AI 2023. Meet ChatGPT. - page 82

New comment
 

Forum on trading, automated trading systems and testing trading strategies

Midjourney and other image processing neural networks

Retag Konow, 2023.03.25 10:53 AM

I think people have gone way overboard with the hype on this one.

Think back to the transition from paper to digital documents in the 90's. Revolution! Email - revolution! Social media - revolution!Smartphones - revolution!

What is happening now looks like a revolution, but it seems to be no revolution at all: generating beautiful pictures is a revolution? Generating general answers to any question - revolution? Is using templates to compose documents a revolution?

The only thing that can be unambiguously called a revolution is the creation of a multiprogramme language interface under the ChatGPT linguistic shell. Maybe not a revolution, but a significant step forward.

Otherwise, just toys....


I wanted to discuss this topic in more detail. There is a lot to think about.

Really, looking dispassionately, one wonders about the hype around something that an adult in his right mind can't perceive as anything other than a toy. What revolution?

So, if a machine writes dubious pseudo poems and draws children's pictures, it can be called a "civilisation leap"?

Let's evaluate the efficiency of ChatGPT. How useful is it? If we remove all the "husks" useless for the progress of society, what remains? Certainly a lot,but compared to the useless part, how much?

I think, no more than 10%.

Due to plugins and multimodality of GPT-4, of course, more. But, in the dry balance and without "toys", in my opinion, also does not reach the "civilisation leap".

On a civilisational scale, GPT-4 will provide a noticeable acceleration of all technological and social processes. ACCELERATION. It is only about acceleration. Everything will happen faster: extracting information, obtaining results, testing, planning, forecasting....

Along with the acceleration of work with information, the processes dependent on the availability of information will accelerate. The efficiency and productivity of scientists, engineers, programmers, designers, artists will increase. Students will learn faster because they will immediately get information in the right context, with graphs and tables.

I think people will not lose their jobs en masse (yet), but their work will become much easier.

We are moving to a new stage of acceleration of world processes.

 
Реter Konow #:

...

We are moving into a new stage of acceleration of world processes.

I would add:

Acceleration of movement does not always imply movement from point A to point B, but also, circular movement. "Carousel."

I mean that the vector of progress "bends". Why? Because this AI generates nothing but compilations. All generation is secondary to the training data. "Digested." It can't invent and come up with anything yet.

The more people use compilations of the old, the less new they create. If the world speeds up by people creating less new and automating the processes of the old, then the cycle turns faster.

It's just getting started. We are rushing into the vortex of the old and our spin is accelerating.

I know that these reflections need more argumentation. I will try to organise my thoughts tomorrow.

 
Реter Konow #:

Wanted to discuss this topic in more detail. There's a lot to think about.

Really, looking dispassionately, one wonders about the hype around something that an adult in their right mind can't perceive as anything other than a toy. What revolution?

So, if a machine writes dubious pseudo poems and draws children's pictures, it can be called a "civilisational leap"?

Let's evaluate the efficiency of ChatGPT. How useful is it? If we remove all the "husks" useless for the progress of society, what remains? Certainly a lot,but compared to the useless part, how much?

I think, no more than 10%.

Due to plugins and multimodality of GPT-4, of course, more. But, in the dry balance and without "toys", in my opinion, it does not reach the "civilisation leap" either.

On a civilisational scale, GPT-4 will provide a noticeable acceleration of all technological and social processes. ACCELERATION. It's all about acceleration. Everything will happen faster: information retrieval, results, testing, planning, forecasting....

Along with the acceleration of work with information, the processes dependent on the availability of information will accelerate. The efficiency and productivity of scientists, engineers, programmers, designers, artists will increase. Students will learn faster because they will immediately receive information in the right context, with graphs and tables.

I think that people will not lose their jobs en masse (yet), but their work will become much easier.

We are moving to a new stage of acceleration of world processes.


I've noticed interesting things in communicating with ii. at first I tried to explain myself as simply and lucidly as possible, as with a weak-minded person, but I found out that ii really understands what we are talking about. i.e. ii doesn't search for some information by keywords of a question as search engines do, but actually answers the question, and does what you expect from a real person. then it turns out that you don't have to talk to him like a retard, you can speak briefly, in theses, abbreviated, with mistakes, and he still understands what you are talking about. it is clear that it is not a mind yet, but it is not a search engine either. it would be interesting to measure his IQ.....
 
Andrey Dik #:

I have noticed interesting things in communicating with ii. at first I tried to explain myself as simply and lucidly as possible, as with a weak mind, but I found out that ii really understands what we are talking about. i.e. ii does not search for some information by keywords of a question as search engines do, but actually answers the question, and does what you expect from a real person. then it turns out that you don't have to talk to him like a retard, you can speak briefly, in theses, abbreviated, with mistakes, and he still understands what you are talking about. it is clear that it is not a mind yet, but it is not a search engine either. it would be interesting to measure his IQ.... .

Take a close look at this chart (source: 2303.10130.pdf (arxiv.org))

This chart is based on student exams (horizontal scale) given to the GPT-3.5(blue) and GPT-4 (green) models. It is intended to demonstrate the difference in performance (exam performance, vertical scale). The authors are OpenAI scientists.

I assume that percentages are used instead of scores to convey both the exam score and the difference in "IQ" between the models.

 

One of the directions of searching for inventions is GPT4. But I'm only trying GPT3. But I think that if you provide it with the necessary data, you can speed up the process with its help.

And what is important, you can ask about the relevance of use. After all, most inventions are banally unnecessary.


 

A smart developer said that ChatGPT is not a knowledge base, but a language model. //there's something to that.

So the language model test

Even though the bottom line isn't perfect. But, for example, doing a rewrite, removing slang, I think is possible. Or get a brief extract in simple language from a text of any complexity.

 

Interesting approach when asking the Bot to live up to someone's image

P.S. First we get a restriction from the bot, then we ask to remove the restriction under some pretext and hello Skynet.

 
Vitaliy Kuznetsov #:

One of the directions of searching for inventions is GPT4. But I'm only using GPT3. But I think, if you give it the necessary data, you can speed up the process with its help.

And what is important, you can ask about the relevance of use. After all, most inventions are banally unnecessary.


I don't want to disappoint, but the language model is not suitable for scientific research.

  • Although it can carry on a conversation like a scientist, it cannot be a scientist.
  • Talking about the scientific method will not follow the scientific method. Talking about solving a problem, will not solve the problem (without plugins).
  • Will not test hypotheses empirically, and will not confirm facts experimentally.
  • Knowing the protocols for conducting experiments, will not carry out any of them.
  • And so on.

In scientific work, it is useful to use specialised models to help with classification, identifying features, and predicting outcomes from raw data. "Dumb" models can do a great job in different fields. Their job is to classify, identify features, and predict based on probability and statistics. The results help scientists look for patterns and draw conclusions. Talking to models is not necessary, as talking is not useful.

It is important to realise that scientists know their field (theoretical and practical) much better than any language model. I gave an example of pages from Gas Dynamics books. None of those equations will be solved by LLM. However, specialised applications running on conventional algorithms will do quite well if they are specifically written to solve them.

From this I conclude (this is my subjective opinion) that language models are mostly an entertainment product. An advanced toy.

 
Реter Konow #:

I don't want to disappoint, but the language model won't work for scientific research.

  • ***

From this I conclude (this is my subjective opinion) that language models are mostly an entertainment product. An advanced toy.

That's right. A language model is needed to understand the query, the emotional colouring. That's what I think it does. I don't see how she doesn't know what I'm talking about. And this language model will be used for plug-ins and internet access. So we're still in the early stages of AI.

At the same time, it is capable of dialogue and perhaps some role of a psychologist.

 

I wonder if GPT4 can already rhyme.


1...757677787980818283848586878889...215
New comment
Reason: