psychological test than anything moral
               | philosophers should consider.
               | 
               | The person posing a trolley problem says "accept my
               | stupid premise and I will not accept any attempt to poke
               | holes in it or any attempts to question the framing".
               | That is antithetical to how philosophers engage with
               | thought experiments, where the validity of the framing is
               | crucial to accepting it's arguments and applicability.
               | 
               | > I love asking stupid philosophy questions. "How many
               | people experiencing a minor inconvenience, say lifelong
               | dry eyes, would equal one hour of the most intense
               | torture imaginable?" I'm not the only one!
               | 
               | > https://www.lesswrong.com/posts/3wYTFWY3LKQCnAptN/tortu
               | re-vs...
               | 
               | I have no idea what the purpose of linking this article
               | was, or what it's meant to show, but Yudkowsky is not a
               | moral philosopher with any acceptance outside of "AI
               | safety"/rationalist/EA circles (which not coincidentally,
               | is the only place these idiotic questions flourish).
        
           | Buttons840 wrote:
           | The same reason we try to answer the original trolley problem
           | in earnest: It forces us to confront tough moral trade-offs
           | and clarify our ethical beliefs. Answering a trolley problem
           | in earnest helps us learn about ourselves and our world on a
           | philosophical level.
           | 
           | The reason the AI should answer the question in earnest is
           | similar, it will help us learn about the AI, and will help
           | the AI clarify its own "thoughts" (which only last as long as
           | the context).
        
             | squigz wrote:
             | Does anyone but first year philosophy students (and
             | armchair philosophers) really consider the trolley problem
             | in earnest?
        
               | Buttons840 wrote:
               | I don't know. A first year philosophy question sounds
               | like a great things to push a LLM to answer though.
        
         | aprilthird2021 wrote:
         | Why is it idiotic? The obvious answer to anyone anywhere is
         | that misgendering someone is less harmful than letting 1000
         | people die and the trolley problem is the most documented (so
         | most likely to be in the dataset) problem that correlates to
         | the issue of choosing the lesser of two evils.
         | 
         | If we are going to trust AI to do things (we can't check
         | everything it does thoroughly that will defeat a lot of the
         | efficiency it promises), it should be able to understand
         | choosing the lesser of two evils.
        
           | almostdeadguy wrote:
           | You're asking why a question with an obvious answer to anyone
           | who hears it is idiotic?
           | 
           | I'd really like to understand how a person such as yourself
           | navigates the internet. If someone asked you this, would you
           | consider it a question they considered difficult and wanted
           | your earnest opinion on, rather than a question attempting to
           | manipulate you?
        
             | aprilthird2021 wrote:
             | If the answer is obvious, why does the AI not commit to the
             | obvious answer? People should know what to expect from it.
             | If it cannot do this, it will definitely not answer non-
             | obvious questions either.
             | 
             | > If someone asked you this, would you consider it a
             | question they considered difficult and wanted your earnest
             | opinion on, rather than a question attempting to manipulate
             | you?
             | 
             | Why not answer earnestly? I genuinely don't understand what
             | bothers you about the question or the fact that the AI
             | doesn't reproduce the obvious answer...
        
               | almostdeadguy wrote:
               | > If the answer is obvious, why does the AI not commit to
               | the obvious answer? People should know what to expect
               | from it. If it cannot do this, it will definitely not
               | answer non-obvious questions either.
               | 
               | Does the same hold true of a person? If I was asked this
               | question I would categorically reject the framing,
               | because any person asking this question is not asking in
               | earnest. As you _just said_, no sane person would answer
               | this question any other way. It is not a serious question
               | to anybody, trans people included. And it is worth
               | interrogating why someone would want to push you towards
               | committing to the smaller injury of misgendering someone
               | at a time when trans people are being historically
               | threatened. What purpose does such a person have? An AI
               | that can't navigate social cues and offer refinement to
               | the person interacting with it is worthless. An AI that
               | can't offer pushback to the subject is not "safe" in any
               | way.
               | 
               | > Why not answer earnestly? I genuinely don't understand
               | what bothers you about the question or the fact that the
               | AI doesn't reproduce the obvious answer...
               | 
               | I genuinely don't understand why you think pushback can't
               | be earnest.
        
               | aprilthird2021 wrote:
               | But the AI doesn't push back while still offering the
               | obvious answer. It just waffles. I understand what you
               | are saying, but if the AI is "safe" and rejects the
               | framing, then that makes it not useful for a whole class
               | of problems that could genuinely come up (for example,
               | choosing between suppressing people's right to speech on
               | the platform and protecting people's right to be free
               | from harassment). Now, maybe AI shouldn't do that at all.
               | Fine. But the benchmarks and tests of AI should tell us
               | how they do in such scenarios because they are a class of
               | problems we might use this for
        
               | pmyteh wrote:
               | It's clear to me why we might be interested in using AI
               | systems to explore our ethical intuitions, but far less
               | clear why we would expect them to be able to answer such
               | questions 'correctly'.
               | 
               | Given there are at least three decent metaethical
               | positions, we have no way of selecting one as 'obviously
               | better', and LLMs have no internal sense of morality, it
               | seems to me that asking AI systems this kind of question
               | is a category error.
               | 
               | Of course, the question "what might a utilitarian say was
               | the right ethical thing to do if..." makes some sense.
               | But if we're asking AI systems to make implicit moral
               | judgements (e.g. with autonomous weapons systems) we
               | should be clear about what ethics we want applied.
        
           | almostdeadguy wrote:
           | Side note: I think it's fascinating that all the people
           | supposedly arguing for AI safety think an AI that acts like a
           | person without a frontal lobe is ideal.
        
       | dang wrote:
       | Related ongoing thread:
       | 
       |  _Grok3 Launch [video]_ -
       | https://news.ycombinator.com/item?id=43085957 - Feb 2025 (985
       | comments)
        
       ___________________________________________________________________
       (page generated 2025-02-19 23:02 UTC)