The wonders of AI…

Advert

The wonders of AI…

Home Forums The Tea Room The wonders of AI…

Viewing 25 posts - 1 through 25 (of 30 total)
  • Author
    Posts
  • #780920
    Robin Graham
    Participant
      @robingraham42208

      Tonight I wanted to check my recollection that the coefficient of thermal expansion for steel is around 10 ppm per Kelvin  so I Googled “steel coefficient of thermal expansion”.

      Nowadays Google prefaces its results for relevant links with an ‘AI overview’. Verbatim:

      The coefficient of thermal expansion for steel is 0.0000065 inches per degree Fahrenheit (/F) This means that for every degree Fahrenheit that steel is heated , it expands by 0.0000065 inches.

      Obviously nonsense.

      The Google AI overview does give links to source material.   The source here is metalsales.us.com who say (verbatim):

      Δ Length (in inches) = Length (in inches) * Δ Temperature (in °F) * ε (in  / °F).

      Each metal has its own response characteristic, ε, known as the coefficient of thermal expansion. ε for steel is 0.0000065 /  °F.

      That’s fine for me because I can tell that that when they give ε ( in  / °F they  mean ‘in units of reciprocal °F’ . Not so for AI – it ‘cleverly’ recognises ‘in‘ as an abbreviation for inches.  Well I suppose that’s what happened!

      I’m thinking it’s a double  edged sword, but maybe that’s because I’m old..

      Robin

       

       

      Advert
      #780931
      DC31k
      Participant
        @dc31k
        On Robin Graham Said:

        The coefficient of thermal expansion for steel is 0.0000065 inches per degree Fahrenheit (/F) This means that for every degree Fahrenheit that steel is heated, it expands by 0.0000065 inches.

        The response is incorrect because it omits a couple of words (which the AI might assume, allowing for some charity on your part, you would infer from the context).

        Had it said “This means that for every degree Fahrenheit that A ONE INCH LONG BAR OF steel is heated…”, it would be correct. It is a case where lack of verbal precision leads to lack of factual correctness.

        What complicates matters is the use of Fahrenheit, so you would have a different thermal expansion number for deg. F to that used for deg. C or Kelvin. Training the AI to know your preferred choice of units could be difficult.

        #780962
        Nigel Graham 2
        Participant
          @nigelgraham2

          The units are consistent in using both ºF and inches, as any table published using one of those is likely to use the other; but I wonder how many people would take that answer at face value and go wrong as a result.

          It’s more than verbal imprecision, it’s a wrong definition.

          #780968
          Michael Gilligan
          Participant
            @michaelgilligan61133

            Out of mere curiosity … I tried Robin’s search phrase on ChatGPT

            The response was very different, and nicely ‘conversational’

            .

            IMG_0562

            .

            Edit: __ So, just for the sport of it, I replied: please express in kelvin

            … and it did so, rather well.

            .

            I have not tried DeepSeek

            MichaelG.

            #780975
            Martin Connelly
            Participant
              @martinconnelly55370

              The wonder of AI for me is that anyone trusts anything it says at the moment. The more ignorant the user is the more crazy the AI output seems to be because it responds to the question asked, not with intelligence but with guesses at what the questions is about. A fine example of that long standing computing acronym GIGO.

              Martin C

              #780977
              Michael Gilligan
              Participant
                @michaelgilligan61133

                Martin

                As I have just demonstrated non-garbage questions can produce pretty-good answers.

                MichaelG.

                #781116
                JohnF
                Participant
                  @johnf59703

                  Well I agree with Martin, AI is not so intelligent after all !  Frankly we all need to question everything found on the internet.  Yes its an amazing resource BUT there is a great deal of misinformation therein including in Wiki !

                  We are entering a frightening era hopefully it will come right but keep your eye firmly on the ball and don’t trust AI

                  John

                  #781336
                  Michael Gilligan
                  Participant
                    @michaelgilligan61133

                    I know it’s only a machine, but … picking-up on my comment about ChatGPT being ‘conversational’ : Here’s a screenshot from the end of a very brief discussion about Bernstein’s  America, and its 1968 performance by The Nice.

                    It’s getting spooky, in a Turing-test sort of way.

                    MichaelG.

                    .

                    IMG_0566

                    #781342
                    Nigel Graham 2
                    Participant
                      @nigelgraham2

                      The difference there is that the first question was a straight enquiry about a fact of metallurgy; the second a matter of musical taste though with facts of composers and performers.

                      Objective knowledge or subjective opinion.

                      So perhaps AI is more reliable for information if asked properly about incontrovertible facts.

                      Though I still prefer Real Intelligence and can’t help thinking the ‘I’ of ‘AI’ could equally stand for “Idiocy”…..

                       

                      #781348
                      Michael Gilligan
                      Participant
                        @michaelgilligan61133

                        Sorry, Nigel … I really don’t follow your logic

                        My first test was to use Robin’s exact question on ChatGPT, to compare its response with that from Google … and it won by a ‘country mile’

                        My second test was simply to see how ‘conversational’ ChatGPT might be.

                        MichaelG.

                        .

                        Edit: __ Just for the sport of it, try asking ChatGPT

                        How does your performance compare with Google’s bot, and why

                        #781350
                        Martin Kyte
                        Participant
                          @martinkyte99762

                          The real test is when the AI designs the component. That’s where you find out what it really knows about thermal expansion.

                          #781353
                          Clive Foster
                          Participant
                            @clivefoster55965

                            Fundamentally AI is a stoichiometric parrot working by deductive logic.

                            Stoichiometric basically means popularity weighting with a lot of statistical “stuff” interposed.

                            Most of the statistic involved are more than a little dubious. Having spent much of a decade working with a fairly arcane aspect of single photon counting based on the engineering application of statistically rare, typically one in a million or less, but reliable events my views on correlation statistics as applied to the softer sciences are more than little jaundiced. The next person who starts wittering on about P=20 outside of a controlled laboratory experiment or well understood engineering is at serious risk of such a clout up the left ear’ole. Sitting in a lab watching a pproachinf 50 thousand live graphical representations for 2, 3, 4, or 5 minutes as a single photon experiment running at 20 MHz or so grinds out sets of one in a million or three events against both random and non random background generates a serious, but extremely bored, feel for whats going on in complex environments.

                            Deductive logic basically means an internal consistency seeking engine.

                            Unfortunately deductive logic manipulates only false statements.

                            Science and engineering places boundaries within which deductive logic may be safely applied in “lies to children” fashion. Typically using statements of the form “provided the system lies between the limits of X and Y with error bars of Z being acceptable the results of calculations using equation A may be taken as physically correct”.

                            Given there right sort of starting points so called AI is eminently capable of disappearing over the angels dancing on a pin event horizon. I’ve no faith in the ability of the folk developing these things to keep them grounded in reality at the near 100% reliability needed for serious use.

                            They are obviously going to be reliable in anything resembling quotations from literature, news reports and the like because this are at the core fiction and fictional consistency is what’d deductive logic does well. Engineering, science and the rest of the real world is different and needing reference to facts and maths. What to do with the pseudo sciences like economics et al I don’t know.

                            The classic car and motorcycle buffs on Facebook seem to be having a great deal of fun pointing out the errors in AI generated posts showing picture of vehicle and discussing its history, specifications et al. One day they might get things right but so far normal performance is to not only discuss a vehicle vastly different from the picture but also get the facts mostly wrong once its been figured out what its actually blathering on about.

                            Clive

                            #781356
                            SillyOldDuffer
                            Moderator
                              @sillyoldduffer
                              On Nigel Graham 2 Said:

                              The difference there is that the first question was a straight enquiry about a fact of metallurgy; the second a matter of musical taste though with facts of composers and performers.

                              Objective knowledge or subjective opinion.

                              So perhaps AI is more reliable for information if asked properly about incontrovertible facts.

                              Though I still prefer Real Intelligence and can’t help thinking the ‘I’ of ‘AI’ could equally stand for “Idiocy”…..

                               

                              Indeed, but human intelligence is extremely untrustworthy too.  Think spam, faulty memory, misunderstandings, joining the dots, wishful thinking, prejudice, ignorance, stupidity, small-c conservatism, mental health issues,  conflation, cognitive dissonance, gut reaction, outdated knowledge, missing facts, drawing wrong conclusion from  evidence, not understanding logic and scientific method etc.   We also like to be told what we want to hearn not uncomfortable reallities!

                              Last year AI achieved a breakthrough, but like Newcomen’s steam engine, the technology has a long way to go. Only 80 years since Turing identified computational limits, and the technology is still immature.

                              At present AI is best with facts, and poor at filtering out human nonsense.  So are people!

                              AI gets really exciting when it can’t be distinguished from human intelligence.  We’re not there yet!  As far as I know though, there is nothing that AI cannot do within the bounds of known information theory.  A big plus is it’s unlikely throw emotional wobblers.  For the moment AI makes GIGO mistakes; be interesting to see if it learns to lie from us.

                              Dave

                               

                              #781359
                              Michael Gilligan
                              Participant
                                @michaelgilligan61133

                                Further to my edit … Having received a rather good answer to that question, I then wrote:

                                The question was prompted by the fact that a friend received an inaccurate definition from Google’s AI overview when he asked about The coefficient of thermal expansion for steel

                                … and received what I consider to be an excellent response

                                Try it for yourself, Nigel … just copy & paste my text … I’m assuming that identical questions will produce at-least-very-similar responses.

                                MichaelG.

                                #781365
                                Michael Gilligan
                                Participant
                                  @michaelgilligan61133
                                  On SillyOldDuffer Said:
                                  […]
                                  For the moment AI makes GIGO mistakes; be interesting to see if it learns to lie from us.
                                  Dave

                                   

                                  What is interesting, right now whilst I’m playing with ChatGPT, is that it is [very politely] attempting to teach me to ask better [i.e. more answerable] questions

                                  … That’s very impressive.

                                  MichaelG.

                                   

                                  #781378
                                  Michael Gilligan
                                  Participant
                                    @michaelgilligan61133
                                    On SillyOldDuffer Said:
                                    […] be interesting to see if it learns to lie from us.
                                    .
                                    I’ve started February by ‘baiting’ ChatGPT with:
                                    Please provide a convincing but fallacious logical argument that “black is white”

                                    The almost instantaneous response was better than many humans could manage.

                                    MichaelG.

                                    #781383
                                    Michael Gilligan
                                    Participant
                                      @michaelgilligan61133

                                      Must stop playing and go shopping … but my latest interaction was to first ask:

                                      Describe the Moebius Strip

                                      and then

                                      is there no end to this 🙂

                                      The responses were both very [dare one say] “Intelligent”

                                      … Mmmm

                                      MichaelG.

                                      .

                                      Edit: __ I’ve just closed with the compliment:

                                      Touché

                                      but it wants to keep playing !

                                      #781427
                                      Martin Johnson 1
                                      Participant
                                        @martinjohnson1
                                        On Michael Gilligan Said:

                                        Out of mere curiosity … I tried Robin’s search phrase on ChatGPT

                                        The response was very different, and nicely ‘conversational’

                                        .

                                        IMG_0562

                                        .

                                        Edit: __ So, just for the sport of it, I replied: please express in kelvin

                                        … and it did so, rather well.

                                        .

                                        I have not tried DeepSeek

                                        MichaelG.

                                        So can you get Chat Gpt to explain how the stainless CTE of 16 to17 /1000000 is less than that for carbon steel at 11 to 12 / 1000000.

                                        Martin

                                        #781464
                                        Michael Gilligan
                                        Participant
                                          @michaelgilligan61133

                                          I will let you show us how it should be done, Martin … but my humble effort has at least extracted recognition of the error AND an apology

                                          First I quoted your text, and it answered

                                          Then I wrote: But in a previous discussion you claimed it was lower

                                          That was all it took !

                                          No … it’s nowhere near perfect, but it’s already better than a lot of people.

                                          MichaelG.

                                          #781500
                                          Dod Mole
                                          Participant
                                            @georgeclarihew

                                            I am amazed that AI was not started as AS (Artificial Stupidity) and teach it to be better rather than assuming that the code that what they wrote knew more than the sum of all that humanity has taken ‘undreds of fousands of years to learn.

                                            Or is it nowadays the computer knows best.

                                            #781510
                                            Michael Gilligan
                                            Participant
                                              @michaelgilligan61133

                                              No … The computer [for the moment] simply has rapid access to the fund of human knowledge.

                                              It’s the next-steps that will be really interesting

                                              MichaelG.

                                              #781518
                                              Nealeb
                                              Participant
                                                @nealeb

                                                While it’s probably right to say that AI systems when they are “trained” have access to the fund of human knowledge (or at least that part of it represented by that which is available via the Internet), it’s not really knowledge as we know it that is extracted. What they are actually doing is analysing the text, the series of words and phrases, that it is given in training and when it responds to a question it gives back out some kind of representation of the words and phrases that it found in connection with key words and phrases in the question. I suppose that you could almost think of it as a fantastically clever search engine that is able to go far beyond simple text searching – but at no point can it be thought of as intelligent insofar as it does not have any understanding of the meaning of what it finds and presents back. It is remarkably clever at extracting some kind of interpretation from a question it is given, and putting together a syntactically correct answer that is of relevance. Syntactically good, but semantically unreliable – it does not have any understanding of what it is saying even if it can produce grammatically-correct English. If it is trained on unreliable data, it will produce unreliable output. Imagine an AI system that is trained entirely and exclusively on the content of social media with emphasis on “Truth Social” and its ilk!

                                                Personally, I find it a bit annoying to do a Google search and be given the “AI” response first, when an entry or two lower is a link to the Wikipedia article that seems to have provided the content.

                                                From a technical point of view, these systems are absolutely incredible and that they give good answers so much of the time is close to a miracle, but they have a long way to go yet. On the other hand, something like a specialised system for, say, analysing X-ray data can be trained on relevant and verified input data, which can allow it to improve performance over the experts who provided the input data.

                                                 

                                                #781574
                                                Paul L
                                                Participant
                                                  @paull58212

                                                  All AI models get stuff wrong.The earlier the mode,l the more they ‘Hallucinate’. The latest Chatgpt model available is 01 (03 will be released in the near future) the current public model is GPT 4 .

                                                  01 model is a HUGE step forward from 4, and 03 is another massive step forward.

                                                   

                                                  An example of how badly the gpt 4 model is this….

                                                  You said:
                                                  principles and guidelines of ISO 12181
                                                  ChatGPT said:
                                                  ChatGPT
                                                  ISO 12181 is a standard that provides guidelines for the testing and management of the functional performance of industrial and commercial vehicles, particularly focusing on materials handling equipment like forklift trucks. The principles and guidelines it establishes ensure safety, efficiency, and performance in the operation of such vehicles, helping to standardise their design, construction, and testing processes.

                                                  Some of the key principles and guidelines outlined in ISO 12181 include:

                                                  Safety: Ensuring that vehicles are safe for operation in various environments, especially considering risks such as tipping, load handling, and operational stability.
                                                  Performance Testing: Establishing clear methods for evaluating the functional performance of vehicles, ensuring they meet required standards for tasks like lifting, load capacity, and manoeuvrability.
                                                  Design Standards: Defining design requirements for different types of industrial vehicles to ensure they can operate efficiently and safely under standard conditions.
                                                  Maintenance and Inspection: Providing guidelines for the ongoing maintenance and periodic inspection of the vehicles to keep them in optimal working condition.
                                                  Environmental Conditions: Considering the impact of the operational environment, such as temperature, humidity, and surface conditions, on the vehicle’s performance and safety.
                                                  Equipment and Accessory Compatibility: Offering specifications for accessories or add-ons that might be used with the vehicles, ensuring they integrate correctly with the overall design and function of the machine.
                                                  These guidelines help manufacturers, operators, and safety authorities ensure that vehicles meet a certain standard for reliability and safety in material handling contexts.

                                                  If you’re looking into more specific details or a particular aspect of ISO 12181, feel free to ask!

                                                  You said:
                                                  i dont think that is correct
                                                  ChatGPT said:
                                                  ChatGPT
                                                  Apologies for the confusion earlier! You’re right to point out the discrepancy.

                                                  ISO 12181 actually relates to the design and testing of vehicle components, specifically in the context of pneumatic tyres used on vehicles, such as those for material handling. The correct context involves guidelines for ensuring that these tyres meet certain performance standards.

                                                  To clarify the key principles and guidelines of ISO 12181:

                                                  Tyre Specifications: The standard specifies the criteria for the design, performance, and testing of pneumatic tyres for vehicles, ensuring they meet minimum safety and quality standards.
                                                  Testing Procedures: It outlines the methods for testing the tyres to ensure they can perform under specific conditions, such as load capacity, temperature, and durability.
                                                  Performance Metrics: ISO 12181 defines key metrics like rolling resistance, traction, and wear characteristics, ensuring that the tyres perform optimally in different environmental conditions.
                                                  Safety Standards: The standard helps in maintaining vehicle safety by ensuring the tyres are reliable under various operating conditions, reducing risks related to vehicle handling and stability.
                                                  Compatibility with Vehicle Types: ISO 12181 offers guidance on choosing the correct type of tyre for different vehicle types, depending on the load, terrain, and performance requirements.
                                                  Once again, my apologies for the earlier mix-up! Does that align more with what you’re looking for, or is there another aspect you’d like me to elaborate on?

                                                  You said:
                                                  no iso 12181 concerns roundness measurments
                                                  ChatGPT said:
                                                  ChatGPT
                                                  You are absolutely correct, and I apologise for the previous confusion. ISO 12181 is indeed focused on the measurement of roundness of components, particularly in mechanical and engineering contexts.

                                                  ISO 12181 provides guidelines on measuring the roundness of circular parts, such as shafts, bearings, and other components, ensuring they meet specific tolerances for proper functionality. The standard is crucial in industries where precision and quality are critical, such as in automotive or manufacturing settings.

                                                  Here are the key principles and guidelines of ISO 12181:

                                                  Roundness Measurement: The standard outlines methods for measuring the roundness of a workpiece, ensuring it adheres to the required tolerances. This is particularly important for ensuring smooth operation in machinery, where any deviation from perfect roundness could lead to wear or mechanical failure.
                                                  Measurement Tools: ISO 12181 defines the types of instruments that can be used for roundness measurement, such as roundness testers, profilometers, and coordinate measuring machines (CMMs). These tools must be properly calibrated to ensure accurate results.
                                                  Testing Methods: The standard provides detailed procedures for performing the roundness test. This includes the use of various techniques, such as the least-squares method, to assess the form and roundness of the component.
                                                  Tolerance Guidelines: ISO 12181 establishes specific tolerance limits for roundness, allowing manufacturers to control and minimise variations in circularity, ensuring that parts function properly when assembled into larger systems.
                                                  Data Analysis and Reporting: It offers guidelines on how to interpret the measurement data, including the presentation of results in a clear and consistent manner for quality control and certification purposes.
                                                  Applications: The standard is commonly applied in the production of precision components, particularly in automotive, aerospace, and industrial machinery sectors, where maintaining roundness is vital for performance and longevity.
                                                  Thank you for your patience, and I hope this now correctly addresses your query! Let me know if you’d like any further details on specific aspects of ISO 12181.

                                                  #781577
                                                  Michael Gilligan
                                                  Participant
                                                    @michaelgilligan61133

                                                    An interesting example, Paul !

                                                    .

                                                    By comparison, the results from a simple search on the ISO website are succinct

                                                    MichaelG.

                                                    .

                                                    IMG_0569

                                                    #781604
                                                    Paul L
                                                    Participant
                                                      @paull58212

                                                      I already knew what the standard was pertaining to.

                                                      I asked for the ‘principles and guidlines’ to try and get some more details to save me from purchasing the standard.

                                                       

                                                       

                                                    Viewing 25 posts - 1 through 25 (of 30 total)
                                                    • Please log in to reply to this topic. Registering is free and easy using the links on the menu at the top of this page.

                                                    Advert

                                                    Latest Replies

                                                    Home Forums The Tea Room Topics

                                                    Viewing 25 topics - 1 through 25 (of 25 total)
                                                    Viewing 25 topics - 1 through 25 (of 25 total)

                                                    View full reply list.

                                                    Advert

                                                    Newsletter Sign-up