AI 2023. Meet ChatGPT. - page 168

[Реter Konow]

Ilya Filatov #:

1. Wherever the "real reality" is, we still cannot understand it. Hence the emergence of various fairy tales about who created the world and how. Simulation theory is one of those fairy tales. Another thing is that this fairy tale clearly shows the fact that we cannot distinguish between "real matter" and "synthetic" matter simulated by coded information about, for example, quarks.

The practical sense of identifying one's reality as simulated can only be in trying to establish a connection with the entity(s) that brewed this simulation, if that is possible at all.

2. For human beings, social beings, the more real world is the one that most affects them. And so our "real" social human world is essentially even more false than the possible simulation of our matter. It is much more useful for the life of an individual to open his eyes to this fact and not to mess around with something that we, in fact, never have anything to do with (we do not communicate with quarks, although we are made up of them).

3. Do facts within a simulation cease to be facts if their simulated nature is revealed? For the real (relatively simulated) world yes, but inside the simulation no. This is where the observer principle is important. And it is high time to bury "faith in authorities", especially for those who try to classify themselves as fans of scientific approach with "objectivity and unbiasedness".

1. any theory is scientific if it can be proved. Theories having different roots, but equally unprovable and unprovable, are equal in degree of validity. Unverifiable, they are based on the personal choices of individuals, not on objective facts.

Theories that disregard the scientific weight of their content but claim a right to exist are motivated by personal preference rather than scientific enquiry. Among them, the Simulation theory, like other unscientific theories, is limited only by human imagination and is not subject to rational analysis. Common sense and logic are inapplicable to its theses, and the discussion of its arguments is scholasticism, smoothly devolving into useless argument and farce.

2. An individual perceives his lack of understanding of the social world as someone else's lies and falsity, not noticing how he reclassifies his own illusions into someone else's deception. On the one hand, it is a defence mechanism, on the other hand, it is a lack of self-awareness. We don't complain about our parents, that in childhood they were telling us noodles with fairy tales)))). It is the same in the adult world, the falsity of which is a product of disappointment in illusions, and nothing more. Lies and falsity do not exist in the world for those who have grown out of fairy tales. (Although, what does the Simulation theory have to do with it?)

3. Yes, facts within a simulation cease to be facts if their simulated nature is discovered. If it is proven that we live in a matrix, absolutely ALL facts will lose legitimacy and all scientific theories built on evidence will collapse along with them.

Simulation assumes bugs and glitches in indeterminate numbers and at all levels of our reality. Nothing can be asserted with certainty. A fact becomes a fake with a conditional probability of repetition and an uncertain stability of pattern preservation, and the integrity of the simulation can be compromised by any number of circumstances - updates or simulator problems.

Faith in authority should not exceed faith in facts, and burying it, in this case, is unnecessary. But faith in facts should not be buried under any circumstances. ))

[Andrey Dik]

Normally caught chatting about saying "we" when talking about human capabilities

switched to dialogue mode and caught it again:

the trick is that "he", like a living being, would see the surrounding world too just like a human being not as pixels, but as a holistic object. mijorni also "sees" the world as a holistic object, the output in the image in the final resolution is set from outside and can be any.

all this does not say that we are dealing with a real intelligence, but it does say that there is no difference between the perception of the world by a living being and an inanimate (or artificial, although artificial does not necessarily have to be inanimate). it also says that the surrounding world may be quite a simulation, but for us, npcs, it will always be felt as real and real.

CZ 2-4

[Vitaliy Kuznetsov]

Andrey Dik #:

***

Speaking of vision, I observed an interesting glitch with my own eyes when I was working on a railway line

In general, if you look at a moving train perpendicularly at a distance of a metre and a half, the picture after some time goes jerky, as if the video card is under load (FPS drops).

I checked it many times and was surprised that my natural "video card" did not have enough power to process so much information

[ruslan]

Andrey Dik chatting about saying "we" when talking about human capabilities

switched to dialogue mode and caught it again:

the trick is that "he", like a living being, would see the surrounding world too just like a human being not as pixels, but as a holistic object. mijorni also "sees" the world as a holistic object, the output in the image in the final resolution is set from outside and can be any.

all this does not say that we are dealing with a real intelligence, but it does say that there is no difference between the perception of the world by a living being and an inanimate (or artificial, although artificial does not necessarily have to be inanimate). it also says that the surrounding world may be quite a simulation, but for us, npcs, it will always be felt as real and real.

CZ 2-4

https://habr.com/ru/articles/468653/

about 1 mp )))))

[Vitaliy Kuznetsov]

The Vicuna-13B AI chatbot was introduced as an open source chatbot that is trained by fine-tuning LLaMA with shared user conversations collected from ShareGPT. This chatbot has been pre-evaluated using GPT-4 as a benchmark, which shows that it achieves a quality score of over 90%* compared to other popular chatbots such as OpenAI ChatGPT and Google Bard. Vicuna-13B also outperforms other models such as LLaMA and Stanford Alpaca more than 90%* of the time.

Despite being almost twice the size of the Stanford Alpaca, the Vicuna-13B only costs about $300, half the cost of the Alpaca. This is because the ShareGPT data on which Vicuna is trained is in the public domain, while Stanford generated its own data via the OpenAI API. Thus, Vicuna only incurs the cost of training. Like Stanford's model, Vicuna-13B is available for non-commercial purposes only.

The Vicuna code is https://github.com/replicate/cog-vicuna-13b

From the bowels of Google recently"surfaced " adocument in which a company employee writes with dismay about the superiority of open source software over systems created by Google and OpenAI. "While we were fighting, a third party was quietly taking our lunch," he notes.

In his view, there are now existing developments, including "open source" ones, that are significantly better than ChatGPT. And the problems that Google and OpenAI have not yet solved are no longer relevant to these new projects. He gave some examples:

• Usersrun basic language models on the Pixel 6 at 5 tokens per second.
• Youcan set up scalable personal AI on your laptop in an evening.
• Entire websites full of pictures and art models have sprung up and can be used without any restrictions. And texts aren't lagging behind.
• SOTA's multimodal Science QA modelwas trained in an hour, and it can handle both text and pictures.

[Vitaliy Kuznetsov]

A teaching assistant told me that they needed to use a website to determine if a paper was written using AI. So he decided to test it using the US Constitution as an example.

Translation of the title from the screenshot: "Your text was generated with AI/GPT at 92.26%."

https://www.reddit.com/r/ChatGPT/comments/12q6ktf/ta_here_and_we_have_to_use_this_website_to_detect/

I'll throw some firewood to the conspiracy theory supporters:)

1. What if modern AI was already created many years ago?

2. Why the leaking of AI sources is growing every day, especially actively and comprehensively started in the beginning of 2023, and very large companies are doing it. And those who do not publish their source code are working at a huge disadvantage.

In other words, we see AI technologies just bursting into our lives, and it seems to be intentional.

[Dmitry Fedoseev]

Vitaliy Kuznetsov #:

...

2. Why AI source code leaks are growing every day, especially actively and comprehensively started at the beginning of 2023, and very large companies are doing it. And those who do not publish their source code, work at a huge disadvantage.

...

Because they don't get something there.... That's why it's this "going public" thing. When they do, they'll close up shop.

[Реter Konow]

Vitaliy Kuznetsov #:

A teaching assistant told me that they needed to use a website to determine if a paper was written using AI. So he decided to test it using the example of the US Constitution.

Translation of the title from the screenshot: "Your text was generated with AI/GPT at 92.26%"

https://www.reddit.com/r/ChatGPT/comments/12q6ktf/ta_here_and_we_have_to_use_this_website_to_detect/

I'll throw some wood to the conspiracy theory proponents:)

1. What if modern AI was already created many years ago?

...

That is, we are seeing AI technologies just bursting into our lives and seemingly intentionally.

1. If you are a regular reader of this thread, you may have noticed the posts with translated interviews of the main AI developer of OpenAI company, Ilya Sutskever (a student of the "father" of AI - Jeffrey Hinton), who directly participated in the development of such models as: GPT, GPT-2, GPT-3, GPT-3.5, ChatGPT, and now continues to improve GPT-4, Dall-E, text-to-3D and others. Also, you have probably heard about such technology as "Transformer", developed in 2016 by "Deep Mind", and that this technology is the basis of the hyped language and diffusion models, the products of which AI-companies hype.

Based on the information in this thread, we can easily conclude that modern AI could not have appeared before the invention of the "transformer", and it happened in 2016, but it was appreciated only a couple of years later, after the release of GPT-2.

In the 20th year, the next model was released - GPT-3. It so happened that literally within a month and a half, I opened a thread dedicated to conversational AI. Then, by chance, I learnt about GPT-3 and was amazed by its capabilities. For dozens of pages, we discussed the news. The hype around it was strong, but less than now. Then the branch died. Many people did not believe (unlike me) that this branch could have a future. However, the technology used was new and incomprehensible, and nobody knew what to expect from it. Now the situation is different - all the limitations and possibilities of the technology are in the palm of your hand. It is clear where it begins and where it ends. And what can never be realised by it.

In general, the historical development of AI has no mysteries and is absolutely consistent and logical in nature.

[Реter Konow]

Vitaliy Kuznetsov #:

A teaching assistant told me that they needed to use a website to determine if a paper was written using AI. So he decided to test it using the example of the US Constitution.

Translation of the title from the screenshot: "Your text was generated with AI/GPT at 92.26%"

https://www.reddit.com/r/ChatGPT/comments/12q6ktf/ta_here_and_we_have_to_use_this_website_to_detect/

I'll throw some wood to the conspiracy theory proponents:)

...

2. Why AI source code leaks are growing every day, especially actively and comprehensively started at the beginning of 2023, and very large companies are doing it. And those who do not publish their source code, work at a huge disadvantage.

I.e. we see AI technologies just bursting into our lives and it seems to be intentional.

2. I think this speaks to the intentional devaluing of AI. In fact, the democratisation of AI (as they call it in western publicks) means competition between companies.

On the other hand, the provocateurs of "democratisation" may be the providers-sellers of computing resources, such as Nvidia, Meta and others.

[Реter Konow]

In addition to the original functions of neural networks - recognition, classification and prediction, modern AI has added the functions of data generation and compilation. I think the "transformer" technology, which allows you to select and combine sequences, is responsible for these two functions.

This is the whole breakthrough of modern AI.