AI 2023. Meet ChatGPT. - page 71

 
The closer you look at the software implementation of neural networks, the clearer it becomes that there are no "networks" there. A lot of vectors and matrices with data and weights, and all sorts of parameters with calculation functions.

It's funny.


The article is VERY good.
 
Реter Konow #:
We have yet to make sense of the technology of language models and understand what's behind them.

Evals is an OpenAI model evaluation platform and open source benchmarking registry.

You can use Evals to create and run assessments that:

  • use datasets to create prompts,
  • measure the quality of completions provided by an OpenAI model, and
  • compare performance across different datasets and models.
 

Now I'm going to give you a real scare.


Remember a few years ago a law came out and since then all mobile and internet providers store all information, conversations, photos, videos.


Now that AI understands the context of what's written and what's pictured, it can find in a couple of minutes:

1. Criminals.

2. Bribe takers.

3. Those who don't pay taxes

4. How many in per cent trust the authorities, etc.

5. Make a list of trends of what people are concerned about

etc.


Just one AI question and it will identify everyone you need according to the query.


And some in-app spy app (from official shop of course) will help anyone to know if wife/husband is cheating etc.

 
Vitaliy Kuznetsov #:

I'm going to give you a real scare.


Remember a few years ago a law came out and from now on all mobile and internet providers store all information, conversations, photos, videos.


Now that AI understands the context of what is written and what is pictured, it will be able to find it in a couple of minutes:

1. Criminals

2. Bribe takers

3. Those who do not pay taxes

4. How many in per cent trust the authorities, etc.

5. Will make a list of trends of what people are concerned about

etc.


Just one AI question and it will identify all the right people according to the query.


And some in-app spy app (from official shop of course) will help anyone to find out if wife/husband is cheating etc.

Ford will start trolling drivers.

 
Lilita Bogachkova #:

Evals is an OpenAI model evaluation platform and an open source benchmarking registry.

You can use Evals to create and run evaluations that:

  • use datasets to create prompts,
  • measure the quality of completions provided by an OpenAI model, and
  • compare performance across different datasets and models.

Thanks. I'll look into it.

 
Vitaliy Kuznetsov #:

I'm going to give you a real scare.


Remember a few years ago a law came out and from now on all mobile and internet providers store all information, conversations, photos, videos.


Now that AI understands the context of what is written and what is pictured, it will be able to find it in a couple of minutes:

1. Criminals

2. Bribe takers

3. Those who do not pay taxes

4. How many in per cent trust the authorities, etc.

5. Will make a list of trends of what people are concerned about

etc.


Just one AI question and it will identify all the right people according to the query.


And some in-app spy app (from official shop of course) will help anyone to find out if wife/husband is cheating etc.

I guess we are on the way to giving AI the functions of social controllers, regulators and investigators?

Let's imagine that AI will search and find anomalies in data, gaps in reports, deviations in results, outliers and violations of protocol rules in all spheres of human activity. So much for Lem's social homeostat. But, it's not far from Orwell....

 
Реter Konow #:

I guess we're on our way to giving AIs the role of social controllers, regulators and investigators?

Let's imagine that AI will search for and find anomalies in data, gaps in reports, deviations in results, outliers and violations of protocol rules in all spheres of human activity. So much for Lem's social homeostat. But, it's not far from Orwell....

In the end, all questions will be asked and all mysteries will be solved. If there are still geometry and other problems for which there is a reward, you can try to make a fortune.

 
Vitaliy Kuznetsov #:

In the end, all questions will be asked and all mysteries will be solved. If there are still geometry and other problems for which there is a reward, you can try to win a prize.

It's AI gold rush time. The world is excited by a new achievement of technical progress. In the west it is called "Disruptive techonology", i.e. technology that undermines established businesses and economies. It creates confusion at the beginning, then changes the way of life of millions of people.

IT giants were the first to react and in their insatiable quest for profits have heated up the consumer market to the max. In the west, AI news and reviews flow endlessly. The public is in a state of excitement and expectation of new wonders. Taking into account the specificity of their mentality (they are obsessed with technology) it is not surprising.

Interest will be kept, but it is not known how long. Perhaps a few years. If a new scientific and technical breakthrough is made during this time, it will guarantee the company market demand. We can count on promising contracts and a bright future. But it will not be possible to be independent.

It is necessary to understand WHO the competitors are and what financial, scientific and technical resources they possess. At best, the fate of OpenAI is inevitable. You can even learn from their example. They grabbed a promising development of Google scientists (transformer), developed and scaled it. A simple plagiarism with modifications turned into a grand success. Well done (I praise with a degree of sarcasm).

That's why I stopped believing in the development of local AI on personal computers. Even if it becomes technically possible, the language model will not be independently integrated into the Microsoft or Google development environments and workspaces on a computer. And its own environments, neither the user nor a third-party company will create their own.

Consequently, the maximum potential of LLM can be extracted only in environments belonging to IT giants, and it is such integrations that the consumer public will be oriented towards. Even a relatively large company will not be able to break through without having a toolkit in which to implement AI. This toolkit must not only exist, but also be demanded by users around the world. At this point, we can put an end to this issue.


P.S. Today it is more promising to make a "kush" in the field of Artificial Intelligence than in Geometry....

 

I tried Bing to write different codes. I caught myself thinking that we will all turn from coders into editors of codes written by AI. There are a lot of errors in large codes, but almost none in short tasks. Here's an example: (find the discrepancy)

Bing, write a function for determining the handle of a moving average indicator in MQL5, using the standard library, with checking the correctness of getting the handle. Use a class to write it.

Message sent. Bing, write a function to determine the handle of the moving average indicator in MQL5, using the standard library, with the check of the correctness of obtaining the handle. Use a class to write it.

Of course! Here is an example of a class for defining the handle of the moving average indicator in MQL5 using the standard library and checking the correctness of getting the handle:

#include <MovingAverages.mqh>

class MA_Handle
{
private:
   int handle;
public:
   MA_Handle(string symbol, ENUM_TIMEFRAMES timeframe, int period, ENUM_MA_METHOD ma_method)
   {
      handle = iMA(symbol,timeframe,period,0,ma_method,PRICE_CLOSE);
      if(handle == INVALID_HANDLE)
      {
         printf("Error: failed to get MA handle");
      }
   }
   
   int GetHandle()
   {
      return handle;
   }
};

This class has a constructor MA_Handle , which takes symbol symbol , timeframe timeframe , period period and moving average method ma_method as arguments and defines the moving average indicator handle. If the handle fails (the value INVALID_HANDLE is returned), the constructor displays an error message. You can use the GetHandle() method to get the indicator handle and use it for further work with the indicator data. Please note that before using this class you should connect the standard library MovingAverages.mqh .

 

Do you check code written in MQL5 in MetaEditor?


Message sent.

No, I cannot check the code in MetaEditor or any other development environment. I generate code based on my knowledge and information from web search. However, I cannot run the code or verify that it works. Please make sure the code works properly before using it.