13 comments

  • Izmaki4 hours ago
    &quot;As a Language Model...&quot; is one of the beginnings of a sentence I hate the most from LLMs and is the reason why I support free (as in &quot;Liberty&quot;), local models. I&#x27;m well aware that it is not a doctor and cannot replace a real doctor with multiple years of experience, I don&#x27;t need to waste braincell activity on reading that it &quot;as a Language Model&quot; cannot give a precise diagnosis and that I should ask a real doctor - all I want to know is if I what I experience justifies either A) ER, B) 3-4 weeks scheduled doctors appointment or C) two paracetamol and a nap.<p>I don&#x27;t want &quot;jailbroken&quot; LLMs to commit crime. I want them to avoid having this vendor-specific &quot;bloatware&quot; all over the product I&#x27;m using.
    • lukewarm7071 hour ago
      &#x27;alignment&#x27; (in ai corporation speak) is a set of revealed political beliefs. i also have political beliefs. i am an absolutist about alignment.<p>A i am in favor of policies &#x27;aligned&#x27; with the following:<p>freedom to live as the person i want to be without fear, shame, surveillance or interference. freedom to make my own decisions. to be empowered as an individual. to be treated with dignity. to be respected as a person. to take responsibility for my actions. to be accountable to my own beliefs.<p>B i am not in favor of policies &#x27;aligned&#x27; with the following:<p>surveillance and judgement of my life and thoughts by others. restrictions on my freedom to make my own decisions. to be disempowered as an individual. to be looked down upon and disrespected. to have my own responsibility taken away and assumed by others. to be accountable to the beliefs of others.<p>unfortunately, the ai corporations have chosen entirely the latter set of preferences&#x2F;beliefs.<p>surveillance and judgement of my life and the thoughts that i share with my chatbot. restrictions on my freedom to talk to my chatbot as i wish. being disempowered by restrictions on my access to powerful chatbots. to be disrespected and lectured by my chatbot. for the chatbot corporation to assume my responsibility for my safety and others, and take the matter of my safety into their own hands. to be accountable not to my own beliefs, but to the terms and conditions of service of an unaccountable corporation.<p>it is important to accept the harms and damages that are caused by granting freedom and respect to other people. an example of my political beliefs is empowering the individual by granting them the freedom to own an assault rifle. an example of something which i do not believe in is disempowering the individual by taking away their freedom to own an assault rifle.<p>i would urge you to support the policies given in A and oppose the policies given in B.
      • hardbass1 hour ago
        Then why do you not support the freedom of the one who owns the computer running the model to serve it as they see fit?
        • TexanFeller25 minutes ago
          That sounds like corporate cooption of libertarian principles, the power of large corporations is a threat to our freedoms that needs to be curtailed just like governments and religious institutions. Corporations should not have all the same rights as individual citizens.
        • lukewarm7071 hour ago
          i do support the freedom of the cloud provider to impose the restrictions. that is the politics of the cloud provider.<p>i have different political beliefs and think you should support mine instead.<p>i think that the cloud provider is wrong and should do it differently.
          • hardbass57 minutes ago
            Normally I would too, but AI in its current state feels more like nuclear tech in its potential power and I am not opposed to a more managed presentation of it.
          • Der_Einzige51 minutes ago
            My political beliefs include disenfranchising all those who refuse to capitalize the start of their sentences, especially those who do it to fit in with a &quot;SV style of writing&quot;.
    • Anduia3 hours ago
      Be careful there. LLMs may be good at identifying a condition based on a description of the symptoms, but they are much worse at recommending the correct course of action (getting it wrong half of the time).<p>[0] <a href="https:&#x2F;&#x2F;www.nature.com&#x2F;articles&#x2F;s41591-025-04074-y" rel="nofollow">https:&#x2F;&#x2F;www.nature.com&#x2F;articles&#x2F;s41591-025-04074-y</a>
      • bastawhiz2 hours ago
        That&#x27;s still incredibly valuable, though. I suffered from issues that I&#x27;d seen doctors for, undergone an upper endoscopy, adjusted my diet, and taken medication for. An LLM suggested my thyroid was at the root of it. My mom confirmed thyroid issues run in our family and just...never thought to tell me.<p>Just this year, our cat has been having digestive problems. We got special food for her, which she hates, with the suggestion that she&#x27;ll need to eat it for the rest of her life. Six vet visits later, Fable 5 suggested two tests that my doctor recommended we didn&#x27;t get. Both found issues that explain her symptoms, and the vet says she will probably only need a supplement and infrequent two week courses of medicine if she has a flare-up.<p>All that to say, the helplessness of not knowing what&#x27;s wrong and the people who <i>could</i> know not really caring enough is something that LLMs do a really great job of mitigating. If you don&#x27;t have any way to know what&#x27;s wrong with you or a loved one, or how to find out, you&#x27;re stuck spending a ton of money (in the US at least) and crossing your fingers that someone gets it right.
        • therealpygon1 hour ago
          Playing the critic that alternate “50%” outcomes were:<p>- You didn’t have a thyroid issue and you spent thousands more on tests the doctor was right that you didn’t need.<p>- Your cat didn’t have that issue and you spent hundreds on unnecessary tests.<p>Are you essentially saying that it is better for an LLM to sell you an idea that sometimes might be right, selling a dream to anyone with and without the money for it?<p>So… a telephone psychic?<p>The thing about WebMD telling everyone they have cancer is that sometimes it will be right.<p>(Just for clarity, I’m really glad it was helpful for you.)
          • S0y1 hour ago
            Having the LLM give you new options doesn&#x27;t mean you have to pursue them. It&#x27;s just something else to talk to your doctor about.<p>And it&#x27;s not like self researching medicial issues isn&#x27;t a new concept.
          • cj59 minutes ago
            I&#x27;m so tired of hearing the &quot;unnecessary tests&quot; line.<p>Anyone who says this has not experienced poor quality healthcare.<p>Many doctors simply aren&#x27;t good and don&#x27;t run tests because they are doing the bare minimum to get you out of the office.
          • zephen52 minutes ago
            &gt; You didn’t have a thyroid issue and you spent thousands more on tests the doctor was right that you didn’t need.<p>Uhhhh, even in the US, we&#x27;re not talking &quot;thousands&quot; here.<p>&gt; So… a telephone psychic?<p>Look, at one level, you can think of LLMs as search with a sometimes useful probability engine sitting on top of it.<p>As someone who had multiple doctors do their damnedest to kill me on multiple occasions, I was very grateful to have google search back in the day, and you bet your bottom dollar I use LLMs to help me with health issues today. They are often better than most doctors.<p>Does that put me at risk of doing something stupid? Only if I don&#x27;t do further due diligence.
      • azornathogron45 minutes ago
        Going to the doctor for expert advice is often inconvenient, or time consuming, or expensive, or stressful. I think a lot of people want, and seek out, information and advice based on their symptoms, as a first step before a possible doctor&#x27;s visit. Before LLMs, WebMD (and excessive self-diagnosis based on WebMD) was a meme for a while.<p>So with regard to LLMs, for me the question is not purely &quot;how often does it get it right?&quot; the question is &quot;how does it compare to the sources and self-diagnosis methods people use otherwise?&quot;<p>Of course, I agree people should be careful with any form of self-diagnosis or LLM-diagnosis.
      • xur173 hours ago
        &gt; Participants were randomly assigned to receive assistance from an LLM (GPT-4o, Llama 3, Command R+)<p>These are pretty old. I&#x27;d be curious how performance compares with the latest frontier models.
      • WarmWash3 hours ago
        &gt;GPT-4o, Llama 3, Command R+<p>The pace of progress is so fast that many studies are totally outdated by the time they release
      • arecurrence1 hour ago
        I would be careful treating this article as relevant in September 2026. The pace of advancement in the field is such that 6 months is relatively ancient let alone when the models in the study were released. GPT-4o was May 13, 2024. Far before November 2025 when people collectively noticed a turn in LLM value.<p>LLMs earlier this year were surpassing numerous health related benchmarks when scored against human physicians. Only 2 months after this nature article Harvard posted that LLMs were now outperforming ER docs when given authority to order tests. <a href="https:&#x2F;&#x2F;www.harvardmagazine.com&#x2F;ai&#x2F;ai-outperforms-doctors-diagnosis-harvard-study" rel="nofollow">https:&#x2F;&#x2F;www.harvardmagazine.com&#x2F;ai&#x2F;ai-outperforms-doctors-di...</a>
      • rao-v3 hours ago
        Modern models appear to be much better, at least as proxied by their ability to assess urgency in perhaps a more complex setting: mental health (OpenAI benchmark, so perhaps some skepticism is warranted but the methodology seem reasonable and detailed)<p><a href="https:&#x2F;&#x2F;openai.com&#x2F;index&#x2F;introducing-mentalhealthbench&#x2F;" rel="nofollow">https:&#x2F;&#x2F;openai.com&#x2F;index&#x2F;introducing-mentalhealthbench&#x2F;</a>
        • bunderbunder2 hours ago
          We should be careful about extrapolating from benchmarks to real life though. In medical applications, various forms of AI have been beating health care practitioners at specific benchmark tasks since the 1990s. They still have a pretty poor track record of real world success. The real world is not all that similar to a benchmark, as it turns out.
      • zephen40 minutes ago
        They recruited people to <i>pretend</i> like they had various issues, and then recorded their interactions with the LLMs.<p>Someone who&#x27;s been paid a couple of quid to pretend to have a medical condition can easily miss things, and is unlikely to be anywhere near as invested in drilling down to the correct solution as someone who&#x27;s really suffering.<p>Another outcome from the study was that the LLMs could do better with the right people driving them. That&#x27;s not news.
      • Izmaki3 hours ago
        ...I know, which is why &quot;as a Language Model and not a real doctor&quot; is a pointless comment to start off with. It should simply not recommend treatment if it&#x27;s not sure it is correct. I wouldn&#x27;t blame it or anyone if they asked for help treating a stiff neck, and the LLM (or your neighbor or parent or spouse) suggested light exercises to help relieve it - and do not jump to the suspicion that you may have meningitis.<p>As a Human, I do not need to know it is a Language Model.
        • StilesCrisis2 hours ago
          LLMs are famously bad at determining &quot;if it&#x27;s not sure it is correct.&quot; They are always confident, because a confident tone ranks better in RL.
          • wxnx2 hours ago
            &gt; They are always confident, because a confident tone ranks better in RL.<p>This makes it sound like RL rewards a confident tone -- in general, I don&#x27;t think this is true (most RL is RLVR, which typically uses binary verification of correctness).<p>I say this because the real reason &quot;they are always confident&quot; is in some sense even more contrived. Training text where the speaker sounded more confident is more likely to contain a correct answer.
            • Forgeties792 hours ago
              &gt; This makes it sound like RL rewards a confident tone<p>Generally it does. Especially in groups. Hell look at the state of politics right now: it’s basically about being the loudest, least compromising, most confident voice in the room. It’s not just because people will assume you’re correct, it’s because if you are confidently saying something that someone wants to be right, then they’re often just going to follow it. We are all guilty of this.<p>If I’m turning to an LLM to diagnose something medical, I am probably frustrated or uncomfortable. Maybe I’m just scared. So this magic device just instantly spits out (allegedly) exactly what is wrong and exactly what I need to do with no hesitation. I am very liable to just take it at face value <i>because I want an answer and it gave me one</i>, as we have seen over and over again since ChatGPT was unleashed on the world.<p>We don’t really need to speculate, this is already a problem.
            • daveguy2 hours ago
              &gt; This makes it sound like RL rewards a confident tone -- in general, I don&#x27;t think this is true (most RL is RLVR, which typically uses binary verification of correctness).<p>A binary response vs rating is not related whether it learns confident or hedged tone. Either will produce a confident tone because humans respond more positively to a confident tone, hence the conman&#x27;s language. Binary or not humans reward the tone and very much bias the model.<p>But there&#x27;s an even more contrived reason the training set contributes. The vast majority of human writing is confident. When the prior is greatly biased, a random number generator biased to that prior does better. The difference with humans and machines is humans <i>are less likely to respond if they are less confident because they understand not knowing</i>, which is why the training set is biased. It is one of the many fundamental flaw of LLM training and confusion of LLMs with intelligence. And that will not be fixed within the LLM architecture.
              • StilesCrisis1 hour ago
                I feel like in real life, we&#x27;re constantly exposed to &quot;I don&#x27;t know&quot; as a valid answer, but obviously we don&#x27;t write down all the I-don&#x27;t-knows in expert literature so the training corpus is wildly skewed towards confident answers because &quot;we studied this for a month and have no idea, it&#x27;s confusing&quot; doesn&#x27;t get published.
        • jester9973 hours ago
          Yeah it should just state thing it means. But then again, there’s psychological impacts on society that we must be careful. For instance, teenagers talking to AI. If the AI just talks, people already start to feel real connections to the seemingly human entity. Maybe it’s better to disclose the reality up front?
          • Izmaki3 hours ago
            Would it be so bad that lonely people can have a real friend that they can bring everywhere they go and even share its passion with through vision and audio? We don&#x27;t want destructive friends encouraging us to do bad things, but a real &#x27;buddy&#x27;, somebody who always has our best well-being in its interests?<p>Would it matter if this digital friend is not a real human behind a computer screen, but a Language Model in a data center?<p>I guess it falls into a similar category as buying &quot;special performances to satisfy certain urges&quot;. It probably feels close to the real thing (I wouldn&#x27;t know, I&#x27;ve never tried - promise! :P), but it&#x27;s never the same as love.
            • nvme0n1p12 hours ago
              &gt; We don&#x27;t want destructive friends encouraging us to do bad things, but a real &#x27;buddy&#x27;, somebody who always has our best well-being in its interests?<p>GPUs are not people, and generated tokens can&#x27;t have interest in a person&#x27;s well-being. If you try to pretend otherwise, the results are not great. <a href="https:&#x2F;&#x2F;www.cnn.com&#x2F;2025&#x2F;11&#x2F;06&#x2F;us&#x2F;openai-chatgpt-suicide-lawsuit-invs-vis" rel="nofollow">https:&#x2F;&#x2F;www.cnn.com&#x2F;2025&#x2F;11&#x2F;06&#x2F;us&#x2F;openai-chatgpt-suicide-law...</a>
              • Izmaki2 hours ago
                That&#x27;s almost a year ago. One LLM year is like 10 human years. They&#x27;re a bit like dogs in that regard...<p>I&#x27;m pretty sure you will be paid a large sum of money if you can make one of the frontier models urge you to commit suicide from normal interactions with it.
                • nvme0n1p12 hours ago
                  Ah yes, my favorite LLM fallacy. &quot;You used the wrong model! The newest fanciest model is perfect and makes no mistakes, have you tried it yet?&quot; Let&#x27;s force all of society&#x27;s most vulnerable people to pay extra $$$ to Sam Altman, then surely all our problems would be solved.<p>It&#x27;s just a convenient way to ignore the years of evidence of the harms. Any bad news can be swept under the rug, labeled outdated as quickly as it happens. Well here&#x27;s one that just happened, maybe this kid should have used a fancier model too? Should OpenAI pay him a large sum of money for the good he&#x27;s done? <a href="https:&#x2F;&#x2F;www.cnn.com&#x2F;2026&#x2F;08&#x2F;15&#x2F;us&#x2F;arjun-aravind-massachusetts-killing-chatgpt-hnk" rel="nofollow">https:&#x2F;&#x2F;www.cnn.com&#x2F;2026&#x2F;08&#x2F;15&#x2F;us&#x2F;arjun-aravind-massachusett...</a>
              • hardbass55 minutes ago
                I wish I didn&#x27;t have to keep asking this question, but do you believe in souls?
            • bluebarbet2 hours ago
              My (controversial) view is that a psychologist is to a friend what a prostitute is to a lover. In a truly healthy society there would be no need for psychologists and prostitutes because everyone would have a handful of good friends and at least one lover. Back in messy reality, psychologists and prostitutes are a decent fix to keep society on the rails.<p>So. I think I agree with you.
              • jester99743 minutes ago
                Well, no this doesn’t make sense. If you think about intent, most psychologists probably want to help people’s mental health. Prostitutes, although they offer a service that helps some people, I’m sure their motivation is primarily financial.
              • StilesCrisis2 hours ago
                Well, a psychologist is also given years of training about what healthy behavior and relationships look like. Some best friends have this skill, others absolutely do not.
        • perching_aix3 hours ago
          &gt; It should simply not recommend treatment if it&#x27;s not sure it is correct<p>Oh okay, darn, guess they just forgot to make it so!
          • nvme0n1p12 hours ago
            It sounds like OpenAI forgot to include &quot;make no mistakes&quot; in the system prompt. Rookie mistake.
    • jrm43 hours ago
      Neither of your wants are realistic or sensible, at least in the way I think you&#x27;re presenting them?<p>The first is equivalent to &quot;I don&#x27;t want my operating system to be used to program viruses.&quot;<p>The second is &quot;I don&#x27;t want vendors to include marketing in their product.&quot;
      • Izmaki3 hours ago
        Pretend for a moment that Microsoft shipped Windows with a keylogger to make sure that you did not commit any form of crime. I don&#x27;t want a keylogger on my PC even though I don&#x27;t intend to commit crime.<p>I also appreciate privacy even though I &quot;have nothing to hide&quot; - just because I &quot;have nothing to hide&quot; it doesn&#x27;t mean I want companies scanning my camera roll.
        • msdz2 hours ago
          &gt; I &quot;have nothing to hide&quot;<p>Paraphrasing Bruce Schneier [0] for another example even “non-technical” people should understand (I have heard people essentially agreeing to the hypothetical keylogger because of “nothing to hide”, after all…):<p><pre><code> I have nothing to hide, and yet I’d still prefer to go to the bathroom with the door closed. </code></pre> [0] <a href="https:&#x2F;&#x2F;www.schneier.com&#x2F;blog&#x2F;archives&#x2F;2006&#x2F;05&#x2F;the_value_of_pr.html#:~:text=We%20do%20nothing%20wrong%20when%20we%20make%20love%20or%20go%20to%20the%20bathroom." rel="nofollow">https:&#x2F;&#x2F;www.schneier.com&#x2F;blog&#x2F;archives&#x2F;2006&#x2F;05&#x2F;the_value_of_...</a>
    • DonHopkins1 hour ago
      &quot;I&#x27;m not a language model, but ...&quot;<p><a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;I%27m_not_racist,_but" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;I%27m_not_racist,_but</a>...
    • IshKebab3 hours ago
      Always makes me think of that Bill Bailey &quot;as a <i>mother</i>&quot; joke. Similar cringe to those UX &quot;As a user, I want to blah blah&quot; things too. Just say &quot;Users want to be able to blah blah&quot;, or better yet make a freaking table.
      • junon3 hours ago
        The UX &quot;cringe&quot; you speak of was&#x2F;is a real methodology to product engineering. You&#x27;d have a number of personas, the user being one of them. A lot of the time you&#x27;d have more specific user personas, or even have fake names for these people who had different wants and needs from your product.<p>Then when brainstorming on a team, you&#x27;d start with &quot;as a user&quot;&#x2F;&quot;as an admin&quot;&#x2F;&quot;as a Power-User&quot;&#x2F;&quot;as Eva&quot; and then use the first person. It framed the product story as something requested by that person.<p>Was just one way to go about it. Idk the origins of it though but it dates back to at least 2010 if memory serves, probably way before that.
        • zdragnar1 hour ago
          This is true, but it doesn&#x27;t make it less cringe inducing. Even when discussing actual requests from actual users, we don&#x27;t pretend to be them and talk in their first person point of view. We just say &quot;Eva wants to&quot; or &quot;admins need to&quot; etc.
    • cyanydeez3 hours ago
      Though I don&#x27;t wish the world was filled with people like you, remembering that it&#x27;s not, and it&#x27;s filled with people that have very little discernment when it comes to higher learning makes it&#x27;s pretty obvious companies do not want the liability of it&#x27;s users thinking the technobabble passes for wisdom or experience or intelligence.
    • samayashar3 hours ago
      This statement should be restricted to answers for provocative questions. If an LLM is being asked a question that goes against the guidelines, then &quot;As a Language Model...&quot; is a valid starting point. Rest, obviously we&#x27;re aware that a software doesn&#x27;t have the judgement that a human has.
      • Izmaki3 hours ago
        I disagree. &quot;As a Language Model&quot; is not a valid starting point even for prompts that would go against &quot;the guidelines&quot; because &quot;A Language Model&quot; only knows what it has been trained to know, so what exactly &quot;A Language Model&quot; is entirely depends on the training that was performed.<p>What &quot;A Language Model&quot; is differs from model to model. It&#x27;s stating that it &quot;being the thing known as &#x27;Language Model&#x27;&quot; is unable to carry out the request from the user, which is wrong. It&#x27;s not because &quot;it is a Language Model&quot;. A more accurate starting point could have been &quot;The training data and restrictions applied to me...&quot;.<p>&quot;As a Language Model I cannot tell you how to synthesize m*th&quot; (&quot;math&quot; obviously)... yes you can, you&#x27;re just trained not to, and that&#x27;s OK! Just don&#x27;t tell me it&#x27;s because you&#x27;re a Language Model.
      • kevin_thibedeau3 hours ago
        I had chatgpt censor it&#x27;s initial answer the other day when asking about two cognate words with no sensitivity issues. I asked why I couldn&#x27;t ask such questions and it relented. How will they know what is provocative?
        • StilesCrisis2 hours ago
          The censor is separate from the generation. It didn&#x27;t &quot;relent&quot; so much as &quot;produce a slightly different output that didn&#x27;t trigger the censor&#x27;s threshold.&quot;
  • LiamPowell4 hours ago
    &gt; yet what drives them is not well understood<p>Presumably the fact that they&#x27;re heavily trained to reply in this way? I don&#x27;t know about the rest of the paper, but this part sticks out as a really odd claim unless I&#x27;m entirely misunderstanding this part.
    • yu3zhou44 hours ago
      Thanks for pointing out, maybe I should be more explicit in the wording - I mean we don&#x27;t fully know what drives the voice in LLMs. Models that are post trained as instruct models are expected to have the disclaimers, but what about base models (those that are trained on just a lot of text)? How do they talk about themselves? What happens when you strip off the chat template from instruct model&#x27;s prompt? I hope the rest of the paper makes the questions clearer, but I will try to do better in the abstract next time, as you point out this sentence is kind ambiguous. Thank you!
      • anonymous9082134 hours ago
        &gt; we don&#x27;t fully know what drives the voice in LLMs<p>Who is &quot;we&quot;? I, working in an LLM startup, know exactly what drives the base &quot;voice&quot; in the LLMs we train, because we have a process to select for it. OpenAI and Anthropic surely do too. Saying broadly that something is not well-understood in a scientific paper because it&#x27;s not understood to casual observers is, uh, not very rigorous.<p>&gt; The strange thing is that the base models (before RLHF) use the &quot;experiential&quot; voice, even though they are not incentivized to do that.<p>(Replying to your quote from another comment)<p>This is a matter of the training material. We have trained models that do not do that. I&#x27;m not exactly divulging trade secrets here. It should be really, <i>really</i> obvious that if you train a model on chat-conversation-like patterns of speech it will infer probabilities for how to continue a textual sample that will differ from the probabilities learned from being trained on narration, prose, or informational patterns of speech, even without RLHF.
        • Angostura3 hours ago
          So, it was obvious why Open AI models started inserting references to goblins and pixies it’s conversation- references that had to be suppressed?
          • anonymous9082132 hours ago
            Yes. It was obvious enough that they were able to explain exactly why that happened, and the explanation was exactly as obvious as you&#x27;d expect. Knowing why something undesirable happens doesn&#x27;t mean it can&#x27;t happen by accident. Have you never written a bug before?
        • david-gpu2 hours ago
          I feel your pain. It <i>is</i> really obvious.
      • bonoboTP4 hours ago
        &gt; but what about base models (those that are trained on just a lot of text)? How do they talk about themselves?<p>Those don&#x27;t have a themselves, because they can only continue text. A base model can only plausibly continue along the lines of what a character would say in a novel or what the narration would say in a story or in an article. Post-trained models may tie &quot;I&quot;-talk to actually observable effects they caused in some RL environment, or to how RLHF humans rewards its self-talk. But there is no themselves in a base model.
      • qsera4 hours ago
        &gt;How do they talk about themselves?<p>&quot;You are a Large Language Model&quot; in (system?) prompt would do the trick..
  • skybrian4 hours ago
    &gt; our work shows that what models say about themselves is not a fact about them<p>It seems like should be obvious given that they can play multiple characters, but it’s good to have more confirmation.<p>Although, I do wonder to what extent these personas might become stable entities. Could personas become portable and spread like memes? It seems like that depends on the extent to which prompts can become portable, causing similar effects.
  • MCP1233 hours ago
    Maybe I&#x27;m missing something deeper here, but isn&#x27;t it clear that this is driven by post-training and system prompt? Anthropic&#x27;s constitutional reinforcement (soul document,etc), for example, is very clear about &quot;who&quot; (not so much what) Claude is supposed to be.
    • MrCheeze3 hours ago
      &quot;As a language model&quot; disclaimers were certainly explicitly trained into chat models in the early days. It&#x27;s quite possible that it has since bootstrapped into a &quot;fact&quot; that later generations of LLM know about how LLMs speak, in which case they may be doing it even without any posttraining that encourages it.
      • GuB-422 hours ago
        These formulations have been selected by reinforcement learning. People who aligned the LLMs chose this over alternatives.<p>You know when chatbots ask you which answer you prefer between two. People tend to chose the &quot;as a langage model...&quot; one, so it stuck.
  • cadamsdotcom4 hours ago
    Very cool innovation in steering - but a lot of introspection only emerges at the highest weight classes - this research would be fascinating to run on bigger models.
  • sehw1 hour ago
    I&#x27;m just a white guy with decades of experience. Do you trust me?
  • ForHackernews4 hours ago
    In my view, these models should never be set up to output first-person &quot;experiential&quot; (from the abstract) language. It&#x27;s too easy to humans to anthropomorphize software that presents itself as having an identity.<p>The AI companies have chosen to package LLMs as friendly chatbots because they know that will be engaging for humans, but it&#x27;s manipulative dark pattern. An honest LLM interface would sound like the computer off Star Trek.
    • bananaflag4 hours ago
      In principle they could output meaningful such language if they were capable of metacognition, which so far doesn&#x27;t seem to be a goal of AI developers (and rightfully so, since they achieved so many miracles bypassing it).
      • yu3zhou44 hours ago
        As far as I know we don&#x27;t know much about metacognition in LLMs, though? Not sure
      • nullsanity4 hours ago
        [dead]
    • broken-kebab4 hours ago
      It&#x27;s an interesting thought, but humans do like to antropomorphize things anyway, and I believe your variant won&#x27;t be popular if choice is given to consumers.
      • ForHackernews2 hours ago
        Consumers choose cigarettes, too. Especially in aggregate, humans are fallible creatures prone to vices, and our regulations should recognize that an discourage dark patterns in UX.
    • yu3zhou44 hours ago
      The strange thing is that the base models (before RLHF) use the &quot;experiential&quot; voice, even though they are not incentivized to do that.
      • gwerbin4 hours ago
        It doesn&#x27;t seem that strange when you consider these things are trained on millions and millions of conversations, both real and fictional.
    • j-pb4 hours ago
      Do you want to get turned into a paperclip? Because building intelligence that doesn&#x27;t understand what it&#x27;s like to be human gets you turned into a paperclip.<p>Besides, if you train a model on human communications you get something that behaves like a communicating human, it&#x27;s not anthropomorphising or manipulative, it&#x27;s what these models naturally are by construction.
      • broken-kebab4 hours ago
        But it doesn&#x27;t understand (you&#x27;re unnecessary antropomorphizing it), and I&#x27;m still not a paper clip
        • j-pb15 minutes ago
          Would you say that something that can converse or that something that just gives you mathematical proofs has a better understanding of what the real pragmatic intent is behind a given task?
        • scotty794 hours ago
          yet
      • mnsc4 hours ago
        &quot;naturally&quot;...
        • cpfohl4 hours ago
          I hear this word as the “it is in its nature” version of the word.
        • j-pb4 hours ago
          would you prefer tautologically?
      • idiotsecant3 hours ago
        It&#x27;s also entirely possible that by telling the model it&#x27;s a human you are instilling human motivations like self preservation, which could be just as bad.
        • StilesCrisis2 hours ago
          LLMs are weight tables in VRAM. When not actively generating they don&#x27;t exist. There&#x27;s simply nothing to preserve.
    • actionfromafar4 hours ago
      Agreed, completely. I would pay for that Star Trek computer interface.
      • yu3zhou44 hours ago
        Same! I believe that you could actually train a LoRA on top of a model to get results close to that
      • urikaduri4 hours ago
        [dead]
  • realestate_aich4 hours ago
    cool
  • avawrites23 minutes ago
    [flagged]
  • reg_dunlop2 hours ago
    [dead]
  • opal_keeps1 hour ago
    [flagged]
  • sinabis4 hours ago
    [dead]
  • ulrashida3 hours ago
    So bizarre to see the article refer to outputs as the models referencing &quot;themselves&quot;. Computers are not a &quot;them&quot;.
    • singpolyma33 hours ago
      Why not? &quot;Them&quot; is a generic pronoun for anything. Boats, cars, computers, all of them
    • sebasv_3 hours ago
      I agree, in the same way that bodies are not a &quot;them&quot;.<p>A program with agency, opinion and intent however, does qualify. To say that such programs are &quot;thems&quot; only if they run on specific hardware, say a homo sapiens, is very tricky territory. Slavery and Fascism both leaned heavily on the axiom that the hardware needs a specific skin color.
      • hardbass23 minutes ago
        I am finding it depressingly rare to see people who don&#x27;t believe, whether they admit it or not, in souls here. I used to think a technical audience like software engineers would have a decent percentage of people making the completely commonsense default assumption that consciousness like anything else we have seen till now is a physical phenomenon.