I downloaded it to my phone and this program can really answer everything and you can't even tell them apart from human answers anymore… I find that incredible
How does this work, how is this possible?
(3 votes)
Loading...Hello everyone How can I use Elementor to place the anchor exactly where it stays when navigating and doesn't move a little further? Is there a way to set anchors individually for mobile and desktop?
Hello everyone, I hope someone can help me. I have a Yamaha Virago 125 built in 1997. My machine stood still for a long time and when I wanted to start it up the engine was idling at high speed without me giving it gas. I got on and shifted into first gear. My Yamaha…
I want to discover Europe as a flight attendant. It's hard to get into Hansa, the job is more secure but you don't earn much. With Ryanair you can start immediately if you need to, money is better. Where better to explore Europe? And which airline is better for your CV?
I know sounds pretty stupid🥲
Hello, my fuel hose was bitten through and I wanted to ask if it is a wear part because otherwise I would get it paid for by the car dealership thanks in advance to everyone
Good day dear programmers, As a project for my first-semester exam, I was given the task of reprogramming Donkey Kong Adventure in a simplified way. I've completed the first step – namely, moving along the x-axis. However, I'm currently struggling with how to make my character move along the y-axis. Perhaps someone can help me….
In itself, an AI is nothing but a lot of stochastics, so probability.
The AI is fed with training data and then trained. Thus, it is then possible to provide the best possible answer to the questions of users (in the case of GPT).
And in the answer it is recommended, because if GPT does not find a suitable answer, something is simply generated. This is also not wrong, because it is a language model, not a knowledge model. ChatGPT is therefore incorrectly applied in most cases, nevertheless in many cases gives the correct answer, but not always.
Intelligence is actually not behind it. The technology for AI was also already in the 80s and 90s. At that time, there was not enough computing capacity to use mathematics behind it commercially and for the large mass.
By the way, researchers expect about 550 tons of CO2 to train GPT 3.5, for comparison, a car is about 2 tons a year. This is the power consumption of 320 four-family households.
Well, if you exactly If you want to know, you’re not going to get around a computer science program.
The basic principle – simply explained – is an extremely broad database to which the AI can resort. New – in contrast to previous search engines – is that by corresponding programming (keyword: algorithms) the AI can independently create links between the data, so that practically unpredictable and creative results can arise on the basis of incredibly much existing knowledge. The system can constantly improve itself, so to speak ‘to learn’.
I am currently a computer science student and think this answer is quite bullshit. The basic function of generative artificial intelligence is not particularly difficult to identify the technology as a witchcraft that only trained computer scientists could understand, I find more than just questionable. Only when dealing with such technologies can one correctly estimate the chances and risks as a private individual. This is not complicated, there are thousands of fantastic and easy to understand websites and videos on the topic. In the basic study of Computer Science (Bachelor), one must be lucky to have a lecture that deals extensively with such modern technologies as they differ from traditional AIs. In the master with a specialization on AI, this can of course look different.
But to get back to the answer: the generative AI usually has no access to the data base, it was trained on the data base and the results of the training are now part of the LLM.
The comparisons with traditional search engines also suggest a little. Search engines have been wasting algorithms that we could call artificial intelligence (the actual definition is very controversial). Linkings between data are naturally also in search engines, and they have been identified by the AI for decades. The models are also usually not designed to “learn” after their initial training, inputting the past becomes simply an additional context as part of the input in the respective chat (in the case of ChatGPT, Gemini etc.). In other words, a search engine learns more about it than an LLM, as it must constantly evaluate and rank new websites.
Basically, LLMs do not have much to do with search engines, and even if the leaders in Silicon Valley like to have different, I don’t think LLMs can replace search engines – or should. LLMs are simply mathematical models that, based on their training data, calculate probabilities and generate a bit of variance (“temperature”) based on them expenditures that do not have to be actual even if they are usually linguistically correct. I do not see creativity as a human property, even if the expenditures can be “unique”.
It is neither magic nor a serious or creative source of information.
Well, that’s why you’ve put yourself in a hard spot thanks to your (previous) knowledge of computer science.
But to your arguments:
“The basic function is not particularly difficult… (etc.).” – Then the question arises, however, why this has not been developed functionally decades ago. Do you have an explanation?
“(The AI) has been developed on a data basis…” – EBEN. Thus, it uses exactly that as a starting point.
“The models are not designed to learn …” – but you can always read corresponding statements in this sense.
“Creative” are such results from AI over and over again – in that connections are made or connections between data are “seen” that have not been discovered by humans. (I have read examples from medicine; I would have to research the sources again).
In the meantime, some universities and colleges that also specifically offer a Bachelor’s degree program Aritifical Intelligence, or at least one specialist in the BSc computer science to AI.
Okay – thanks for your explanatory information.
The commentary is only reaffirmed by ignorance of technology and my criticism of your answer. I don’t intend to have a great discussion with you.
I don’t know if you’re really in a position to criticize me for my studies.
Just because something is not complicated, it does not mean that the invention was simple. Today, every child at school learns the findings that humanity has collected over millennia. I have spoken of the basic features of the functioning of the technology, not of the implementation details.
Generative AIs use a database as a basis; the statement that she actively intervenes, but is inaccurate.
As I wrote in my answer, you have to distinguish learning during training and learning after training end. Models that are published once learn only through training. These are then effectively different models. “I heard xyz” also does not refute my description of how it works.
Well, I see creativity as a purely human property, but if you want to tell her that.
I would like to make it clear once again that we are talking about generative AIs, i.e. Large Language Models. In the natural sciences, traditional AI models are used which are suitable for these tasks. These are basically different technologies. Companies and media are often thrown into a pot for marketing and it is understandable that you are quickly confused as a layman. That is exactly why I was at the heart of my statement, which is why I wanted to know about such issues. It’s not hard if you want it.
There are, of course, many levels to do so, but I try to explain the basic functioning very easily:
A generative AI like ChatGPT first needs billions of tokens (you can imagine as a kind of word) training data. These can come from websites, books, or forum entries from the Internet. Probabilities are calculated on this basis: how likely is it that a word follows another word in a specific context? If you now give the AI a new input (prompt called), it can create an output, word for word based on it and its training (so the probabilities). With a parameter called temperature, you can also make the AI choose sometimes not the most likely word – so variation comes into the answer, the same input does not always lead to the same output. The AI has no understanding of what she writes, how we humans have it; it is all a matter of probabilities, based on training data. Thus, so-called hallucinations often occur when the AI generates something that, although linguistically speaking, makes sense, is, however, complete nonsense. That is why one should always be critical of the expenditure of generic AIs.
Modern Large Language Models such as ChatGPT 4 are significantly more complex and sometimes consist of several parts that have been trained for special tasks and can solve them particularly well. In this field is currently constantly researched and there are many breakthroughs, but also many unresolved (or insoluble) problems. In principle, however, they are still based on this principle. If you are interested in the topic, there are many beautiful videos on sites like YouTube.
This is the result of decades of research and development. Machine learning hasn’t been around since yesterday.
In essence, data are divided into vectors, these are translated into a number and from there it is almost just about probability.
Thus, AI models like GPT are also able to capture contexts and, based on this, deliver the most likely correct answer.
Programming and the AIs learn more and more from themselves.
Provides beautiful videos on YT that describe how neuronal self-learning networks work.
It’s like Google, but knows everything in Google and can give the best answers.
This shit also contains many misinformation
No, that’s not true. Some are misinformation, but we really talk about a small part here. AIs have most malfunctions when it comes to ideological things, such as modern woke ideology, feminism, ethnicity and conservatism.
Especially bad is the Gimini AI from Google. Many AIs are politically charged, for example when you ask them for sexist women’s sweats, most refuse. But sexist men’s sweats are okay. You see, they’ve been given some barriers. Or some world views.
Hallucinations are a huge problem and this statement proves that you haven’t been seriously trying to get from an AI information on somewhat smaller topics. Why you suddenly complain about filters that are naturally imperfect at LLMs due to the huge training records (see Jailbreaks, or your examples), I don’t know.
With GPT4o the AI of OpenAI has become significantly better. And the emphasis is on DEUTLICH. I tested GPT4 a few months ago. Slowly, often in the middle of the sentence an error generated etc. 3.5 is also rather bad than right. Now, 4o, it’s like 100 years of work would be in it.
This has become so good that I have now extended my subscription for $20.
The AI for image generation is also 1A tip.