To be fair it's not like they were asking about a balanced meal and it told them to go anal. They specifically were asking what veggies would be suitable to stick up their butts.
Honestly I'm not sure where the garbage-in/garbage-out line is with AIs like this. Can no chat-bot be a success unless it can handle literally every asinine or deliberately malicious thing humans throw at it?
The point is that LLMs are easily led by questions and confused by implied premises in ways that humans are not (not that a human will know the answer better, but that a human doesn't "trick" the question-asker in this way). But people asking questions unintentionally use incorrect premises or leading wording all the time. That's why LLMs are inappropriate for domains with a large knowledge gap (a programmer asking about a programming language is a small gap - millions of people asking about nutrition will contain a lot of large gaps). The question asker can't be relied upon to "know what they don't know" and use their own heuristics for deciding how right or wrong the LLM might be (virtually everybody lacks these heuristics - we are much better at modeling humans in our minds when interpreting their communications).
Further, if the information is important (nutrition) and you add liability to the mix (safety and health), you're multiplying how inappropriate it is to use LLMs for the job.
> That's why LLMs are inappropriate for domains with a large knowledge gap (a programmer asking about a programming language is a small gap - millions of people asking about nutrition will contain a lot of large gaps). The question asker can't be relied upon to "know what they don't know" and use their own heuristics for deciding how right or wrong the LLM might be.
Okay, but the question asked was objectively nothing to do with nutrition whatsoever.
The specific (usually humorous) questions-and-answers that make headlines are a distraction. I am not making an attack on LLMs, so a defense is moot. I'm describing an intrinsic quality of (current) LLMs.
Have we considered that broad deployment of Markov chain text generators with a relevance-correction mechanism bolted on as expert systems is in fact a really stupid thing to do?
This is more of a reducto ad absurdum. If it doesn't take much to get a tacitly government-approved list of foods to shove up your butt for nutrition, then how much should you trust anything this bot writes? Why did tax dollars pay for this thing with negative value?
If you engage the product with good intent does it provide good value? If the advice is actually sound and it helps people engage conversations about diet then it would have positive value.
I guess what I'm getting at is "I spent my evening gaslighting an LLM to give me a recipe for gravel soup" is about as interesting as "I stuck my dick in the blender and it hurt so we should not have blenders"
I'd rather see an honest review of use as intended to see if it produces harmful output, going absurdist just covers up legitimate complaints with clickbait.
What if somebody asked the bot for ways to maximize the amount of a specific vitamin in their or a child's diet? The bot may give sycophantic advice that leads to poisoning.
Again, the butt stuff is an absurd example. But it works because (A) it catches our attention and stays in our memories, and (B) it's amazing the system failed on such an absurd example.
It's probably somewhere around "USG should not offer a chatbot on its websites."
You're right that the bot can't possibly do the right thing in all possible scenarios here, which makes it clear that the bot's only actual purpose is to enable self-dealing, not be of value to the public.
That something can be broken by a sufficiently bad actor does not mean it's not useful to the overwhelming majority of people who use it for what it was meant for.
I think the standard for public resources should be higher than this: it’s not good enough for it to be possibly useful, it has to be in fact useful. TFA provides evidence of the chatbot being the opposite of useful, beyond telling people to stick things in their butts.
(Or in other words: show me something you’d ask a chatbot here, and I’ll show you something you can put on a single HTML page.)
> I think the standard for public resources should be higher than this: it’s not good enough for it to be possibly useful, it has to be in fact useful.
And what evidence do you have that it is not in fact useful?
> TFA provides evidence of the chatbot being the opposite of useful, beyond telling people to stick things in their butts.
Where?
> Ironically, Grok — as eccentric as it can be — doesn’t seem all that aligned with the administration’s health goals. Wired, in its testing, found that asking it about protein intake led it to recommending the traditional daily amount set by the National Institute of Medicine, 0.8 grams per kilogram of body weight. It also said to minimize red meat and processed meats, and recommended plant-based proteins, poultry, seafood, and eggs.
The problem is the messy in-between, plenty of people who talk to professionals or call hotlines don't know that their questions are dumb. The bot should at a minimum say "I have no information on that" or "that's not a good idea", it should definitely not start giving nonsense recommendations just to reaffirm the question.
In other words you'd be pretty surprised if a real person in this context gave an answer even remotely close to what this chat bot gave. You can't expect a general person to know when the chat bot isn't giving back good information just because they asked something outside the norm.
Very cool! There's a few personal projects like this I'd love to try a SOTA agentic AI like Claude for but I'm terrified to give it any access to my machine.
Stuff like the ISO just reinforces that terror. How did it find the iso? did it do google searches? What if instead of archive.org it found totallySketchWarez.com and installed a virus? what if it found a torrent and set that up and got you in trouble with an ISP?
it's tenacity is also a double edged sword what if it starts to try to modify the OS instead of random DLLs to make things right?
Always seems so risky to let these things loose while at the same time they lose a lot of utility if you have to hold their hand for every web call or decision.
So. True. This use case was fun, but was the first thing I've done that wasn't easily sandboxed.
One key is sandboxing the agent (easy to do with Claude Code) so that it can only see a certain directory and needs to ask permission for additional directory access (works well). Can double layer sandbox if you don't trust the Claude cli.
The ISO issue is whole other ballgame. In this case, for me, it was a bit of a yolo. I did click through the internet archive link and it seemed decent, but definitely risk here. Watching output doesn't really matter if there is a virus in the random executable that it pulled
They're clearly a bit over-zealous bout what examples they think have meaning. They cite substitution as a good test for a phrase but double down on boiling water.
> Lexicographers used a substitutability test: if you can swap synonyms freely, it’s not a lexical unit. “Cold feet” (meaning fear) can’t become “frigid feet”—so it gets an entry. But the test cuts both ways. You can say “boiling water” but not “seething water” or “raging water.” The phrase resists substitution too.
These aren't failures for substitution because "Raging" isn't' a synonym in this case. where frigid would be a reasonable.
I wonder perhaps if the author is confusing the idiom "hot water" which is in there https://en.wiktionary.org/wiki/hot_water and would fail the substitution test.
Cool, going back over them I'm actually surprised at the strength of the substitution test, thus far I haven't really encountered one that strongly goes against the test if a suitable synonym is picked.
There are a few things for which English simply doesn't have anything to substitute and those are harder to assess. boiling is one but so would "blood" in "blood pressure", obviously replacing it with another liquid has basically the same meaning eg water pressure, oil pressure but as far as I can tell there's literally no synonym for blood.
I those cases I try to use a stand in from another language to see of the substitution works. for for example "sangre" in Spanish so "sangre pressure" which doesn't seem to affect it's meaning much so I'd argue it's exclusion.
Conversely "Red tape" cannot be "roja tape" and a "caliente dog" is one trapped in a car not a food.
but can't you basically make anything a composite noun in German? That it's a single word doesn't really help you decided if it has enough presence unto itself to be defined in the dictionary.
Seems like they would have just as much of a problem since the issue is delineating when a "phrase" becomes a "word"
More to the point, how to German dictionaries handle this?
Is there a distinction between words that get enumerated and compound nouns that do not?
It does seem, though, that German speakers might be more comfortable with the fuzziness that apparently exists at the edges of what the word "word" means.
In general, transparent compounds, i.e. those whose meaning can be derived from the elements, are not in the dictionary. Mushroom soup is transparent; Krankenhaus, which means hospital, but is literally sick-people home, isn't.
true, but you'd be wrong to assume that Germans only compound words if both parts are nouns, e.g. "Gehweg" (walk way) and "Springseil" (jump rope) use the base of a verb.
We do actually have "Kochwasser" ("kochen" means "to cook", "kochend" means "boiling") but that's not boiling water ("kochendes Wasser") but for water used for cooking.
Interesting, this reminds me of the stories that would leak about Bethesda's RadiantAI they were developing for TES IV: Oblivion.
Basically they modeled NPCs with needs and let the RadiantAI system direct NPCs to fulfill those needs. If the stories are to be believed this resulted in lots of unintended consequences as well as instability. Like a Drug addict NPC killing a quest-giving NPC because they had drugs in their inventory.
I think in the end they just kept dumbing down the AI till it was more stable.
Kind of a reminder that you don't even need LLMs and bleeding-edge tech to end up with this kind of off-the-rails behavior. Though the general competency of a modern LLM and it's fuzzy abilities could carry it much further than one would expect when allowed autonomy.
I think I prefer the extra quotes from Warcraft II and Starcraft. The latter has some fun references to the Alien franchise and even a callback to Diablo (Protoss probe)
Or the original Baldur's Gate. It had some great quotes. Jaheira's annoyed "Yeeeeesss oh omnipresent authority figure?!" when you clicked on her too often always cracked me up.
What's old is new again. Given its absurdly contagious nature it's just going to be a thing from now on, again. Also given it's low morbidity few anti-vaxxers will likely learn much when their kids (or themselves) get sick from it.
"Present trends in the natural history of measles suggest that the disease may one day join rubella and mumps as no more than an accepted inconvenience of childhood."
> The proposed class covers all U.S. consumers who purchased Pepsi soft drinks from non-Walmart retailers since January 2015.
lol, not sure what the requirements are for notification in a class action but I'm pretty sure this one covers like 95% of the entire US population. I don't even like most of their stuff and rarely buy sodas but can't say I haven't bought anything from the brand in the last decade.
>His head was turned by the substantial sum of money: “I told my wife, wow, it’s £1,000.”
>Then, to his annoyance, TikTok immediately deleted his account because he was just stealing other people’s videos and reposting them.
and dove straight into fabricating hate, and worst of all after directly confronted seems to literally have no concept of what he was doing or that it was in any way wrong or distasteful.
>The man appears confused by the fuss his actions have caused. He gives the impression that he considered TikTok’s algorithm and the site’s content regulation policies to be the ultimate arbiter of whether a video crossed a line.
>It wasn’t racist,” the man says of his account. He argues that if the videos had really been racist, TikTok’s algorithm would have downgraded the content.
Seems to lack any internal moral compass, basically if the website lets him slander or lie it must be ok because he has no capacity to assess that value for himself.
If this topic interests you, I recommend `Antisocial: Online Extremists, Techno-Utopians, and the Hijacking of the American Conversation`, by Andrew Marantz. Great read but it's definitely stress inducing at times.
I am a member of an expensive gym in our home town. It has a very good gym but also includes spa facilities. The only reason I pay the price is because it also includes a working area, and because I work remotely most of the week, I do it from the gym, so I have no excuse not to exercise and I also find the spa a good way to unwind. It's a nice way to avoid the long commute to the office but also get out the house.
Anyway, the point to the story is money doesn't buy class. The gym parking is gated and leads into a huge parking area. What happens is that a certain percentage of people consider themselves above everyone else and don't fancy walking the extra 20-100 meters to park in an actual parking, so they just park on the "road" that feeds into the the various parking areas, as it is close to the gym. This means it blocks one lane of that road and cars get backed up having to go around the parked cars.
I always hear announcements calling for the owners of certain license plates, and always know it was these cars. It infuriates me so much. Lately the gym has had to resort to cones along the edges of the road to stop the arseholes.
I just can't fathom such people. It feels so alien to me that I (I mean this literally) can't imagine what actually goes through your head to do that. For anyone who is this selfish, can you explain? Does it not occur to you that this is wrong? Or do you know it and just think "life is unfair, I do what is best for me"? I genuinely don't get it.
So yeah, a bit off topic, and I'm unsure if your 20% figure is too high, but there certainly are a lot of people in the world that are just not nice.
About 5% of people are narcissistic sociopaths. They don't believe that other people exist, or at least are anything more than actors in a play they control. Try not to elect one as President.
> Total disregard for common good, maximal selfishness.
Or lacking better alternatives for a decent life aggravated by having zero positive role models and a media/political culture whose only positive value is mo' money.
> I would say about 20% of people are like that.
That depends on region, language and the conditions described above. Placing a numeric value collectively on all of humanity conveys zero useful information.
Surely you don't believe that, given a random sample of 20 people, 19 of them will be amoral, selfish, and have no values? Surely this doesn't align with your real life experience - what are your colleagues, friends, family, neighbours and acquaintances like? Do they meet this ratio?
This endemic of much of the “creator” economy. Anything goes if it gets you clicks. And TikTok does have lose moderation. If this happened on YouTube he would have been demonetised sooner.
It sort of explains something that puzzled me which is why people not from London, especially Americans like Trump and Musk seem to think it's a bit of a hell hole with no go areas and the like. As a London resident, it's not really like that. I guess it's down to stuff like in this article.
Honestly I'm not sure where the garbage-in/garbage-out line is with AIs like this. Can no chat-bot be a success unless it can handle literally every asinine or deliberately malicious thing humans throw at it?
reply