Thanks so much for putting this talk online, I now have a much better grasp of what generating sentences by using the next likeliest 'word' (really token) means in practice and why ChatGPT works and doesn't work at the same time.
Interesting! Just for the fun, I asked chatGPT whether you are right. Here's the conversation: Me: Is gerben wierda correct with his explanation of how gpt works? GPT: Without specific details or context about Gerben Wierda's explanation of how GPT works, I can't provide a definitive answer. Gerben Wierda is known for his work in various fields, including information technology and project management, but without a specific explanation to reference, I can't comment on its accuracy. If you provide more information about what exactly Wierda said about GPT, I can try to offer insights or assess its accuracy based on current understanding of GPT models as of my last update in January 2022. Me: He states that llm works on tokens and the gpt output is not based on understanding, but on calculating the most probably next answer in a sequence. GPT: The description provided by Gerben Wierda seems to align with a fundamental understanding of how large language models (LLMs) like GPT operate. 1. Tokens: LLMs indeed work on tokens. These tokens are essentially discrete units of text, often representing words or subwords. The model processes input text by breaking it down into these tokens and then generates output by predicting the next token(s) in a sequence. 2. Output Generation: The output of LLMs like GPT is indeed generated based on probabilities. The model doesn't "understand" language in the way humans do; rather, it has been trained on large datasets and learns statistical patterns in the text. When generating output, the model calculates the probabilities of various tokens occurring next in the sequence based on its training, and it selects the token with the highest probability as the next output token. This process continues iteratively to generate a sequence of tokens that form the output text. So, in summary, Wierda's description appears to provide an accurate high-level explanation of how GPT operates, focusing on token-based processing and probability-based output generation rather than on true comprehension or understanding of language.
Thank you for this concise and enlightening look behind the scenes of GPT - deserves to be seen by many more. Entirely based on facts and down-to-earth you lifted the curtain of ChatGPT's 'magic' and helped me understand some of the nuances of how it works. Recommended to anyone to better position realistic use cases for GPT.
Absolutely fascinating talk. Compelling, well-crafted. I do have to wonder where the line is? Listening to young children learning language really quickly demonstrates that they are basically statistics machines. They learn the words and the patterns far quicker than they gain the understanding of all those words. I've seen kids just throw words together with no clear understanding of what they are actually saying but simply seeking a desired response. I've always wondered how much of us remain those statistic-machines that mostly speak in set patterns. This talk actually alludes to it when our presenter speaks on how "conviction" often precedes observation and reason. We go shopping for the facts that fit our currently held beliefs. Our pattern detection systems can certainly be disrupted but it's often challenging. What does this mean? I don't know. I want to believe that LLMs aren't conscious, it's a far more comforting thought and the evidence here is strong but how far are they from understanding? Are the centers, the hubs, in our own brains that form around languages or other skills just LLM-like statistics-machines that work to predict the patterns of stimuli we encounter in the environment? Could we string multiple LLMs together, trained and tuned differently, in a way that started to approach some rudimentary 'intelligence?' After all, one of the reasons teachers and students "show their work" with math problems is because our own prediction machines sometimes skip steps or intuit a wrong answer. Is a math-tuned model possible? Maybe not as an LLM but a separate type of system that an LLM could draw upon? I certainly don't know the answers here but I do feel my intelligence is emergent and not so special as to be impossible to reproduce. Very thought-provoking (and speaking of prediction machines, it's funny that RU-vid put this in my suggested videos even though it has only 580 views at the moment and none of my other suggested videos are AI-related -- seems YT's machine-learning is working pretty well).
Thank you. The remark on children learning makes me think about how Uncle Ludwig handles meaning in (hence understanding of) language. It boils down to "Meaning of language elements lies hidden in what a community of people with shared experiences consider correct use". Basically, children learn by *using* elements and finding out from feedback if they are considered *correct*. (Uncle Ludwig really was spot-on if you ask me). If that is correct, then token prediction resembles understanding because it is directed at 'correct' use. That is why it is so often so close. But it is not correct use of the same elements we see as elements (words and phrases, and even intonations and such), but something very technical: the tokens (n-grams, I recall that in the early 1990's there even was a patent on using n-grams, but I digress). While the correct use of words and phrases is how we humans learn understanding (answerable with correct/incorrect, so true/false), the 'guessing-type' use of tokens (effectively out of a set of close alternatives) is how these models do it. There are some deep questions lurking here. For instance, can there be understanding without experience? (My guess would be 'no or almost entirely no' - which would have consequences for LLMs as well) Can we build something better out of multiple LLMs? We might (The cool Cicero application from Meta bearing some resemblance to that approach), but the issue of size explosion might make it impractical, in the same way that an almost infinite set of rules (GOFAI's Cyc approach) runs into physical limitations quickly (especially on performance). We might have to build things that are 10k-100k as large and will that ever be feasible? Digital approaches are quickly overwhelmed, which is a fundamental problem too if you ask me. More on the conviction-story here: ea.rna.nl/2022/10/24/on-the-psychology-of-architecture-and-the-architecture-of-psychology/ And if you like it (which you do) I ask people to actively spread it in their circles. I (and others) think it might help if this explanation was a little wider seen.
Thank you, very interesting and well presented. I would loved to hear something on multi-modality. Stephen Wolfram thinks that the GTP-4 has a language-independent metamodel of ”words”. So no sparks of the AGI yet?
@@ppulkkinen5308 Thank you. I wonder what "a language-independent metamodel of 'words'" would mean (I could not find it), unless that stands for the embeddings of tokens. I ask everyone who thinks this is worthwhile to watch to actively spread it in their own circle. It's pretty hard to get realism a place, given the fever currently out there.
Sorry @@ppulkkinen5308, I forgot AGI. No. These models 'understand' token/pixel distributions. That is like understanding 'ink distributions' of a printed text. With this understanding of 'ink distributions' the can approximate the results of 'text/image that can be understood by human understanding'. There are no correct and erroneous results (no 'hallucinations') only successful and failed approximations. See ea.rna.nl/2023/11/01/the-hidden-meaning-of-the-errors-of-chatgpt-and-friends/ (and more in the series of which that post is a part, and that grew out of this presentation)
I would add a comment about the Change Architecture section regarding the heart transplant. There is a distinction between switching out the ESB at a project/department level vs the Enterprise. Decisions at the enterprise that happen at the CIO level are normally non-negotiable (due to vendor issues typically) and impacted teams need to make judgment calls. Best impacted teams can do is delay the up coming heart surgery and cannot opt to not do it.
Heel interessant verhaal Gerben! Je geeft aan dat sociale media ingericht zijn op positieve feedback en derhalve niet aan de negatieve feedback eigenschappen van roddel voldoen. Dat lijkt me grotendeels een juiste waarneming maar veel toepassingen verschillen van gossip omdat ze een “publiek” karakter hebben ipv een (semi-) -vertrouwelijk karakter. Publiek corrigeren in je eigen bubbel is moeilijker dan vertrouwelijk negatieve feedback genereren. Of zie ik dat verkeerd. Andere mooie waarneming is dat we ons organiseren rond onze IT-bergen en dat moeten accepteren ipv IT follows business. Hartelijke groet.
Ik denk dat je een goede toevoeging hebt. Er is in social media vaak wel een klein private gossip kanaal, DMs b.v. Maar grotendeels is het inderdaad publiek en dat heeft een andere dynamiek. Jouw observatie klopt dus ook wel met mijn waarneming dat je in die social media slecht aan 'private negative feedback' toekomt, de zwakte van het private (DM) kanaal versus het publieke kanaal zou dan een aanvullende reden zijn. Dus een mooie toevoeging. Overigens is roddel niet de enige vorm van stabilising negative feedback, het is meer het voorbeeld uit de Dunbar-size dat ik er bij heb gehaald. En ik heb het in mijn verhaal vaak over 'talk' en niet alleen over 'gossip'. Maar de sheer size van public over private is een aardig inzicht om hier in mee te nemen.
Great presentation! What is your take on the people thriving in the positive feedback loop, refusing to accept the truth, content with the state of things and using their position to make sure that you fall in line.
That they are typical humans. That all of this says something about humans and our limitations (all of us). As said at the end of the presentation: the IT revolution is teaching us some things about ourselves and above all our limitations. I think the question before us is if we as a species are capable of learning the lesson, and manage the essence/limits of our psychology/intelligence (take your pick). As far as I'm concerned, the jury is still out, but the prospects are not looking good unless a new paradigm, a new shared belief, conquers humankind (which can happen).
More in depth (and more up to date, e.g. on agile/devops and architecture: ea.rna.nl/ And have a look at the other (longer) videos in this channel. There are only a few, but they are more in depth (thus also longer).
Fascinating, especially the insight on negative feedback needed to stabilise social media systems. This is a loophole that has enabled presidents to get elected based on misinformation.
Dear Friends, I have read TOGAF v9.1, I have some questions to ask again so that I understand correctly and fully, the questions are as follows: 1/ The ABB (architecture building blocks) will be technology independent, right?, and it is used to describe Target models in Phase B,C,D? 2/ The SBB (Solution building blocks) is tied to specific technologies (eg Cloud, Microservices, etc.), right?, and it is used in Phase E?. 3/ The Technical reference model is used in Phase E, right?.
Dear ytuong, EA is not the same as TOGAF, nor is the reverse true. In my opinion TOGAF is seriously ineffective in the real world. You may have a look at the Chess and the Art of Enterprise video (EAC 2016 keynote) in my channel. Feel free to use TOGAF, but in the countries where it was first promoted it has never gotten a serious foothold for the reasons explained in that keynote.
@@ntcuong01ct1 There is not a framework I can advise. I do not use a framework at all, unless you consider 'EA Chess' a framework ;-). ru-vid.com/show-UCTqv-dxJc_nMsZqBYJW3iCw has two keynotes from Enterprise Architecture Europe on EA (Chess and the Art of Enterprise Architecture from 2016 and Architecture in an Age of Agile form 2018). SAFe pays some attention to architecture, this contains some useful suggestions but for me has the status of ideas, more than a real mature practice.
Not every form of meeting is 'committee & bureaucracy' and the animation does not propose that. We need collaboration (as nobody can know everything about everything) and we need a consent-based checks & balances. A short video like this cannot go into depth on these and the example given of the checks & balances we need is not by definition bureaucracy. It can be very pragmatic and fast. Having said that, this video is 8 years old and was made to explain it to people in a setting with projects and up-front project design. The architecture approach itself was already agile, but the projects at that time were still classic (in that case, EA became agile before the change organisation did).
@Gerben Wierda, really appreciate the effort in making such a simple and straight forward video. Hats off to all the drawings and the person behind the drawings :)
Just awesome! Enterprise architecture is much more than just IT. Its basically defining how the enterprise should function to be effective, Robust and flexible at the same time. Enterprise here mean Business/NGOs/Government/Group. In plain and layman terms, Enterprise Architecture is needed to translate human idea/s into reality in a systematic and time bound manner.
See ea.rna.nl/2015/04/27/the-great-escape-ea-is-not-about-it/ and ea.rna.nl/2017/01/13/agile-architecture/ and the video ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-d10A21OepgA.html
That, I think, is understandable. Same goes for any person that is non-white, however that is hard to do with this technique as it is all line drawing on a white background.
If you ask me, mainly one property: most of it will be about agility. See ea.rna.nl/2020/02/11/a-tipping-point-in-the-information-revolution/ and the final link in that article's PS.
This has not been created by software but by a real human. The human was filmed while creating this (it is a real human's hand you are looking at). Some scenes were pre-drawn as a start, then making additions were being filmed. Some after-processing and editing has taken place.
WARNING! This is what happens when a self-proclaimed expert spreads false dogma and tries to convince people this is a fact. This is a horrible explanation. Another waste of time. Volumes could be written on what’s wrong with what this narrator is saying. Move on!!!
You seem to be angry about the story. Angry enough to comment publicly, anyway. So, the subject is important to you. Being content-oriented, I'm curious what you think is wrong with the story. Mail me if you don't want to comment in public and we can have a fair discussion.
I agree with Gerben. I am an early career EA and this already (only 20 minutes in) hits the mark in what I read and practice thus far in the industry. There are way too many documents and "principals" that no one revisits or follows hence the "LEAN" movement and other such concepts. When you say things like "Volumes could be written" or "spreading false dogma" and want to be taken seriously, you should bring facts and not just random opinions. Being someone that thinks everyone should just take you at your word is why EA gets a bad rap in the first place. The my way or the highway mantra gets no one anywhere.
I could help, but then I have to know who 'he' is and what the problem is. Maybe a little more than just that single sentence you wrote so that there is something to understand and/or react to?
I see it's a simple innovative entrance to EA there are many other elements like governance - reosurces - security - business processes ...etc but it still good entry to EA
Thanks. Yep, definitely too simple of course (less than 4 minutes after all). A more in depth treatise is in my other video, but then again, that takes about an hour to sit through.
This video is a good introduction BUT by the end it suggests an architect role ends when the "ideal future state" plan is created. It is misleading. Architect is involved from planning, designing, through the development process. Unless we assume waterfall approach, we know results & big risks of that approach.
It's definitely not intended that way. And I think it also doesn't say that as most of the animation is about the architecture role during a change process. It just at the end touches on future state. It is based on architecture in an up-front design world (project architecture and so forth) and not the agile methodology for changes, but that doesn't mean architecture itself cannot be agile. Have a look at the other video in this channel for more in-depth and ea.rna.nl/2015/03/08/losing-a-limpet-what-happens-when-we-dont-have-enterprise-architecture/ for a follow up on that for architecture in a setting with agile methodology.
I am thinking not about Agile particularly but what is considered Evolutionary Architecture. I like video very much. However, coming back to it a year later, I feel something important is missing, perhaps "Architecture is a snapshot of an ongoing process" rather that a future vision or design.
Yes, I know that one. Nice story. I've used the definition for architecture mentioned in there ("the important stuff") once in a while. It has also a typical software-people problem, the idea that "Software is not limited by physics". While theoretically it isn't (software is logic) the idea that physics plays no role is something that has produced many failures and problems in IT. Software engineers often ignore the physical limits of IT in their thinking (e.g. latency) because *logically* there is no price to pay. But I digress. I'm all in favour of architecture being very 'fluid' reacting to what happens, with wise things about "the important stuff". But it is about the design decisions that are made. I'm not in favour of using a 'future state' to guide you, instead I am in favour of using the here and now and look at how it can be strengthened (but all of that is in the other video). The suggestion that it is about "where we go" is in my opinion the weakest part of this video, but it was impossible to put all the subtleties in there (less than 4 minutes after all), so we cut a few corners. Though we did mention that you cannot predict the future and you have to be robust under change. But "where we're heading" is indeed problematic in the audio. You've effectively put the finger on a small sore spot in the video, for me too :-)
Hi everyone, i'm having a research about EA, there's a problem that I couldn't find any real example of EA in reality. Can you please help me out by naming some EA solutions ? and also some company/organizations using these solutions ? Thank you so much :)
It is not quite clear what you mean by 'example of EA in reality', do you mean actual EA documents from organisations? These are generally not public as they are strategic. The same holds for 'EA solutions', probably (depending on what you mean). There are several 'solutions' for setting up EA functions and processes. Even for the most popular/widespread of these there is little proof that they actually (can) work as intended. You can watch ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-tCTqNZW0wIk.html and you might look at documents by Kotusev kotusev.com/ (e.g. 'history of' and 'fad' documents).
A lot of insights on the practice of EA. Useful for those that have already experience on EA initiatives providing a lot of ideas for discussion or reflection.
