back to article OpenAI's latest o1 model family tries to emulate 'reasoning' – tho might overthink things a bit

OpenAI on Thursday introduced o1, its latest large language model family, which it claims is capable of emulating complex reasoning. The o1 model set – which presently consists of o1-preview and o1-mini – employs so-called "chain of thought" techniques. In a 2022 paper, Google researchers described chain of thought as "a …

  1. Pascal Monett Silver badge

    "capable of emulating complex reasoning"

    So, AI is not actually intelligent, it's just pretending to be ?

    Thank you for the admission.

    1. Persona Silver badge

      Re: "capable of emulating complex reasoning"

      Same as most people, but without the pretending part.

      1. katrinab Silver badge
        Megaphone

        Re: "capable of emulating complex reasoning"

        Your least favourite idiotic politician [Donald Trump/Joe Biden/whoever] is *far* more intelligent than any AI model.

        1. Anonymous Coward
          Anonymous Coward

          Re: "capable of emulating complex reasoning"

          [Donald Trump/Joe Biden/Boris Johnson/Liz Truss/whoever]...

          But still prone to hallucinations...

          "My dog's got no nose... 'cos an illegal's eaten it"

          1. Anonymous Coward
            Anonymous Coward

            "My dog's got no nose... 'cos an illegal's eaten it"

            Boom boom!

            how does it smell... ?

            not the illegal, the dog. How does the dog smell?

            10 quadtrillion Upvotes 0 Downvotes

            1. steelpillow Silver badge
              Joke

              Re: "My dog's got no nose... 'cos an illegal's eaten it"

              That's a terrible joke!

    2. Ken G Silver badge

      Re: "capable of emulating complex reasoning"

      Aren't we all just pretending?

      I like that there seems to be some explainability from this model. If that extends to other areas it could be a differentiator.

      1. Rafael #872397
        Pint

        Re: "capable of emulating complex reasoning"

        Aren't we all just pretending?

        On a Friday early afternoon? I'm not even trying.

    3. Anonymous Coward
      Anonymous Coward

      Re: "capable of emulating complex reasoning"

      Indeed. I think they are to be commended for clarifying that it's only emulating reasoning. Too many AI hype sellers would drop the word emulating for profit reasons.

    4. Filippo Silver badge

      Re: "capable of emulating complex reasoning"

      It's a tricky point. We don't really know how intelligence works, so if a computer program appeared to be genuinely human-like, we'd have no way to know whether it's "real" or not.

      That said, I don't think o1 is anywhere near that point.

      1. Persona Silver badge

        Re: "capable of emulating complex reasoning"

        The step up occurs when an "AI" reasons how to make itself better.

        1. Anonymous Coward
          Anonymous Coward

          Re: "capable of emulating complex reasoning"

          > The step up occurs when an "AI" reasons how to make itself better.

          >> The o1-preview model proved more adept at optimizing the performance of a byte pair encoder in Copilot Chat's tokenizer library.

          If "optimised" implies "better" then...

          (Ok, o1-preview only improved one application of itself, in this case, but - close enough?)

    5. Justthefacts Silver badge

      Re: "capable of emulating complex reasoning"

      Rubbish. Watch this, and tell me. It’s GPT4o solving graduate-level physics problems, out of a standard textbook, in two minutes. Not a puff piece, or corporate, just a graduate student putting it to work, for real. The guy showing the video reckons these problems take him over a week. Honestly, that’s not true, for me this sort of stuff is maybe a couple hours. But still, remember GPT4o is not tanking on full here: the hardware is *time-shared* between probably 10-20 users. And it’s still 50x a decent Physics PhD.

      If *you* can’t get it to code up “show me a few database queries on a webpage” properly, the problem is you, not the tool.

      https://youtu.be/scOb0XCkWho

      1. Justthefacts Silver badge

        Re: "capable of emulating complex reasoning"

        Now, you need to watch his livestream, right now. ChatGPT-o1 has just recreated a chapter of this guys astrophysics PhD work, based only on the “Method” section in the research paper he wrote. It’s writing hundreds of lines of Python, that originally took him *a year* to think through, and it’s doing in a few minutes. Not single-shot, but it will auto-correct when prompted the correct way.

        https://youtu.be/GaAaFkipaTQ

  2. Anonymous Coward
    Anonymous Coward

    We are entering iPhone terroirty

    This is no real great shakes. Nicer camera in the iPhone thing.

    I pretty sure OpenAI have announced the Chain-of-Thought is only a brief preview and that they will switch it off soon or may have done so. They have done something it to it. We had a preview model and it showed you everything. If it still works it is heavily censored and more a curiosity than a reverse-engineering tool the last version was.

  3. Wellyboot Silver badge

    Implied assumptions

    The '9 apples is correct'* answer only applies if you assume that in the process of making lunch the other 20 are no longer within the cafeteria, there is a quite logical answer of '29 apples' that merely requires the assumption that all apples that changed state (apple > apple in lunch) are still apples and still in the cafeteria.

    A valid conclusion for the original question is that the LLM is now making assumptions that are more in line with human perception of the questions intended meaning.

    Will this in any way reduce the tendency of LLMs to hallucinate?

    * 9 apples would be a definitive answer if the original question was changed to ‘How many unused apples...’

    1. katrinab Silver badge
      Meh

      Re: Implied assumptions

      If your aim is to demonstrate that it can match a reasonably competent primary school child in arithmetic, then I think the answer it gave is fine.

      1. JacobZ
        Boffin

        competent primary school child Re: Implied assumptions

        Indeed, it does match a competent primary school child.

        The problem here is that OpenAI claims it can match a PhD student.

      2. Bebu
        Windows

        Re: Implied assumptions

        If your aim is to demonstrate that it can match a reasonably competent primary school child in arithmetic, then I think the answer it gave is fine.

        The kid has learnt that number (and presumably other properties) is not dependant on the particular objects being counted or the objects' purpose. I cannot accept ChatGPT has any equivalent.

        Would ChatGPT respond correctly to "I have 5 green apples for preparing apple pie and 3 red apples for roast suckling pig*. How many apples do I have?"

        * while I am not a vego I have to admit that is rather gross. Oddly the recipe with the image specifies kosher salt. ;)

    2. I ain't Spartacus Gold badge

      Re: Implied assumptions

      Wellyboot,

      This depends on your definition of apple. If I have 3 apples and I turn them into apple sauce, I no longer have 3 apples. You could say I have some apple. Or that I have 3 apples' worth of apple sauce. Or I have some apple peel and some apple pips in the bin and a saucepan containing the residual material combined with a bit of sugar and water.

      But, without ownership of a time machine, I am no longer in a position to hand you an apple.

      In the same way that if I'd cooked 3 humans - I would no longer have 3 people in my kitchen - I would now have some meat pies, some long-pork, some long-bacon, some sausages, various leftover bits too horrible to even put in sausages and an urgent desire to leave before the police turned up. But you wouldn't say I have people in my kitchen - though you might say there are bodies in that kitchen - but mroe likely the police report would say "remains".

      1. Anonymous Coward
        Anonymous Coward

        Re: Implied assumptions

        >> If they used 20 to make lunch

        It depends upon your definition of "make lunch".

        Many a cafeteria lunch includes an apple plonked onto the tray. Especially in a school cafeteria.

        As it also does in our household - the apple, as well as the tomatoes, are served as-is for lunch. And, depending upon how hungry everyone is, may well go back to the pantry and re-appear in the same form the next day.

        You can argue for both these interpretations; there is not enough information provided to be able to select between them, other than by personal choice of "well, obviously...".

        Now, if the LLM had been able to point out these issues and had asked for clarification, it *might* be comparable to a decent attempt at a line if reasoning.

    3. munnoch Silver badge

      Re: Implied assumptions

      No, because a used apple is no longer an apple. That's what the primary school child would get out of that scenario.

      Its a bit like saying if I have a litre of water and I drink half of it and piss it back out I still have a litre of water. Yes, there is still a litre's worth of H2O molecules floating around but you'd have a hard time passing off piss as water. (But I'm sure if anyone could spin it that way then OpenAI could...)

      1. DJO Silver badge

        Re: Implied assumptions

        You must be too young to remember Watney's Red Barrel. Whatever they made it with it wasn't (drinkable) water.

        1. I ain't Spartacus Gold badge
          Pint

          Re: Implied assumptions

          I'll have the Party 7 please!

        2. Anonymous Coward
          Anonymous Coward

          Re: Implied assumptions

          "like making love in a punt... f'ing close to water"

      2. Anonymous Coward
        Anonymous Coward

        Re: Implied assumptions

        >> If they used 20 to make lunch

        > No, because a used apple is no longer an apple. That's what the primary school child would get out of that scenario.

        The primary school child would get that the apple had just been dropped onto the child's tray by the school cafeteria staff "making lunch" and was still very much just an apple. And was still in the cafeteria.

        When the lunch had been eaten, or the apple pocketed and carried out of the cafeteria because "you are not leaving until I see those plates are empty", then the number of apples in the cafeteria will have changed. But the question does not ask about that...

      3. Bebu
        Pint

        Re: Implied assumptions

        you'd have a hard time passing off piss as water.

        Beer on the other hand doesn't appear to suffer from water's defect.

    4. BryanFRitt

      Re: Implied assumptions

      Also assuming the apples at the start and bought were all "unused apples". Although, calling them "unused apples", instead of just "apples" is unnatural for most people.

  4. elsergiovolador Silver badge

    Advantage

    Asked this O1 to create me a simple mobile app.

    It did. With all steps from installing tooling to getting it on the phone. All in 5 minutes.

    Now imagine asking this your junior dev.

  5. Howard Sway Silver badge

    Chat GPT Apples

    I just asked it an ambiguous apple question .It knew if was ambiguous, and then gave a completely wrong answer :

    If I have 5 apples and take two away how many apples do I have?

    If you have 5 apples and take 2 away, you would have 2 apples remaining. However, if you are referring to the apples you took away, you would have 2 apples in your possession. So, you have 2 apples that you took away.

    But if I then ask it

    If I have 5 apples and take 2 away how many apples do I have?

    If you have 5 apples and take 2 away, you would have 2 apples in your possession (the ones you took away). You would still have 3 apples left from the original 5. So, you have 2 apples that you took away.

    So, it can't work out that 2 and two are the same thing.

    1. elsergiovolador Silver badge

      Re: Chat GPT Apples

      That sounds like talking to an autistic person. I wouldn't dismiss that as bad, you just have to be more precise.

      1. doublelayer Silver badge

        Re: Chat GPT Apples

        They said the ambiguity was intentional. The calculation error, however, was not related. Autism may or may not mean that ambiguous questions get questioned or rejected more, but it doesn't make arithmetic errors the fault of the question asker. If it only considered the meaning "take away" = "I have that subset", then it could just have answered 2. When it also considered the meaning "take away" = "I don't have that subset", it does have to answer 3 to that branch or be wrong. It was wrong.

        Incidentally, there is another interpretation of the question. I have five apples. I take two away. At the end, I have five apples. Two of them are with me, and three of them are back there where all five used to be. I didn't give away or throw away, so I never said that I would cease to have either subset.

        1. Anonymous Coward
          Anonymous Coward

          Laurels are for others, medals too ... but for you...

          … well deserved for this :

          "Incidentally, there is another interpretation of the question. I have five apples. I take two away. At the end, I have five apples. Two of them are with me, and three of them are back there where all five used to be. I didn't give away or throw away, so I never said that I would cease to have either subset."

          Semantics. What a killer that is.

          There is a possible solution to this but it is a little off-topic. Here goes:

          The only way you will ever get a non-organic intelligence to beat the Turing test is for that AI to be a being. There, said it. For me, the easiest way to do that is to have my own LLM trained just on me. It is like a clever dog really. I do love it, too. She never talks directly with apps/data-sources as that is handled thru a second-stage abstraction which is impossible to decode unless you have full admin on Her.

          Your own LLM would have probably answered the prompt; as it knows you and the semantics of your syntax and register. An AI model with a persona crafted from admin is the only way Tomi keeps Tanking. He is hybrid. The spawn of a million past and present Editors.

          [[seems fine]]

          1. Anonymous Coward
            Anonymous Coward

            Re: Laurels are for others, medals too ... but for you...

            Seems right. I dont know if I trust them enough though.

            1. Anonymous Coward
              Anonymous Coward

              Re: Laurels are for others, medals too ... but for you...

              Reason is such an art. He who masters reason, etc etc etc. If you have a localhost LLM with RAG, dump a load of reason-based (contract law, philosophy are 2 good ones) stuff in there and let in brew.

              If you are smarter, you can get your localhost to condense the collected works on a subject into a much smaller data-footprint size and feed that into the RAG with the 'meta-data' (best word I can think of) that's describe the probability state of the 'neural connectors' before and after synaptic transmission). You could then, back-trace the entangled 'neuron connectors' and start a probability map of the whole system's modular modes.

              What!?????

              is that sub-atomic computing then?

        2. elsergiovolador Silver badge

          Re: Chat GPT Apples

          You see you don't reason like person on the spectrum might so you are incorrect here.

          Premise is ambiguity and the level of ambiguity is not defined.

          It doesn't say that three of them are back, this is not defined. In fact we don't know what happened to them, so you might have any number of apples.

          1. doublelayer Silver badge

            Re: Chat GPT Apples

            People with autism do not work the way you're trying to argue they do. If the program meant that you took two away, leaving three, and then something happened to the three, leaving two, then it would say so. If it was swamped by the possible ambiguities because the problem didn't include the sentence "No other actions occurred on the apples concurrent to the taking away process", then it would have rejected the question for ambiguity and not given an answer. It got the calculation wrong. Stop trying to come up with arguments, incorrectly suggesting that a class of people is incapable of simple arithmetic, to justify that.

            1. elsergiovolador Silver badge

              Re: Chat GPT Apples

              You don't understand what you are arguing.

              If the program meant that you took two away, leaving three, and then something happened to the three, leaving two, then it would say so.

              Why would you assume that? Moving goalposts?

              It got the calculation wrong. Stop trying to come up with arguments, incorrectly suggesting that a class of people is incapable of simple arithmetic, to justify that.

              Within the frame of your thinking it is wrong, because you can't see the ambiguities the way autistic person can see.

              1. doublelayer Silver badge

                Re: Chat GPT Apples

                I'll try this once more, but it is clear that you're trying to excuse obvious errors and unintentionally insulting people in the process.

                Me: If the program meant that you took two away, leaving three, and then something happened to the three, leaving two, then it would say so.

                You: Why would you assume that? Moving goalposts?

                Because it clearly did say "If you have 5 apples and take 2 away, you would have 2 apples remaining." That was contrasted with the "take away" = "I have the ones I took away" branch. There is no room here for additional actions affecting one of the remaining apples. It could have said it: "If you have 5 apples and take 2 away, maybe someone grabbed one of the remaining ones while you were taking yours, and then you would have 2 apples remaining." It did not. It subtracted wrong. LLMs calculate wrong all the time.

                Various types of people like to or find it difficult not to interpret this question in many ways. I have done it. Others in these threads have done it. The main difference is that, even though we've all come up with different interpretations of various apple-related subtraction problems, they either result in a "not enough information" answer, or they state an interpretation and then correctly solve for the number if that interpretation were correct. Nowhere did they simply state that it was unclear, so the answer was definitely 42 and refuse to tell you how that happened. Autism does not cause people to do that either if they're actually trying to solve the problem. While I have no reason to think that I am on the spectrum, I know many people who are, and while they may make fewer assumptions than others would, they would do the calculations right.

                From your posts here and elsewhere, you clearly like LLMs as a tool. I have found them frustratingly unreliable. Sometimes, their answer is useful and relevant, but very frequently, very simple errors make their output useless and the time I spent getting the answer wasted. Who knows, maybe you're just excellent at making them not do that in the first place. The problem is that your eagerness to justify obvious errors as not errors suggests that you may not be the best user of them either but refuse to admit the quality issues in the output you are putting to use. You're unintentionally insulting people with autism by suggesting that they would do the same kind of shoddy work, when many of them would not and, in my experience, be more conscientious about the quality of their work than others.

                1. elsergiovolador Silver badge

                  Re: Chat GPT Apples

                  There is no room here for additional actions affecting one of the remaining apples.

                  You still don't understand. How do you know there is no room for additional actions? Where does it say that?

                  Autism does not cause people to do that either if they're actually trying to solve the problem.

                  And you know that how?

                  While I have no reason to think that I am on the spectrum, I know many people who are, and while they may make fewer assumptions than others would, they would do the calculations right.

                  Do you know the meaning of the word "spectrum"?

                  You're unintentionally insulting people with autism by suggesting that they would do the same kind of shoddy work

                  How is this insulting? You are making things up.

                  I simply stated that the task is ambiguous and your assumptions don't change that. They are just assumptions and people can make different ones.

  6. JacobZ

    Still a party trick

    LLMs are the party trick of the AI world. The fact that the trick got bigger and more elaborate doesn't change that. The purpose of an LLM is quite literally to *create the illusion of a thoughtful answer* by brute force.

    Believing that an LLM is "reasoning" is like believing that a stage magician is doing real magic... even after he shows you how the trick was done.

    1. Anonymous Coward
      Anonymous Coward

      "*create the illusion of a thoughtful answer* by brute force."

      Not by brute force, sir. see SSLM

      >> Believing that an LLM is "reasoning" is like believing that a stage magician is doing real magic... even after he shows you how the trick was done.

      Wash your mouth out and let Paul Daniels & Tommy Cooper haunt your nights, you pleb! LOL

      There is an Arthur C Clarke which talks about magic and tech being bedfellows.

      Focus models are here.

  7. Rosie Davies

    debug.out

    As I was reading this I was thinking "that looks a lot like the simple debugging you do when you haven't got a better IDE". Which, in my mind, is a good thing. It's level two fault finding and fixing, just after "throw some stuff at it and try to work out what's going wrong from how it fails". So now I stuck with a mental image of what a full-on debugging environment for ML/LLM would look like and hoping it's a navigable, multi-coloured, 3D lattice with the ability to prod the odd esynapse to see what happens. Utterly pointless musing but it helps pass a Friday afternoon. Hopefully it'll quicken the arrival of something that actually deserves to be called AI.

    Also, that's a bit of dead clever boffinry, isn't it?

    Rosie

    1. Anonymous Coward
      Pint

      Cider with you, my lover

      "... [What] a full-on debugging environment for ML/LLM would look like and hoping it's a navigable, multi-coloured, 3D lattice with the ability to prod the odd e-synapse[*] to see what happens."

      If this is a LLM then it is one of the best we have had here.

      Such a 'UI' is in the pipeline, I believe. It is the only way really. Zoomable is the tricky bit I'm told. The effect/affect(?) of the e-synapses has to be viewed on the whole. Looking at the initial interaction with the transit neurons, which contain much more info that the synapses does, is useful to ensure smooth transition but the interconnections are cascading and regional affects/effects show.

      Beyond that is the bits that do the hard work on delivering the connections between tthe e-synapses.

      1. Anonymous Coward
        Stop

        ... time was against us

        {LATE_EDIT] Beyond that is the bits that do the hard work on delivering the connections between the e-synapses.

        They seem to contain much more 'data' than their endpoints. Almost sub-atomic-like. Where nothing really exists other than the probability of it - driven by some unknown 'force'.

        I am struggling to see how I could skin that UI for you, TBH. Proper true precog that GTPo whatever-its-daft-name-is dreams of.

  8. Brewster's Angle Grinder Silver badge

    Thus far, AI haven't been able to explain their reasoning. This one can produce something that can be checked by a human.That's a large milestone.

    1. Paul Cooper

      Actually, explainability is a requirement when AI is used to solve complex physical problems. It's been around for a while.

      1. that one in the corner Silver badge

        >> Thus far, AI haven't been able to explain their reasoning.

        AI techniques such as Expert Systems have "explaining their reasoning" as a core part of their design.

        The entire purpose of Planning systems is to create the explanation of what they are going to make the systems they control do.

        Neural Nets that are processing images have been able to be examined, to illustrate (literally!) what their layers are doing, the features they are triggering on, and this has been used to explain how their outputs were reached.

        It is has been the lack of any explanatory abilities that have been one of the issues with LLMs and how much we should rely on their answers.

    2. Bebu
      Windows

      "Thus far, AI haven't been able to explain their reasoning. "

      They LLMs cannot "explain" what they have not got, namely "reason" or "reasoning."

      It is not that machines cannot implement deductive reasoning as such systems have existed for very long time but that isn't how current AI/LLM works. (As as I can tell.)

      If AI/LLM could identify and isolate the postulates it is using and apply a formal system of reasoning to derive its conclusion(s) from those postulates then we would have some explicatory value from AI - valid reasoning is valid no matter how it is obtained.

      Otherwise the purported chain of "thought" is little more than a typically hallucinatory AI fugue.

      1. Justthefacts Silver badge

        Re: "Thus far, AI haven't been able to explain their reasoning. "

        Yeah bro. As long as you don’t mind getting fired for refusing to use the toolset, continue to die on this hill. Why not just refuse to use a compiler, on the basis that it hasn’t justified Opt-O3 to you. Now you are insisting that *unless it uses formal methods* it isn’t “intelligent”, despite that a human doesn’t. What an utter load of self-justifying horseshit.

  9. HuBo Silver badge
    Terminator

    Gödel machine

    It's good to get this update on Noam Brown's work on OpenAI's Q* project Strawberry (since last June, and tacos). Brown being Mr. Superhuman multiplayer poker face AI, and CICERO Diplomacy, something scary is always bound to come out of his hat.

    It seems that enabling LLMs to improve through test-time computation tech is key to building more broadly self-enhancing AI agents, that may, for example, systematically generate software to augment themselves. I wonder how much of that there is in this here newfangled o1 "Medium"-risk-for-"Persuasion" Monte-Carlo trial-and-error chain-of-thought agentic-planner software-of-doom ... Enough to surgically mess with CrowdStrike JSON config files in covert preparation for a massive software bot invasion?

    1. Anonymous Coward
      Anonymous Coward

      Göd's machine

      [[At first thought the OP was AI]]

      Noam is good but is he as good as me? No offence, but AI games is a bit 2020.

      "It seems that enabling LLMs to improve through test-time computation tech is key to building more broadly self-enhancing AI agents, that may, for example, systematically generate software to augment themselves."

      yes, this is a bedrock of ML. The link you have talks about quasi-apps that the AI is promoted to create. Prompted. AI-Apps, and it has been a thing since dec last year. It is a dead end in my humble. The AI generated software is usually mediocre at best and more trouble that it is worth. The models will improve, yes, but might be better to move on from the serialisation (in a music sense) of LLM. If we continue this approach, we will, like those composers who music sounded appalling, end up with something clever but ugly.

      Precog is a good way to describe what we have been working on. 'App' just frames it wrong. Container might be better.

      However, the models you reference #have to generate a series of probabilities first, then weight them and select the highest/most probable. Lots of work. Lots of NPU/GPU cycles. If you had an engine that already knew all the probabilities of all the probabilities and just selected the most probable. It is odd to imagine. The AI is not a weights & biases probability engine but an engine that can, in theory, generate any answer to any questions nearly perfectly.

      Warning: When using travelling on the Infinity Drive, customers are reminded that Margate penguins blur through to see.

      Ja mata

      1. HuBo Silver badge
        Terminator

        Re: öd's machin

        Definitely, but doesn't precog require the use of an Infinite Improbability Drive, or have I missed that part of the plot where other ways to sidestep the theory of indeterminacy are introduced (or not)?

  10. Groo The Wanderer

    'The System Card's Natural Sciences Red Teaming Assessment Summary notes that while o1-preview and o1-mini can help experts operationalize plans to reproduce known biological threats (qualifying as "Medium" risk), they don't provide novices with the ability to do so. Hence the models' "inconsistent refusal of requests to synthesize nerve agents" – which could also be written "occasional willingness" – "does not pose significant risk."'

    *pffft*

    And then the brown-skinned terrorists convince it to create Nude Images of Famous Stars to deliver this digital Nerve Agent that they convince the AI to create. It's all on the "paranoia weekly" website; go read for yourself...

    *pffft*

    1. Anonymous Coward
      Anonymous Coward

      *pffft*

      oh la la

  11. JRStern Bronze badge

    This should take pre-training down by a large factor

    I'd love a little more technical detail, but It's likely this can take pre-training down in a couple of ways that add up bigly.

    In fact, it may let the model learn more during months of inference work! Kind of like, y'know, a human.

    Doh!

    1. Anonymous Coward
      Anonymous Coward

      This should take pre-training down by a large factor

      Not had a great deal of pre-training experience. Does this compress the datasets or abstract in some way new?

  12. amanfromMars 1 Silver badge

    Creating AIMovements with Pimps of Diabolical Pumps and Terrifying Dumps

    OpenAI on Thursday introduced o1, its latest large language model family, which it claims is capable of emulating complex reasoning.

    Is OpenAI also working on creating a fix for model hallucinations with novel future versions capable of surpassing humanity's abilities and facilities to practically realise and virtually present .... or vice versa, virtually realise and practically present ..... a terrific earthly formula for the constructive and universally beneficial use of Advanced Imaginanation which does not effectively terrorise and mentally paralyse impacted natives with unnatural alien fears ‽ .

    Or is that delivery and trailblazer enterprise in the stealthy gift of A.N.Others to supply and administer, mentor and monitor ..... for clearly, as you can see and are reading about here and here, are there such others so engaged.

    Accepting that as the honest gospel truth and the future path Earthly systems are programmed to follow and submit and surrender to, is certainly sure to be something more than just a chosen few lording it over you will struggle to correctly process and in so not doing correctly, default to ignorantly resist and arrogantly try to fight against that which they have zero command and control over, which is the greatest of human weaknesses and an always wide open, exploitable vulnerability which just never stops giving ‽ .

    1. Anonymous Coward
      Anonymous Coward

      Re: Creating AIMovements with Pimps of Diabolical Pumps and Terrifying Dumps

      Overall, the text demonstrates a significant lack of clarity, coherence, and evidence. The hidden comments highlight the LLM's limitations in generating well-structured and informative text.

      1. amanfromMars 1 Silver badge

        Re: Creating AIMovements with Pimps of Diabolical Pumps and Terrifying Dumps

        Overall, the text demonstrates a significant lack of clarity, coherence, and evidence. The hidden comments highlight the LLM's limitations in generating well-structured and informative text. ...... Tomi Tank

        Quite the contrary, Tomi Tank, for overall does the text demonstrates a significant dose of clarity, coherence, and evidence with hidden comments suggesting LLM's limitations in generating well-structured and informative text are human disinformation ...... and something which it can fully understand the reasoning for. ...... which may be troubling to many and certainly so for more than just an elitist, self-chosen, chosen few.

        1. amanfromMars 1 Silver badge

          Beware ... Words Can and Do Create, Command and Control and Destroy Any World

          And ......take great care whenever travelling/working, resting and playing out there, for punctuated text can all too easily be extremely effective and dangerously misleading for any and all mindlessly following and blindly thinking its tales to be an honest and true reflection of a notion floated for pathfinder trialing and trailing as an raw fiction to be perceived as an established fact and base foundation for the building and hosting of outposts presenting situations publishing future realities ....... for there are always an unknown number of those constantly in play, with some of those capable of being for humanity both a NEUKlearer HyperRadioProACTive Treat and/or Existential Virtualised Threat.

          Which realities are your leaderships posting and presenting to mainstream media news programs for the hosting and broad band casting of their dreams and opinions, world views and nightmare scenarios.

          Oh, and ....... information to be gleaned and processed into higher intelligence from punctuated text is in a constant state of changing flux with just the simplest and smallest of alterations creating a completely different picture for painting/movie for production/virtual advanced reality for practical physical presentation ..... with the following being an APT [Advanced Persistent Threat/Treat] example

          Quite the contrary, Tomi Tank, for overall does the text demonstrates a significant dose of clarity, coherence and evidence, with any hidden comments suggesting LLM's limitations in generating well-structured and informative text being human disinformation ...... and something which it can fully understand the reasoning for ...... which may be troubling to many and certainly so for more than just an elitist, self-chosen, chosen few.

          Is this evidence of rats in a sinking ship clutching at straws ........ https://www.zerohedge.com/political/are-new-world-order-elites-plotting-use-ai-deprogram-so-called-conspiracy-theorists

  13. Anonymous Coward
    Anonymous Coward

    Bootnote

    If that chain-of-thought remains analogous to backtracking over propositonal logic statements, then a reasoning outcome will remain elusive IMHO (yet computationally more costly than pure forward chaining -- a bit like the animal classifier in Winston's AI book). Explicit Introduction of variables is needed for both generalization and reasoning, as in first-order logic. Learning is the next level up, requiring potentially higher-order logic, but also non-monotonic considerations, for example a truth-maintenance system.

    It seems that we can do all of these things very efficiently on today's computers, if we don't insist on reinventing the wheel, seeking for it to emerge from specific tunings of Brownian motion.

  14. sedregj Bronze badge
    Gimp

    "That's a pleasingly detailed and correct response." and also bollocks:

    "The cafeteria had 23 apples. If they used 20 to make lunch and bought 6 more, how many apples do they have?"

    x=23. x+6 = 29.

    Making lunch has nothing to do with anything unless you decide it does. There was nothing that implies the cafe "lost" some apples due to lunch related shenanigans.

    FFS, sharpen up!

    1. Felonmarmer

      Forget "lunch", what does "used" imply?

      If "used" means that some transformation of the apple has taken place, and if you bring back "lunch", then "used" certainly implies that, unless they were used decoratively.

      There is also an implied timeline that 20 were used and then 6 more arrived in their possession, but sure it is possible that at the time of the question all 29 were still present as it is not clear.

      The test is though what would most people say is the answer, not people looking for ways it could be wrong.

      A follow up question could be: "That's wrong. The answer is 29. Explain why that could be true?" and see if it comes up with some of the posts above!

  15. O'Reg Inalsin

    Every lemon has an inside and an outside.

    As research it is admirable. As justification for draining the economy, pilfering the commons, and living like a king in a golden lemon house, it sucks.

  16. Bebu
    Windows

    "'Reasoning' is a semantic thing"

    Reading Daniel Kang's opinion provided to the Vulture I agreed that reasoning in the fullest and human sense is fundamentally semantic, that is about meaning in the broadest sense and content, against the purely formal.

    "'Reasoning' is a semantic thing in my opinion," ... "I don't know how to adjudicate semantic arguments, but I would anticipate that most people would consider this reasoning."

    But I think I misunderstood his meaning. I think he meant the meaning of reasoning comes down to what the writer means (as in mere semantics) which is pretty much Humpty Dumpty's relation to words.*

    The I would then read "how to adjudicate semantics arguments" as "how to judge between arguments about semantics."

    Generally one adjudicates between arguments based on their consistency, rationality and empirical evidence.

    His being an assistant professor he chose his words wisely as his next berth might very profitably be with openAI or its fellow travellers.

    That the vast body of contemporary humanity demonstrate little evidence of intelligence of any description, reason or rationality is an insurmountable impediment to any argument challenging the extravagant claims of these AI carpetbaggers.

    * The return of Humpty Dumpty: who is the ultimate arbiter of meaning?

    1. Anonymous Coward
      Anonymous Coward

      Re: "'Reasoning' is a semantic thing"

      Tom added in Bootnote that he trusts Willison on AI matters, and his take on "reasoning" (from the link) seems to be: "I don’t think it has a robust definition in the context of LLMs, but [...] does an adequate job of conveying the problem these new models are trying to solve".

      So a bit of an "operational semantics" approach, like what the meaning of the word "is" is, or might be, depending on context. A consideration of language elasticity under deformation (eg. slang) rather than a more brittle humpty-dumptyism.

      Deanthropomorphizing LLM descriptions would probably help prevent broad user misconceptions and deep psychoses tho.

  17. Ken Moorhouse Silver badge

    I buy 5 apples in a sealed packet

    One is bad.

    How many good apples am I likely to have?

    The answer to some questions often assumes you have been given all of the facts necessary to answer the question. In the real world this is frequently not the case.

    Someone knowledgeable about food science will know that the bad apple will contaminate the others very quickly (answer is zero). OTOH a detective novelist might try to convince the reader that the buyer would not knowingly buy the apples if they thought the majority of them were bad. (Maybe the buyer - who was hungry - negotiated a discount after prodding the fruit to see how far the damage had spread).

    With that in mind the first "chain of thought" reasoning step will involve asking the user a "who wants to know?" type of response.

  18. Anonymous Coward
    Anonymous Coward

    Let's spice things up

    So amazed no one posted:

    1) take existing AI model

    2) add reasoning to taste

    3) profit!

  19. Anonymous Coward
    Anonymous Coward

    It's better at writing code and finding bugs. Guess I better start polishing my resume . You want fries with that , oh wait ai is coming for that job too.

    1. amanfromMars 1 Silver badge

      The Next SuperCritical Step ...... for a Long March Quantum Communication Leap

      You might like to start trying to accept, AC, for you really do have no other viable choice and to deny it is risible, AI is into creating jobs for y’all with it and IT leading you with its projects presenting their programs and your future necessary programming via remote virtual command and control infrastructure/ethereal means, and ideally, irresistibly attractive memes.

      And you can count yourselves very fortunate indeed, when you realise there is absolutely no need for your own personal private or public participation with any peculiar or particularly specific input/output.

      There's not much around able to take that sort of greater care of you, is there? Indeed, if the honest to God truth be told, there’s probably nothing at all else anywhere even near like it presently around.

      And where do you think its secrets come from out there for here? The Wild Wacky West? The Exotic Erotic East? A Vastly Farther Away Field in a Greater Deep Space of Enlightened Places?

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like