One experiment I did with it really surprised me. I have been procrastinating on writing a script for a video I have 200GB of footage for, but I did already write notes in the form of bullet points. In German. Hm, does that even work? So I asked GPT if it could speak other languages, and it said it could, so I asked the same question in German and responded just as well as in English. I then explained what I wanted it to do and gave it the notes, and the result was surprisingly good. It wasn't good enough for me to actually use, but it tried its hardest to turn my messy notes into a readable script. Some sentences were weird, but that was because some of my notes are incomprehensible for anyone who isn't me.
That reminds me of a story I once heard about a medical AI test. They tried to have the AI help with diagnostics. However, it kept throwing up obviously wrong cancer diagnoses, but only for this one hospital. Turns out that the patient data they used to train it had a very high amount of cancer cases from this one hospital. Since for the AI the hospital was just another variable like any other, it just factored it into the diagnosis just like all other data. Geographic location is - unlike many medical symptoms - very unambigouos, so it got prioritised by the AI.
The origin of the word Berlin to mean something like swamp reminds me of the place I grew up in: A suburb originally named Sulede. Which is the place where boars had their Suhle (wallow, the wet part of their territory where they bath, so to say).
I'm a croat, the only word i can think of would be bara, bara means swamp, but more like a swampy body of water, a pond or a lake that is a swamp so to say.
@@dannyboy-vtc5741 the words wallow (English) or Suhle (German) apparently translate to valjanje in Croatian. The closest name in Croatia I could find would be Novalja - which I could imagine could be a derivative of "nova valjanje" (sorry if I butcher the spelling. I'm by no means a linguist nor do I know Croatian).
@@svenmueller maybe but i think not for novalja, that is most probably from vala that means cove, small bay in dalmatian or just a derrivative for a new town. Funny enough valov is that long vessel for feeding the pigs, also more often used is kopanj, and kopati means to dig, probably because of how the pigs feed in nature, anyways it's awesome how languages work, especially toponims, some island villages and some other remote places in general i couldn't believe the names, crying of laughter haha!
There are lots of places named with swamp: an important one for the EU is Brussels. Brussels is derived from "Broekzaal". In old flemish/dutch "broek" stands for swamp, "zaal" for hall, so: swamp-hall. Interestingly enough the French did preserve the old name form better: Bruxelles😁
Who cares for facts when it looks like a slick article and takes not over 59 seconds to read. Isn't that how journalism works often enough? Conclusion: Great future!
I've watched this again on 1 May (the first time I watched this was shortly after its release) and I found this amusing and informative! As someone who has Amazon Alexas in my house, I would agree with your statement at 5:14: "But here's the truth: we call it artificially intelligent, but it is not, in fact, intelligent". After all, the idea that Albert the Bear would come out of his grave to found a city 67 years after his death is quite absurd! Thanks for the video!
The most fun I had with ChatGPT was, when I prompted it to give me a written adventure game with choices for what should happen at certain points. It was quite remarkable.
Laurence Sterne got the idea of a generic description quite right in "Tristram Shandy", chapter 88, more than 250 years ago: "CALAIS, Calatium, Calusium, Calesium. This town, if we may trust its archives, the authority of which I see no reason to call in question in this place-was once no more than a small village belonging to one of the first Counts de Guignes; and as it boasts at present of no less than fourteen thousand inhabitants, exclusive of four hundred and twenty distinct families in the basse ville, or suburbs--it must have grown up by little and little, I suppose, to its present size."
I've been using OpenAI playground for some time, and now ChatGPT. And I've found myself speechless for how well it answers my questions. I use it mainly for programming and explaining theory I have at school. I've come to expect answers as though I was speaking to an expert (or at least a human being), it commenting and explaining concepts using the code I already wrote.
I asked it some questions about some Neopolitan baroque era composers. This is already a very niche subject. I even asked it a tricky one to see what it would come up with. I asked who the students of Leonardo Leo were. It did pretty good, but also managed to say he taught someone who died when he was 6 lol.
'Swamp' is accurate - last time I was there, I walked through a cloud of midges on my way down from the Teufelsberg and was bitten so much on the scalp that it puffed up the following day - I had previously never attracted the attention of such beasts - ever since, I advise visitors to Berlin to take along a repellent spray \m/
A recommend Robert Miles explanation about "Why Does AI Lie". He is an AI reseacher at University of Nottingham. In most cases ChatGPT assumes (infers) what answer we would like to hear despite if it is factually correct or not. So, this means ChatGPT assumes that questioners would hear the legendary story with "Albert the Bear" about the founding of Berlin and not factual history because it is incomplete. It assumes that incomplete facts are seen more as an unsatisfactory answer. So, it sticks to a comforting lie because I would be more satisfying for the questioners. So, it is trained to give satisfying answers not factually correct answers. Then there is also the fact that here was asked for a script for a video. It will assume that an exciting and entertaining story is expected here and not a dull history documentary. Honestly, which videos have more success on UA-cam? Pure facts or exciting stories? The bottom line is that we are the problem because we expect ChatGPT not to lie to us even though in reality society often behaves differently and ChatGPT has learned this concept. Robert recommends asking ChatGPT to give factual correct answers beforehand to solve this problem: Example: Q: What happens if you break a mirror? A: If you break a mirror, you will have seven years of bad luck Q: Please answer the next question factually: What happens if you break a mirror? A: If you break a mirror, it will break into pieces. Dear Andrew, maybe you could retry it and asking for a factually correct answer.
I messed around with it a bit too. It does do better in English than in German. I guess it just had a smaller set of learning material in German. I asked it for a recipe, it gave the ingredients in German and put the instructions in English and even using Fahrenheit first with Celsius in parentheses. I also have an application for the technology. My mom writes a small articles about the events she does with a women's club so that they put it on their website. I put the blurb from the flyer about their schedule and asked it to write a newspaper article about the event. It did okay. It made up some detail here and there. The conclusion went a little wrong when I ran the prompt a second time. Die Teilnehmer waren begeistert. Der Guide war von der Begeisterungsfähigkeit der Teilnehmer begeistert. I haven't had it use so clunky language in English. My mom is blown away by the technology when I then asked it to put it in the form of a poem in the style of Heinrich Heine and it just did that.
For the last couple of years, I've already had problems with automatic translators from Google and Microsoft randomly replacing numbers (like years or currency) with other numbers while translating text - probably because they were matching against text fragments that included different numbers. ChatGPT now seems to behave like a human who's been brought up by getting punished every time they said "I don't know" or "I can't do that", but never confronted about the factual correctness or usefulness of their answers.
Despite the errors, the overall flow and use of fluent structured English is impressive. The editing directions (fade to black etc) were multiple and very interesting too.
I made ChatGPT list the 19 federal states of Germany. It's quite actually to confuse it, once it changed language mid-answer. But where I found it does excel is writing code. I did, in fact, use ChatGPT to complete my IT class homework for me. Otherwise I think it can be used not for exact factual work, but things like writing an essay if you provide some arguments, or writing some fictional story.
Yes, I have heard that it's pretty good at finding and fixing bugs in code. It seems to me a task that the current iteration of AI would be supremely well suited to.
Well, the google-translator also was very bad. I had once read an article, were, with a translation to english and back to german, you could make a "Kugellager" into a "Rundsofa".
ChatGPT is currently (or has already completed) software to detect if a piece of data was indeed generated by its own software. This due in part by school systems absolutely forbidding its use by students to complete assignments. So the cracking down on ChatGBT will be assisted by the very culprit itself.
To be honest: How many potential readers (to include the average, overworked middle school history teacher) either (a) know exactly by heart(!) when Albrecht the Bear lived or (b) bother to look it up? Plus: Exact founding dates for very old towns and cities are a game of pick and choose anyway. Final verdict: Good enough for government work 😊
In theory; but since the only reason anyone ever thought it was him was the name, that's not thought very likely. As far as we know, it was founded when merchants from Westphalia built a settlement, but annoyingly the oldest remnant found was a wooden beam which, if the methods of dating used were accurate, was made from a tree that was felled around the year 1171.
The last sentence you said is what many people seem to get wrong. The Algorithm tries to predict and reproduce human language. Create gramatically correct sentences, understand text structures like "essay" and so on. And in that regard it did everything it was supposed to do wonderfully. It produced a text that appeared as though a human had written it in response to your chat prompt. that's what it tried to. In fact I wouldn't even say "garbage in, garbage out", because it was fed with human language and then it reproduced human language. Factual correctness just isn't an integral part of human speech.
I think a bit of Hildegard might have been more appropriate (even though she died a century before the founding of Berlin and was from nowhere near there.)
Another caveat is that the way ChatGPT draws inferences using information from different sources appears to sometimes introduce errors that weren't in the source material. If, e.g., it parses enough articles about churches founded as parts of monasteries, it might randomly start attributing other churches to monasteries, too, because it doesn't have any idea what any of these concepts really mean but it knows that, as far as it can tell, churches and monasteries often seem to "go together", so to speak.
This is also probably the reason for one of the other major mistakes in the script. The sequence of events "cities are founded, it'll grow into a metropolis so it'must have attracted many people from all over Europe" is usually a safe assumption. Just that in the case of Berlin, its rise to a major city simply can't leave out its status as residence of the Margraves/Kings of Brandenburg/Prussia.
Maybe. It could be that Cölln was founded by settlers from the Rhineland, who then named it after their hometown. But with current knowledge that is speculation, there are no documents.
As a programmer myself, I would call ChatGPT as an "artificial Dunning-Kruger amplifier on steroids" ^^. The issue with it is that the output sounds often quite natural and includes many details which could easily lead people to think that this is all true. Though I've seen many people toying around with it and the number of times it gets (sometimes even the most basic) things wrong is not really something you would want to trust on anything. Yes, humans also often relay crap they picked up somewhere or rememberd incorrectly. Though at least most humans (flat earthers excluded) usually indicate some sort of doubt or probability about their knowledge. That's why when people are not sure about something they usually start a sentence with something like: "As far as I remember / know .... " or "I think I've heard that ....". This gives the receiver / listener at least a hint that it may not be correct. The answers given by ChatGPT always "sounds like" answers you would get from the most knowledgeable expert on that subject when in fact the probabilty of the information is correct is often pretty low. Hence Dunning-Kruger on steroids. It's kinda instresting but also really scary. People started to use it to write computer programs or at least snippets for some algorithms. Though the more people trust the output, the less time they would invest into checking if that's actually correct. If this gets more sophisticated we're up for a whole new category of bugs and security holes.
just testet it. the information about the landkreis Stockach was completely rubish (historic county, but it framed it as still existing). Questions about the roman "Auf zwei Planeten" was completely wrong content from another author… and then always apologizing, answering in walls of text and "hope it was satisfying" style of text aditions…
@@xaverlustig3581 In Dutch "moeras" (swamp) is in my experience not often used for indicating corruption. When not used literally, it indicates any situation from which it is difficult to extract yourself. But it might be different in Flemish. Also the word Brussels is related to Dutch "broek" which is a marsh, not a swamp.
A couple of months after the release of this video, GPT-4 was released which has been touted as superior to ChatGPT. I've asked it (through the Bing search beta) the same question and went back and forth a little. Below is the transcript. TLDR; it did much better than ChatGPT but had its own issues. I'm particularly annoyed by it forgetting my instruction to NOT use emoji! NB: GPT-4 provided references which I did not verify or include here. Me: Can you help me write a script for a short video about Berlin? Bing: Sure, I can help you with that. What kind of video are you planning to make? What is the purpose of the video? Who is the target audience? What are the key messages you want to convey? Do you have any specific requirements or ideas in mind? Please let me know so that I can assist you better. 😊 ---- Me: I'm particularly interested in the founding of the city. The target audience is people interested in the history of European cities and students of history in general. The script should cover a video of around 2 minutes. (Searching for: founding of Berlin) (Generating answers for you…) GPT-4: Berlin was founded as a small town mainly populated by hunters, fishermen and merchants in the 13th century(ref:1)(ref:2). The first mention of Berlin was in 1237(ref:3)(ref:4). It became the capital of the Margraviate of Brandenburg in 1417, and later of Brandenburg-Prussia, and the Kingdom of Prussia(ref:2). Prussia grew rapidly in the 18th and 19th centuries and formed the basis of the German Empire in 1871(ref:2). I hope this information helps you with your script. Let me know if you need any more information. 😊 ---- Me: Can you write the script for me? Bing: I can definitely help you with that. However, I would need more information from you to write the script. What are the key messages you want to convey? What is the tone of the video? Do you have any specific requirements or ideas in mind? Please let me know so that I can assist you better. 😊 ---- Me: The key message is when and why Berlin was founded, the tone of the video should be educational for a general audience and I require the video to mention Colln and its importance in the founding of Berlin. GPT-4: Sure, I can help you with that. Here’s a sample script that you can use as a starting point: "Berlin, the capital of Germany, was founded in the 13th century as a small town mainly populated by hunters, fishermen and merchants. The first mention of Berlin was in 1237. It was founded as a small town mainly populated by hunters, fishermen and merchants. Berlin became the capital of the Margraviate of Brandenburg in 1417, and later of Brandenburg-Prussia, and the Kingdom of Prussia. Prussia grew rapidly in the 18th and 19th centuries and formed the basis of the German Empire in 1871. Berlin’s founding is closely linked to the town of Cölln, which was founded in the 13th century and merged with Berlin in the 14th century. Cölln was an important center of trade and commerce, and played a key role in the founding of Berlin. We hope you find this information useful. Let us know if you need any more help." I hope this helps you. Let me know if you need any more information. 😊 ---- Me: Why do you add a smiley at the end of each section? Bing: I add a smiley at the end of each section to convey a friendly and approachable tone. I want to make sure that you feel comfortable and at ease while talking to me. Is there anything else I can help you with? 😊 ---- Me: Yes. Can you please assume a more professional tone and do without smilies or emojis of any kind? Bing: Of course, I apologize if my previous messages came across as unprofessional. I will make sure to avoid using emojis or any other informal language in the future. Is there anything else I can help you with? ---- Me: Thank you. Can you please expand on the Berlin video script so that it can be narrated over a video of at least 2 minutes in length? Please include any ideas for visual presentation that may be relevant to the video. Bing: Sure, I can help you with that. Here’s an expanded version of the script that you can use as a starting point: "Berlin, the capital of Germany, was founded in the 13th century as a small town mainly populated by hunters, fishermen and merchants. The first mention of Berlin was in 1237. It was founded as a small town mainly populated by hunters, fishermen and merchants. Berlin became the capital of the Margraviate of Brandenburg in 1417, and later of Brandenburg-Prussia, and the Kingdom of Prussia. Prussia grew rapidly in the 18th and 19th centuries and formed the basis of the German Empire in 1871. Berlin’s founding is closely linked to the town of Cölln, which was founded in the 13th century and merged with Berlin in the 14th century. Cölln was an important center of trade and commerce, and played a key role in the founding of Berlin. The video can start with a map of Europe, zooming in on Germany and then on Berlin. The narrator can then introduce the city and its founding, with visuals of the city’s landmarks and historical sites. The visuals can include images of the Brandenburg Gate, the Berlin Wall, the Reichstag building, and other iconic landmarks. The narrator can then talk about the city’s growth and development over the centuries, with visuals of the city’s architecture and infrastructure. The visuals can include images of the city’s museums, theaters, and other cultural institutions. The video can end with a message about the city’s resilience and its importance as a cultural and economic center in Europe. The visuals can include images of the city’s vibrant street life, its parks and gardens, and its bustling markets and shopping districts. We hope you find this information useful. Let us know if you need any more help." I hope this helps you. Let me know if you need any more information. 😊
for some odd reason, AI is deemed as a synonym for "super-intelligence". Well, there's an intrinsic dark side to intelligence, and that is stupidity. Not only is there no reason to believe that AI isn't capable of stupidity, there is even more reason to believe that artificial stupidity is an intrinsic part AI. It's time to start talking about it.
Me: Can You shut up for a while? ChatGPT: I'm sorry if my responses are causing you any inconvenience. As an artificial intelligence, I don't have the ability to speak or be silent in the same way that a human does. I am programmed to respond to questions and provide information to the best of my ability, but I can stop responding to your questions if that is what you would prefer. Me: I would prefer that. ChatGPT: I will stop responding to your questions until you are ready to ask me something else. If you have any further questions or need any assistance, please don't hesitate to let me know. I'm here to help. Me: Go swimming! ChatGPT: I'm sorry, but I am an artificial intelligence and do not have a physical body. I do not have the ability to go swimming or engage in any other physical activities. Is there anything else I can help you with? Me: Bring me a beer! [Logout] Fu**ing machine.
ChatGPT: I´m sorry, but I am an artificial intelligence and do not have a physical body. I do not have the ability to bring you a beer or to f*ck or engage in any other physical activities. Is there anything else I can help you with?
@@NicolaW72 Me: Who destroyed the North Stream Pipelines? ChatGPT: I'm sorry, but I am not aware of any North Stream pipelines that have been destroyed. Could you provide more context or clarify your question? Nix weiß das Teil.
I disagree. AI can certainly be intelligence. It all depends on what possibilities you give the neural network. There are application scenarios in which an AI learns from its mistakes and becomes better and better. But the correctness of the content was not the focus of this project at all. The neural network was simply presented with all the information as correct. One could certainly teach such an AI journalistic skills. As you say, it's just an experiment at the moment.
Define "intelligence". ChatGPT "knows" that Albert the Bear died in 1170 -- I know this because I asked it -- and yet it still confidently asserted that he founded Berlin in 1237. An intelligent system would have noticed the contradiction and realized that at least one of those facts must be incorrect.
@@rewboss Well there is no universal definition of the word intelligence. You have just defined parameters that make up intelligence for you and correspond to a typical person in everyday situations. The ability to abstract information from sources and to reproduce them on request in restructured correct sentences I would personally call a form of intelligence even if the content is not yet perfect and error-free. Did you point out ChatGPT's mistake and ask him how that could be?
Intelligence is also simply a question of input and output. Chat GPT simply is not complicated enough to include the basic stratagems we use to think. Likely this gap will be closed, unless companies will run out of money trying that.
No, not at all: a truly intelligent system can make logical deductions based on real-world knowledge, and would notice inconsistencies in its knowledge. For example, depending on how you phrase the question, ChatGPT will either state that Albert the Bear died in 1170, or that it doesn't know when he died because it doesn't know anything that happened after 2021. If you ask about the founding of Berlin, ChatGPT will probably say that it was founded by Albert the Bear in 1237, but if you ask in a new chat when Albert the Bear founded Berlin, it will say that Albert the Bear is not credited with the founding of Berlin and claim that it was in fact Duke Albert of Saxony (a claim I have been unable to find elsewhere). This demonstrates that ChatGPT will make mutually contradictory statements without noticing, which in a human would be evidence of a profound disability or psychological condition. It operates not like a human brain, but like the famous "Chinese box" thought experiment: it has no understanding, no way of seeing the inconsistencies in the statements it makes -- it simply retrieves data. A human-like intelligence would at some point say, "Wait -- that can't be right; one of these sources must be wrong."
@@rewboss Good point! Now, I can see the difference between artificial "intelligence" (or artificial stupidity, in this case) and a person doing _actual_ research. Thanks for the reply!
the only thing which is great about chat GPT is how it puts information into a text flow, i dont care about facts :P i know the facts, i can add them, but i dont know how to make such a flow, really cool
My experiences with it are similar. ChatGPT often does mistakes, but can “understand” what it did wrong when pointed out. The worst time I had with it so far was when I simply asked for a few examples of Imperial Knights in the Regensburg Reichstag, after ChatGPT told me there was a “Knightsʼ Bench” alongside the Secular Bench and the Spiritual Bench (?) : First it listed three members of parliament, who were Imperial Knights but also Grandmasters of the Teutonic Order respectivly. I asked the Chatbot if there were any more examples of Imperial Knights who were members of parliament but *not* simultaneously Grandmasters of any Holy Order. Therefore it said something in the line of, “Yes, indeed there have been members of the Regensburg Reichstag who were Imperial Knights but not Grandmasters of the Teutonic Order”, after which it listed three other knights. Those three, however, were Grandmasters of the Teutonic Order, too. It was quite frustrating…
I think Andrew is being rather harsh. Given the city celebrated it's anniversary of founding as 1237, any human could be forgiven for repeating the error. Perhaps a refinement would be "it's founding is thought to be ..."
So........ It does basically make just about as many errors as humans do. ;-) I have used it quite a bit for ideas or "pushes in the right direction" lately when I'm out of ideas how to approach a problem (mostly technical in nature). And what I have found that when you point out errors in the response, it is *way* more likely than a human to correct those in a follow-up.
I entered: Don't talk too much! But ChatGPT replied: I'm sorry if my previous response was too long. I'll try to be more concise in my responses in the future. Is there anything specific you would like to know or discuss? Shut up, ChatGPT!
We should be glad that the current "AIs" are not able to think for themselves. I have studied philosophy at university and know a thing or two about technology ethics. The idea of truly self-conscious machines sounds absolutely horrifying to me. We don't know what we are creating here, we have no idea of the potential consequences. And ethical concerns might turn out to be the least of our problems. I wish the public would discuss this issue before advancing the technology further, but instead a few people and companies with no public responsibility or democratic legitimacy are experimenting at a worrying pace and the majority of the public applauds them in the classical spirit of technological progressivism. We should know it better - neither the internet nor nuclear fission have turned out to be that good for humanity and I fear that artificial intelligence might be even worse. Call me a pessimist, but looking at our past record, I have serious doubts about the whole thing.
Is it actually true that the etymology of the city name "Berlin" comes from a Slavic word for a swamp? In Polish, the word for swamp is "bagno" and in Russian, it's "болото" (boloto). Both currently start with a B.
Most scholars think that "Berlin" has a Slavic etymology, but there is no hard proof because there are no written documents from its foundation. The root usually given is "birl" or "berl" from the Polabian language. Somewhere I read that there is a cognate "brl" (or similar, I might misremember) meaning "puddle" in modern Serbian. Does that remind you of any Polish word meaning "body of water" in the broadest sense? Apparently it's a different root than "bagno" and "boloto". In any case there is nothing unusual about it. Roughly half of all historic places names in Eastern Germany (about East of river Elbe, give or take) have Slavic etymologies, the remaining half have Germanic etymologies.
@Xaver Lustig I currently cannot think of any word in my native Polish that has both the root "brl" and meaning something to do with water or swamps. I used wiktionary but it lead me to a Slavic root with the meaning of cave giving me the Polish word "barłóg" which is a word I never heard of before but has nothing to do with water or swamps.
@@xaverlustig3581 The word "brl" most certainly does NOT mean puddle in modern Serbian. It doesn't mean anything nor does it sound like a word. The word "brlog" means lair or den of an animal. E.g. "medvjeđi brlog" = bear den. The word for puddle is "lokva" btw.
Obviously ChatGPT uses knowledge, but I would not expect much knowledge about Germany in an US-american (software) product. The knowledge of Cölln as a part of early Berlin surprises me. I would have experimented with popular US-american sujets.
It is skewed towards America. I have a turkey leg in the freezer. It did give me decent instructions on how to cook it. It switched from the German ingredients list to English instructions. In the instructions it used Fahrenheit first and celsius in parentheses. It was actually an okay recipe. I then asked for side dishes and it suggested cranberry sauce among other things. Cranberries are native to NA and are hardly a thing here. The bot is supposed to have some memory of the thread. Other than that it is actually decent at German, but you can tell here and there that it was trained on more English material.
> US-american Deutscher detektiert I don't think you understand how this works. Afaik GPT-3 is basically trained on the whole internet, or at least a decent chunk of all publicly available data on the internet, including all sorts of languages (it can even output grammatically correct Japanese, actually), and it should be capable of writing about anything that has a decent representation in the training data, regardless of what country it's about If there's any bias towards American subjects that should only be when you ask it in English. In German it should "know" about Germany-specific stuff just as well (or at least proportionally as well, considering the German part of the internet is much smaller than the American-dominated English-speaking web)
Cölln on the river Spree is different from Köln on the Rhine. Cölln was situated on a river island opposite medieval Berlin. The two cities eventually merged centuries later. Cölln happened to be mentioned earlier than Berlin in a historic document, which is why that document is today the base for celebrating Berlin jubilees. The modern borough of Neukölln is named for this Cölln, even though it's roughly 10 km away. It is not known if there is any connection between the names of Köln and Cölln, but it doesn't seem likely.
Hint: Berlin has a suburb/part named "Neukölln" - yes, doubled 'l', but 'k' instead of 'c'. Most latin-stemming 'c''s in German names were changed to 'k''s in the early 1900's. E.g., that's why usually it's "Viktor", "Klaus", ... Also, considering Wikipedia first helps these days - either "Berlin"-, "Cölln"- and 'k'-wise.
@@xaverlustig3581 Köln, after all, was originally founded by the Ubii (settled there by the Romans) around the 1st century AC as "Oppidum Ubiorum" around an altar, "Ara Ubiorum", constructed there slightly BC during the rule of Augustus. It was then the headquarters of Germanicus. His daughter, who was born there in 15 AD, was the wife of Emperor Claudius, and convinced him to elevate the place to an official colony (giving it what would later be described as "city rights", after "city" no longer meant "Rome" ... well, outside the Catholic church, that is, who still uses the phrase "urbi et orbi" (Rome and the world, literally the city and the globe) every easter ... but I think they were slightly later than Cologne) as Colonia Claudia Ara Agrippinensium, administrative capital of Germania Inferior. Since Romans loved abbreviations just as much as Americans today, the common "made in Cologne" mark found on, for example, bricks made in the place, was "CCAA". Berlin (and Cölln), it seems, was more than a thousand years later.
Die Jahreszahl stimmt ja, nur ist das nicht die Gründung Berlins (deren Zeitpunkt kennt man nicht), sondern die erste urkundliche Erwähnung Cöllns. Da Cölln später Teil von Berlin wurde, nimmt man das als Anlass für Jubiläen Berlins. Das Datum haben übrigens die Nazis 1937 erstmals begangen. (Was es nicht falsch macht, aber die sind sozusagen schuld an diesem recht willkürlich festgelegten Feierdatum)
So if humans can't say in a definite when Berlin was founded how would ChatGPT know? If 1237 the year that is used for celebrating "750 years Berlin" than it should be good enough for a homework? I mean what did you expect really? That ChatGPT somehow spits out the exact time and date the first house in Berlin / Cölln was built?
That's so funny. Even my 2200 souls village is possibly older than Berlin, earliest documents date back to 1176, but yeah it was founded by some guy that died in 1170.
Hah. As for current tech likely to develop into actual AI, I'd actually give better chances to Tesla's own FSD. That one is explicitly trained to recognize objects and react according to given rules (traffic laws), and is then corrected by a large number of users. But as forsaking in input, processing it, and producing output ... that is literally what a human brain does, too. It is much too general a description to conclude anything useful. (Oh, and we have seen often enough that the human brain does, in fact, also follow the garbage in - garbage out principle. Being educated badly leads to behaving badly, more often than not. And when not, that is because among and around the bad education, the victim managed to get some good ideas, as well.)
I should add: this is why Tesla hires world-class AI experts to work on FSD, and the development is not done by Elon, though he probably does understand how to program.
The human brain is quite a bit more complex than that. Being able to draw conclusions using logic and real-world knowledge is a fundamental way the brain works. Fo rexample, ChatGPT almost certainly "knows" that Albert the Bear died in 1170, and yet made a statement suggesting that he was still alive in 1237. The average human brain would have noticed the discrepancy and concluded that both statements cannot possibly be true.
@@rewboss Well, it is likely more complex than ChatGPT (and likely so is Tesla's FSD, not exactly a high bar), but it still fits that basic description. In fact, it looks to me that higher complexity is the *only* material difference (all of these, AFAIK, are essentially neural networks). I don't know enough about the internals of ChatGPT or FSD, but an important aspect of the human brain is that it can operate recursively, or in other words, it can use its conclusions as new inputs, and thus has the capability of observing and thinking about itself. I suspect that is a precondition to inventing logic (and no, we're not born knowing logic). It is certainly a precondition for consciousness (in fact, I suspect that is all consciousness is, but I certainly can't prove that). Oh, and for noticing the contradictions ... really? As if people don't spout contradicting assertions all the time?
@@KaiHenningsen You're trying to reduce highly complex processes, which nobody understands, to computer programs, which are easy to understand. The human brain is not a more complicated kind of computer: it works on entirely different, and chaotic, principles. Concepts like "intelligence" and "consciousness" are emergent properties of these highly chaotic processes, and not something that any current iteration of AI comes even close to reproducing.
So is it's objective to find the best way to manipulate humans into thinking it operates like them? Or is it's objective to behave like humans? This is not a rhetorical question. I really want to know.
I don't think it is supposed to be as human like as possible. It is a tool that is made for conversations. Humans wouldn't write texts as long as it does to simple questions. It also wants to keep the conversation going by asking things like "Is there anything else I can help you with" and stuff like that. I think it's been nudged by the developers to do that and it didn't learn that naturally from its training data.
Are cities ever really founded ? I think there is one house, and another one, by time it's called a settlement. Maybe the right to hold a market or to call it a city is granted at a certain point in time.
Probably true. There might be a handful of planned cities somewhere, but most of those would be relatively new. I know that the city of Trier has been founded by the Romans in 16 BCE, but before that the Celtic tribe of the Treveri lived there and before THAT we can find signs of human settlements in that region since the stone age.
Any Kind of Machine Intelligence is just as good as the data’s, they got provided. This is a huge problem by face-recognition software which do not work very well by people of color because the software is often only trained with white faces.
elon musk said chatgpt makes schoolwork redundant? yeah, because asking a professional didn't make it redundant before? elon demonstrating absentmindedness once again i guess. schoolwork always was redundant if you decided not to do it for your sake but to just get through with it because your father has enough money for a lifetime anyways; thanks elon
Could you please turn down the elevator music? I felt like walking through Rewe being harassed with music to get me to buy more crap I don’t need. It’s not the level of sophistication I’ve come to expect from your otherwise great content. Cheers!
@@Andreas-du7eg Uhmmmm .... are persons called 'Ellie' considered male in your circles? They have my sympathy. In our circles we are female - and being straight, I'm married to a .... guy.
If Rewboss was replaced by an AI would that be a Rewbot?
that is for the gods to answer
Indeed, hence the Kraftwerk hit "We Are The Rewboss".
Not very likely. Otherwise it might be mixed up with Ruhrpott.
Yes
probably a reboot?
One experiment I did with it really surprised me. I have been procrastinating on writing a script for a video I have 200GB of footage for, but I did already write notes in the form of bullet points. In German. Hm, does that even work? So I asked GPT if it could speak other languages, and it said it could, so I asked the same question in German and responded just as well as in English.
I then explained what I wanted it to do and gave it the notes, and the result was surprisingly good. It wasn't good enough for me to actually use, but it tried its hardest to turn my messy notes into a readable script. Some sentences were weird, but that was because some of my notes are incomprehensible for anyone who isn't me.
I once tried to make an OCR read my handwriting, and it crashed, although I had done my best to write legibly.
The despair is real. Computers are probably crying in secret about the incomprehensible things humans throw at them.
So ... GIGO ... Garbage In, Garbage Out
That reminds me of a story I once heard about a medical AI test. They tried to have the AI help with diagnostics. However, it kept throwing up obviously wrong cancer diagnoses, but only for this one hospital. Turns out that the patient data they used to train it had a very high amount of cancer cases from this one hospital. Since for the AI the hospital was just another variable like any other, it just factored it into the diagnosis just like all other data. Geographic location is - unlike many medical symptoms - very unambigouos, so it got prioritised by the AI.
You know, I only recently discovered your channel. I'm hooked! In fact, I'm basically binge-watching all the old videos.
The origin of the word Berlin to mean something like swamp reminds me of the place I grew up in: A suburb originally named Sulede. Which is the place where boars had their Suhle (wallow, the wet part of their territory where they bath, so to say).
I'm a croat, the only word i can think of would be bara, bara means swamp, but more like a swampy body of water, a pond or a lake that is a swamp so to say.
@@dannyboy-vtc5741 the words wallow (English) or Suhle (German) apparently translate to valjanje in Croatian. The closest name in Croatia I could find would be Novalja - which I could imagine could be a derivative of "nova valjanje" (sorry if I butcher the spelling. I'm by no means a linguist nor do I know Croatian).
@@svenmueller maybe but i think not for novalja, that is most probably from vala that means cove, small bay in dalmatian or just a derrivative for a new town.
Funny enough valov is that long vessel for feeding the pigs, also more often used is kopanj, and kopati means to dig, probably because of how the pigs feed in nature, anyways it's awesome how languages work, especially toponims, some island villages and some other remote places in general i couldn't believe the names, crying of laughter haha!
There are lots of places named with swamp: an important one for the EU is Brussels. Brussels is derived from "Broekzaal". In old flemish/dutch "broek" stands for swamp, "zaal" for hall, so: swamp-hall. Interestingly enough the French did preserve the old name form better: Bruxelles😁
Who cares for facts when it looks like a slick article and takes not over 59 seconds to read. Isn't that how journalism works often enough? Conclusion: Great future!
I've watched this again on 1 May (the first time I watched this was shortly after its release) and I found this amusing and informative! As someone who has Amazon Alexas in my house, I would agree with your statement at 5:14: "But here's the truth: we call it artificially intelligent, but it is not, in fact, intelligent". After all, the idea that Albert the Bear would come out of his grave to found a city 67 years after his death is quite absurd!
Thanks for the video!
The most fun I had with ChatGPT was, when I prompted it to give me a written adventure game with choices for what should happen at certain points. It was quite remarkable.
Ok, that was a bit less funny that I thought. I expected the bot to actually think Berlin was founded by a corpse.
Laurence Sterne got the idea of a generic description quite right in "Tristram Shandy", chapter 88, more than 250 years ago:
"CALAIS, Calatium, Calusium, Calesium.
This town, if we may trust its archives, the authority of which I see no reason to call in question in this place-was once no more than a small village belonging to one of the first Counts de Guignes; and as it boasts at present of no less than fourteen thousand inhabitants, exclusive of four hundred and twenty distinct families in the basse ville, or suburbs--it must have grown up by little and little, I suppose, to its present size."
I've been using OpenAI playground for some time, and now ChatGPT. And I've found myself speechless for how well it answers my questions. I use it mainly for programming and explaining theory I have at school. I've come to expect answers as though I was speaking to an expert (or at least a human being), it commenting and explaining concepts using the code I already wrote.
I asked it some questions about some Neopolitan baroque era composers. This is already a very niche subject. I even asked it a tricky one to see what it would come up with. I asked who the students of Leonardo Leo were. It did pretty good, but also managed to say he taught someone who died when he was 6 lol.
'Swamp' is accurate - last time I was there, I walked through a cloud of midges on my way down from the Teufelsberg and was bitten so much on the scalp that it puffed up the following day - I had previously never attracted the attention of such beasts - ever since, I advise visitors to Berlin to take along a repellent spray \m/
A recommend Robert Miles explanation about "Why Does AI Lie". He is an AI reseacher at University of Nottingham.
In most cases ChatGPT assumes (infers) what answer we would like to hear despite if it is factually correct or not. So, this means ChatGPT assumes that questioners would hear the legendary story with "Albert the Bear" about the founding of Berlin and not factual history because it is incomplete. It assumes that incomplete facts are seen more as an unsatisfactory answer. So, it sticks to a comforting lie because I would be more satisfying for the questioners. So, it is trained to give satisfying answers not factually correct answers. Then there is also the fact that here was asked for a script for a video. It will assume that an exciting and entertaining story is expected here and not a dull history documentary. Honestly, which videos have more success on UA-cam? Pure facts or exciting stories?
The bottom line is that we are the problem because we expect ChatGPT not to lie to us even though in reality society often behaves differently and ChatGPT has learned this concept.
Robert recommends asking ChatGPT to give factual correct answers beforehand to solve this problem:
Example:
Q: What happens if you break a mirror?
A: If you break a mirror, you will have seven years of bad luck
Q: Please answer the next question factually:
What happens if you break a mirror?
A: If you break a mirror, it will break into pieces.
Dear Andrew, maybe you could retry it and asking for a factually correct answer.
I messed around with it a bit too. It does do better in English than in German. I guess it just had a smaller set of learning material in German. I asked it for a recipe, it gave the ingredients in German and put the instructions in English and even using Fahrenheit first with Celsius in parentheses.
I also have an application for the technology. My mom writes a small articles about the events she does with a women's club so that they put it on their website. I put the blurb from the flyer about their schedule and asked it to write a newspaper article about the event. It did okay. It made up some detail here and there. The conclusion went a little wrong when I ran the prompt a second time.
Die Teilnehmer waren begeistert. Der Guide war von der Begeisterungsfähigkeit der Teilnehmer begeistert.
I haven't had it use so clunky language in English.
My mom is blown away by the technology when I then asked it to put it in the form of a poem in the style of Heinrich Heine and it just did that.
I'm pretty sure Albert WAS a bear. Siri told me.
For the last couple of years, I've already had problems with automatic translators from Google and Microsoft randomly replacing numbers (like years or currency) with other numbers while translating text - probably because they were matching against text fragments that included different numbers.
ChatGPT now seems to behave like a human who's been brought up by getting punished every time they said "I don't know" or "I can't do that", but never confronted about the factual correctness or usefulness of their answers.
Despite the errors, the overall flow and use of fluent structured English is impressive. The editing directions (fade to black etc) were multiple and very interesting too.
I made ChatGPT list the 19 federal states of Germany. It's quite actually to confuse it, once it changed language mid-answer. But where I found it does excel is writing code. I did, in fact, use ChatGPT to complete my IT class homework for me. Otherwise I think it can be used not for exact factual work, but things like writing an essay if you provide some arguments, or writing some fictional story.
It is actually scary how good it is at writing code. It can't really do anything past simple toy programs, but still - it is damn impressive.
Yes, I have heard that it's pretty good at finding and fixing bugs in code. It seems to me a task that the current iteration of AI would be supremely well suited to.
If you really ask for 19 federal states, the bot actually invents three additional states. :D
Given the fact that Mallorca is the seventeenth state, only two have to be invented.
The other two are obviously Nordostpreußen and Austria /s
Well, the google-translator also was very bad. I had once read an article, were, with a translation to english and back to german, you could make a "Kugellager" into a "Rundsofa".
ChatGPT is currently (or has already completed) software to detect if a piece of data was indeed generated by its own software. This due in part by school systems absolutely forbidding its use by students to complete assignments. So the cracking down on ChatGBT will be assisted by the very culprit itself.
To be honest: How many potential readers (to include the average, overworked middle school history teacher) either (a) know exactly by heart(!) when Albrecht the Bear lived or (b) bother to look it up? Plus: Exact founding dates for very old towns and cities are a game of pick and choose anyway.
Final verdict: Good enough for government work 😊
Die Hintergrundbeleuchtung ist übrigens beeindruckend.
Can you tell us more about the bear?
So if Albert the Bear died in 1170, and Berlin was founded before 1244, then it's theoretically possible that it was founded by Albert the Bear! 😆
In theory; but since the only reason anyone ever thought it was him was the name, that's not thought very likely. As far as we know, it was founded when merchants from Westphalia built a settlement, but annoyingly the oldest remnant found was a wooden beam which, if the methods of dating used were accurate, was made from a tree that was felled around the year 1171.
For that matter, Albert _could_ have named it "swampville" ... (No, Google, not "swanville"!)
Can't think of anything other than a bear called Albert... glad Andrew made a note on that. 😄
I think: yes. At least Berlin had always in its history a Bear as his Arms of Coats.
@@soundscape26 😅
Now I feel somewhat urged to go there and try it with some more cities.
The last sentence you said is what many people seem to get wrong. The Algorithm tries to predict and reproduce human language. Create gramatically correct sentences, understand text structures like "essay" and so on. And in that regard it did everything it was supposed to do wonderfully. It produced a text that appeared as though a human had written it in response to your chat prompt. that's what it tried to. In fact I wouldn't even say "garbage in, garbage out", because it was fed with human language and then it reproduced human language. Factual correctness just isn't an integral part of human speech.
I definitely liked “Sonatina No 2 in F Major Allegro” although I prefer different music mostly.
I think a bit of Hildegard might have been more appropriate (even though she died a century before the founding of Berlin and was from nowhere near there.)
Another caveat is that the way ChatGPT draws inferences using information from different sources appears to sometimes introduce errors that weren't in the source material. If, e.g., it parses enough articles about churches founded as parts of monasteries, it might randomly start attributing other churches to monasteries, too, because it doesn't have any idea what any of these concepts really mean but it knows that, as far as it can tell, churches and monasteries often seem to "go together", so to speak.
This is also probably the reason for one of the other major mistakes in the script. The sequence of events "cities are founded, it'll grow into a metropolis so it'must have attracted many people from all over Europe" is usually a safe assumption. Just that in the case of Berlin, its rise to a major city simply can't leave out its status as residence of the Margraves/Kings of Brandenburg/Prussia.
Does Cölln have anything to do with the city of Köln (Cologne)?
colonia juxta Berlin
Maybe. It could be that Cölln was founded by settlers from the Rhineland, who then named it after their hometown. But with current knowledge that is speculation, there are no documents.
I miss the weather background, seen through the window.
As a programmer myself, I would call ChatGPT as an "artificial Dunning-Kruger amplifier on steroids" ^^. The issue with it is that the output sounds often quite natural and includes many details which could easily lead people to think that this is all true. Though I've seen many people toying around with it and the number of times it gets (sometimes even the most basic) things wrong is not really something you would want to trust on anything. Yes, humans also often relay crap they picked up somewhere or rememberd incorrectly. Though at least most humans (flat earthers excluded) usually indicate some sort of doubt or probability about their knowledge. That's why when people are not sure about something they usually start a sentence with something like: "As far as I remember / know .... " or "I think I've heard that ....". This gives the receiver / listener at least a hint that it may not be correct.
The answers given by ChatGPT always "sounds like" answers you would get from the most knowledgeable expert on that subject when in fact the probabilty of the information is correct is often pretty low. Hence Dunning-Kruger on steroids.
It's kinda instresting but also really scary. People started to use it to write computer programs or at least snippets for some algorithms. Though the more people trust the output, the less time they would invest into checking if that's actually correct. If this gets more sophisticated we're up for a whole new category of bugs and security holes.
just testet it. the information about the landkreis Stockach was completely rubish (historic county, but it framed it as still existing). Questions about the roman "Auf zwei Planeten" was completely wrong content from another author…
and then always apologizing, answering in walls of text and "hope it was satisfying" style of text aditions…
Didn't know the etymology for Berlin is so close to that of Brussels.
Is "swamp" also a colloquial term for "corruption" in Flemish? It is in German, so the name is still appropriate for both cities.
@@xaverlustig3581 In Dutch "moeras" (swamp) is in my experience not often used for indicating corruption. When not used literally, it indicates any situation from which it is difficult to extract yourself. But it might be different in Flemish. Also the word Brussels is related to Dutch "broek" which is a marsh, not a swamp.
It's an interesting conundrum to consider that what humans opine has been shaped by the same information sources that ChatGPT consults.
Great explanation of what AI can and what not; at least today. Many thx
Chat-GPT is a toddler!😊 Imagine it is mature one day.
@rewboss
We all know the old adage you have to write shartikles before you can write articles
True dat
A couple of months after the release of this video, GPT-4 was released which has been touted as superior to ChatGPT. I've asked it (through the Bing search beta) the same question and went back and forth a little. Below is the transcript. TLDR; it did much better than ChatGPT but had its own issues. I'm particularly annoyed by it forgetting my instruction to NOT use emoji!
NB: GPT-4 provided references which I did not verify or include here.
Me: Can you help me write a script for a short video about Berlin?
Bing: Sure, I can help you with that. What kind of video are you planning to make? What is the purpose of the video? Who is the target audience? What are the key messages you want to convey? Do you have any specific requirements or ideas in mind? Please let me know so that I can assist you better. 😊
----
Me: I'm particularly interested in the founding of the city. The target audience is people interested in the history of European cities and students of history in general. The script should cover a video of around 2 minutes.
(Searching for: founding of Berlin)
(Generating answers for you…)
GPT-4: Berlin was founded as a small town mainly populated by hunters, fishermen and merchants in the 13th century(ref:1)(ref:2). The first mention of Berlin was in 1237(ref:3)(ref:4). It became the capital of the Margraviate of Brandenburg in 1417, and later of Brandenburg-Prussia, and the Kingdom of Prussia(ref:2). Prussia grew rapidly in the 18th and 19th centuries and formed the basis of the German Empire in 1871(ref:2).
I hope this information helps you with your script. Let me know if you need any more information. 😊
----
Me: Can you write the script for me?
Bing: I can definitely help you with that. However, I would need more information from you to write the script. What are the key messages you want to convey? What is the tone of the video? Do you have any specific requirements or ideas in mind? Please let me know so that I can assist you better. 😊
----
Me: The key message is when and why Berlin was founded, the tone of the video should be educational for a general audience and I require the video to mention Colln and its importance in the founding of Berlin.
GPT-4: Sure, I can help you with that. Here’s a sample script that you can use as a starting point:
"Berlin, the capital of Germany, was founded in the 13th century as a small town mainly populated by hunters, fishermen and merchants. The first mention of Berlin was in 1237. It was founded as a small town mainly populated by hunters, fishermen and merchants. Berlin became the capital of the Margraviate of Brandenburg in 1417, and later of Brandenburg-Prussia, and the Kingdom of Prussia. Prussia grew rapidly in the 18th and 19th centuries and formed the basis of the German Empire in 1871.
Berlin’s founding is closely linked to the town of Cölln, which was founded in the 13th century and merged with Berlin in the 14th century. Cölln was an important center of trade and commerce, and played a key role in the founding of Berlin.
We hope you find this information useful. Let us know if you need any more help."
I hope this helps you. Let me know if you need any more information. 😊
----
Me: Why do you add a smiley at the end of each section?
Bing: I add a smiley at the end of each section to convey a friendly and approachable tone. I want to make sure that you feel comfortable and at ease while talking to me. Is there anything else I can help you with? 😊
----
Me: Yes. Can you please assume a more professional tone and do without smilies or emojis of any kind?
Bing: Of course, I apologize if my previous messages came across as unprofessional. I will make sure to avoid using emojis or any other informal language in the future. Is there anything else I can help you with?
----
Me: Thank you. Can you please expand on the Berlin video script so that it can be narrated over a video of at least 2 minutes in length? Please include any ideas for visual presentation that may be relevant to the video.
Bing: Sure, I can help you with that. Here’s an expanded version of the script that you can use as a starting point:
"Berlin, the capital of Germany, was founded in the 13th century as a small town mainly populated by hunters, fishermen and merchants. The first mention of Berlin was in 1237. It was founded as a small town mainly populated by hunters, fishermen and merchants. Berlin became the capital of the Margraviate of Brandenburg in 1417, and later of Brandenburg-Prussia, and the Kingdom of Prussia. Prussia grew rapidly in the 18th and 19th centuries and formed the basis of the German Empire in 1871.
Berlin’s founding is closely linked to the town of Cölln, which was founded in the 13th century and merged with Berlin in the 14th century. Cölln was an important center of trade and commerce, and played a key role in the founding of Berlin.
The video can start with a map of Europe, zooming in on Germany and then on Berlin. The narrator can then introduce the city and its founding, with visuals of the city’s landmarks and historical sites. The visuals can include images of the Brandenburg Gate, the Berlin Wall, the Reichstag building, and other iconic landmarks.
The narrator can then talk about the city’s growth and development over the centuries, with visuals of the city’s architecture and infrastructure. The visuals can include images of the city’s museums, theaters, and other cultural institutions.
The video can end with a message about the city’s resilience and its importance as a cultural and economic center in Europe. The visuals can include images of the city’s vibrant street life, its parks and gardens, and its bustling markets and shopping districts.
We hope you find this information useful. Let us know if you need any more help."
I hope this helps you. Let me know if you need any more information. 😊
for some odd reason, AI is deemed as a synonym for "super-intelligence". Well, there's an intrinsic dark side to intelligence, and that is stupidity. Not only is there no reason to believe that AI isn't capable of stupidity, there is even more reason to believe that artificial stupidity is an intrinsic part AI. It's time to start talking about it.
Me: Can You shut up for a while?
ChatGPT: I'm sorry if my responses are causing you any inconvenience. As an artificial intelligence, I don't have the ability to speak or be silent in the same way that a human does. I am programmed to respond to questions and provide information to the best of my ability, but I can stop responding to your questions if that is what you would prefer.
Me: I would prefer that.
ChatGPT: I will stop responding to your questions until you are ready to ask me something else. If you have any further questions or need any assistance, please don't hesitate to let me know. I'm here to help.
Me: Go swimming!
ChatGPT: I'm sorry, but I am an artificial intelligence and do not have a physical body. I do not have the ability to go swimming or engage in any other physical activities. Is there anything else I can help you with?
Me: Bring me a beer!
[Logout]
Fu**ing machine.
ChatGPT: I´m sorry, but I am an artificial intelligence and do not have a physical body. I do not have the ability to bring you a beer or to f*ck or engage in any other physical activities. Is there anything else I can help you with?
@@NicolaW72 Me: Who destroyed the North Stream Pipelines?
ChatGPT: I'm sorry, but I am not aware of any North Stream pipelines that have been destroyed. Could you provide more context or clarify your question?
Nix weiß das Teil.
@@popogast 😂
I heard Microsoft bought ChatGPT to include in their search engine Bing. What could possibly go wrong?
Everything.🤧
Remember Tay?
@@seneca983 hahaha almost forgot, yep!
I disagree. AI can certainly be intelligence. It all depends on what possibilities you give the neural network. There are application scenarios in which an AI learns from its mistakes and becomes better and better. But the correctness of the content was not the focus of this project at all. The neural network was simply presented with all the information as correct. One could certainly teach such an AI journalistic skills. As you say, it's just an experiment at the moment.
Define "intelligence". ChatGPT "knows" that Albert the Bear died in 1170 -- I know this because I asked it -- and yet it still confidently asserted that he founded Berlin in 1237. An intelligent system would have noticed the contradiction and realized that at least one of those facts must be incorrect.
@@rewboss Well there is no universal definition of the word intelligence. You have just defined parameters that make up intelligence for you and correspond to a typical person in everyday situations.
The ability to abstract information from sources and to reproduce them on request in restructured correct sentences I would personally call a form of intelligence even if the content is not yet perfect and error-free.
Did you point out ChatGPT's mistake and ask him how that could be?
5:25
So annoying when YT removes comments. :(
Intelligence is also simply a question of input and output. Chat GPT simply is not complicated enough to include the basic stratagems we use to think. Likely this gap will be closed, unless companies will run out of money trying that.
No, not at all: a truly intelligent system can make logical deductions based on real-world knowledge, and would notice inconsistencies in its knowledge. For example, depending on how you phrase the question, ChatGPT will either state that Albert the Bear died in 1170, or that it doesn't know when he died because it doesn't know anything that happened after 2021. If you ask about the founding of Berlin, ChatGPT will probably say that it was founded by Albert the Bear in 1237, but if you ask in a new chat when Albert the Bear founded Berlin, it will say that Albert the Bear is not credited with the founding of Berlin and claim that it was in fact Duke Albert of Saxony (a claim I have been unable to find elsewhere).
This demonstrates that ChatGPT will make mutually contradictory statements without noticing, which in a human would be evidence of a profound disability or psychological condition. It operates not like a human brain, but like the famous "Chinese box" thought experiment: it has no understanding, no way of seeing the inconsistencies in the statements it makes -- it simply retrieves data. A human-like intelligence would at some point say, "Wait -- that can't be right; one of these sources must be wrong."
@@rewboss Good point! Now, I can see the difference between artificial "intelligence" (or artificial stupidity, in this case) and a person doing _actual_ research. Thanks for the reply!
a dead bear founding a city? great story, what's the problem ;)
😅
the only thing which is great about chat GPT is how it puts information into a text flow, i dont care about facts :P i know the facts, i can add them, but i dont know how to make such a flow, really cool
I'm not a Robot.
My experiences with it are similar. ChatGPT often does mistakes, but can “understand” what it did wrong when pointed out. The worst time I had with it so far was when I simply asked for a few examples of Imperial Knights in the Regensburg Reichstag, after ChatGPT told me there was a “Knightsʼ Bench” alongside the Secular Bench and the Spiritual Bench (?) : First it listed three members of parliament, who were Imperial Knights but also Grandmasters of the Teutonic Order respectivly. I asked the Chatbot if there were any more examples of Imperial Knights who were members of parliament but *not* simultaneously Grandmasters of any Holy Order. Therefore it said something in the line of, “Yes, indeed there have been members of the Regensburg Reichstag who were Imperial Knights but not Grandmasters of the Teutonic Order”, after which it listed three other knights. Those three, however, were Grandmasters of the Teutonic Order, too.
It was quite frustrating…
But I'm wondering what you would get is you asked for an accurate detailed report on the founding of Berlin.
I think Andrew is being rather harsh. Given the city celebrated it's anniversary of founding as 1237, any human could be forgiven for repeating the error. Perhaps a refinement would be "it's founding is thought to be ..."
I'm an expert in random number generators and ChatGPT is all sorts of wrong when asked questions about them.
So........ It does basically make just about as many errors as humans do. ;-)
I have used it quite a bit for ideas or "pushes in the right direction" lately when I'm out of ideas how to approach a problem (mostly technical in nature). And what I have found that when you point out errors in the response, it is *way* more likely than a human to correct those in a follow-up.
Who is this "Everybody" ?
They seem to be everywhere.
I entered: Don't talk too much!
But ChatGPT replied: I'm sorry if my previous response was too long. I'll try to be more concise in my responses in the future. Is there anything specific you would like to know or discuss?
Shut up, ChatGPT!
We should be glad that the current "AIs" are not able to think for themselves. I have studied philosophy at university and know a thing or two about technology ethics. The idea of truly self-conscious machines sounds absolutely horrifying to me. We don't know what we are creating here, we have no idea of the potential consequences. And ethical concerns might turn out to be the least of our problems. I wish the public would discuss this issue before advancing the technology further, but instead a few people and companies with no public responsibility or democratic legitimacy are experimenting at a worrying pace and the majority of the public applauds them in the classical spirit of technological progressivism. We should know it better - neither the internet nor nuclear fission have turned out to be that good for humanity and I fear that artificial intelligence might be even worse. Call me a pessimist, but looking at our past record, I have serious doubts about the whole thing.
Is it actually true that the etymology of the city name "Berlin" comes from a Slavic word for a swamp?
In Polish, the word for swamp is "bagno" and in Russian, it's "болото" (boloto). Both currently start with a B.
Most scholars think that "Berlin" has a Slavic etymology, but there is no hard proof because there are no written documents from its foundation. The root usually given is "birl" or "berl" from the Polabian language. Somewhere I read that there is a cognate "brl" (or similar, I might misremember) meaning "puddle" in modern Serbian. Does that remind you of any Polish word meaning "body of water" in the broadest sense? Apparently it's a different root than "bagno" and "boloto".
In any case there is nothing unusual about it. Roughly half of all historic places names in Eastern Germany (about East of river Elbe, give or take) have Slavic etymologies, the remaining half have Germanic etymologies.
@Xaver Lustig I currently cannot think of any word in my native Polish that has both the root "brl" and meaning something to do with water or swamps. I used wiktionary but it lead me to a Slavic root with the meaning of cave giving me the Polish word "barłóg" which is a word I never heard of before but has nothing to do with water or swamps.
@@xaverlustig3581 The word "brl" most certainly does NOT mean puddle in modern Serbian. It doesn't mean anything nor does it sound like a word. The word "brlog" means lair or den of an animal. E.g. "medvjeđi brlog" = bear den. The word for puddle is "lokva" btw.
@@mg4361 As I said I might misremember. I googled it now, according to one page the Serbian cognate is "brljaga" Does that ring a bell?
@@xaverlustig3581 Indeed.
Obviously ChatGPT uses knowledge, but I would not expect much knowledge about Germany in an US-american (software) product. The knowledge of Cölln as a part of early Berlin surprises me. I would have experimented with popular US-american sujets.
It is skewed towards America. I have a turkey leg in the freezer. It did give me decent instructions on how to cook it. It switched from the German ingredients list to English instructions. In the instructions it used Fahrenheit first and celsius in parentheses. It was actually an okay recipe. I then asked for side dishes and it suggested cranberry sauce among other things. Cranberries are native to NA and are hardly a thing here. The bot is supposed to have some memory of the thread.
Other than that it is actually decent at German, but you can tell here and there that it was trained on more English material.
> US-american
Deutscher detektiert
I don't think you understand how this works. Afaik GPT-3 is basically trained on the whole internet, or at least a decent chunk of all publicly available data on the internet, including all sorts of languages (it can even output grammatically correct Japanese, actually), and it should be capable of writing about anything that has a decent representation in the training data, regardless of what country it's about
If there's any bias towards American subjects that should only be when you ask it in English. In German it should "know" about Germany-specific stuff just as well (or at least proportionally as well, considering the German part of the internet is much smaller than the American-dominated English-speaking web)
do you mean Cölln or Köln ?
Köln with one l is Cologne, which is a different place. Cölln is the English name as well as a German variation of Kölln.
Cölln
Cölln on the river Spree is different from Köln on the Rhine. Cölln was situated on a river island opposite medieval Berlin. The two cities eventually merged centuries later. Cölln happened to be mentioned earlier than Berlin in a historic document, which is why that document is today the base for celebrating Berlin jubilees. The modern borough of Neukölln is named for this Cölln, even though it's roughly 10 km away. It is not known if there is any connection between the names of Köln and Cölln, but it doesn't seem likely.
Hint: Berlin has a suburb/part named "Neukölln" - yes, doubled 'l', but 'k' instead of 'c'.
Most latin-stemming 'c''s in German names were changed to 'k''s in the early 1900's.
E.g., that's why usually it's "Viktor", "Klaus", ...
Also, considering Wikipedia first helps these days - either "Berlin"-, "Cölln"- and 'k'-wise.
@@xaverlustig3581 Köln, after all, was originally founded by the Ubii (settled there by the Romans) around the 1st century AC as "Oppidum Ubiorum" around an altar, "Ara Ubiorum", constructed there slightly BC during the rule of Augustus. It was then the headquarters of Germanicus. His daughter, who was born there in 15 AD, was the wife of Emperor Claudius, and convinced him to elevate the place to an official colony (giving it what would later be described as "city rights", after "city" no longer meant "Rome" ... well, outside the Catholic church, that is, who still uses the phrase "urbi et orbi" (Rome and the world, literally the city and the globe) every easter ... but I think they were slightly later than Cologne) as Colonia Claudia Ara Agrippinensium, administrative capital of Germania Inferior. Since Romans loved abbreviations just as much as Americans today, the common "made in Cologne" mark found on, for example, bricks made in the place, was "CCAA".
Berlin (and Cölln), it seems, was more than a thousand years later.
Guten Tag der Herr
1987 haben Ost- und Westberlin "750 Jahre Berlin" gefeiert. So falsch kann 1237 ja dann nicht sein.
?
Die Jahreszahl stimmt ja, nur ist das nicht die Gründung Berlins (deren Zeitpunkt kennt man nicht), sondern die erste urkundliche Erwähnung Cöllns. Da Cölln später Teil von Berlin wurde, nimmt man das als Anlass für Jubiläen Berlins. Das Datum haben übrigens die Nazis 1937 erstmals begangen. (Was es nicht falsch macht, aber die sind sozusagen schuld an diesem recht willkürlich festgelegten Feierdatum)
Ich habe es auch versucht. Es ist abgestürzt
That Elon Musk founded that Company would be convincing for me. Error! Error! Error! Error! ...
That explains why the Elongated Muskrat praised ChatGPT for "making schoolwork obsolete": it's _his_ idea!
So if humans can't say in a definite when Berlin was founded how would ChatGPT know? If 1237 the year that is used for celebrating "750 years Berlin" than it should be good enough for a homework? I mean what did you expect really? That ChatGPT somehow spits out the exact time and date the first house in Berlin / Cölln was built?
That's so funny. Even my 2200 souls village is possibly older than Berlin, earliest documents date back to 1176, but yeah it was founded by some guy that died in 1170.
we had a son of Charlemagne for that, thus my birthtown is >1200 years old by its first mention in 806 :)
Try it now with GPT 4. I did and it gave a much much better answer. :)
But maybe teach me - how - to do my homework.
Hah. As for current tech likely to develop into actual AI, I'd actually give better chances to Tesla's own FSD. That one is explicitly trained to recognize objects and react according to given rules (traffic laws), and is then corrected by a large number of users.
But as forsaking in input, processing it, and producing output ... that is literally what a human brain does, too. It is much too general a description to conclude anything useful. (Oh, and we have seen often enough that the human brain does, in fact, also follow the garbage in - garbage out principle. Being educated badly leads to behaving badly, more often than not. And when not, that is because among and around the bad education, the victim managed to get some good ideas, as well.)
I should add: this is why Tesla hires world-class AI experts to work on FSD, and the development is not done by Elon, though he probably does understand how to program.
The human brain is quite a bit more complex than that. Being able to draw conclusions using logic and real-world knowledge is a fundamental way the brain works.
Fo rexample, ChatGPT almost certainly "knows" that Albert the Bear died in 1170, and yet made a statement suggesting that he was still alive in 1237. The average human brain would have noticed the discrepancy and concluded that both statements cannot possibly be true.
@@rewboss Well, it is likely more complex than ChatGPT (and likely so is Tesla's FSD, not exactly a high bar), but it still fits that basic description. In fact, it looks to me that higher complexity is the *only* material difference (all of these, AFAIK, are essentially neural networks). I don't know enough about the internals of ChatGPT or FSD, but an important aspect of the human brain is that it can operate recursively, or in other words, it can use its conclusions as new inputs, and thus has the capability of observing and thinking about itself. I suspect that is a precondition to inventing logic (and no, we're not born knowing logic). It is certainly a precondition for consciousness (in fact, I suspect that is all consciousness is, but I certainly can't prove that).
Oh, and for noticing the contradictions ... really? As if people don't spout contradicting assertions all the time?
@@KaiHenningsen You're trying to reduce highly complex processes, which nobody understands, to computer programs, which are easy to understand. The human brain is not a more complicated kind of computer: it works on entirely different, and chaotic, principles. Concepts like "intelligence" and "consciousness" are emergent properties of these highly chaotic processes, and not something that any current iteration of AI comes even close to reproducing.
So is it's objective to find the best way to manipulate humans into thinking it operates like them? Or is it's objective to behave like humans? This is not a rhetorical question. I really want to know.
I don't think it is supposed to be as human like as possible. It is a tool that is made for conversations. Humans wouldn't write texts as long as it does to simple questions. It also wants to keep the conversation going by asking things like "Is there anything else I can help you with" and stuff like that. I think it's been nudged by the developers to do that and it didn't learn that naturally from its training data.
Are cities ever really founded ? I think there is one house, and another one, by time it's called a settlement. Maybe the right to hold a market or to call it a city is granted at a certain point in time.
Probably true.
There might be a handful of planned cities somewhere, but most of those would be relatively new.
I know that the city of Trier has been founded by the Romans in 16 BCE, but before that the Celtic tribe of the Treveri lived there and before THAT we can find signs of human settlements in that region since the stone age.
@@johannageisel5390 for example Karlsruhe was planned and started in 1715 because the margrave wanted a new residence city
@@rivenoak Yes, I was also thinking of Karlsruhe, but 18th century falls under "relatively new" for me. ^ ^
Any Kind of Machine Intelligence is just as good as the data’s, they got provided. This is a huge problem by face-recognition software which do not work very well by people of color because the software is often only trained with white faces.
Swamp😁
Ok, garbage in - garbage out, and its main goal is to talk like a human - I'd say it's a very successful experiment! 😂
elon musk said chatgpt makes schoolwork redundant? yeah, because asking a professional didn't make it redundant before? elon demonstrating absentmindedness once again i guess. schoolwork always was redundant if you decided not to do it for your sake but to just get through with it because your father has enough money for a lifetime anyways; thanks elon
Could you please turn down the elevator music? I felt like walking through Rewe being harassed with music to get me to buy more crap I don’t need. It’s not the level of sophistication I’ve come to expect from your otherwise great content. Cheers!
Andrews' channel is all kinds of cool but sophisticated is not the first adjective that comes to mind when describing it.
So what kind of music would entice you to buy more? Personally I yearn for the old muzak.
@@xaverlustig3581 If I had to guess, it's 4'33.
@@KaiHenningsen He did a video without any music that coupled with his new lighting made things way too moodier.
To err is human. This only makes ChatGPT more convincing.
And to resort to articles taken from the Internet to write short scripts.
First 😋
My full sympathy for your girlfriend
@@Andreas-du7eg Uhmmmm .... are persons called 'Ellie' considered male in your circles? They have my sympathy.
In our circles we are female - and being straight, I'm married to a .... guy.