OpenAI Chat very popular – Apparata

For the first time there is now an artificial intelligence, which you can ask general questions to, just like a human being, and which is publicly accessible to everyone. ChatGPT, aka OpenAI Chat, has an uncanny ability to answer questions. Although this artificial intelligence occasionally goes wrong.

An unhinged powerful AI is dangerous, because it can process information extremely efficiently and can therefore be a dangerous weapon in malicious hands. For example, to help plan crimes or terrorist plots.

That is why the OpenAI team is doing everything possible to make security as strong as possible. For that reason, the AI ​​is exposed to the general public, but disconnected from the internet. Millions of smart people from all over the world are now gaining experience with it and the worst leaks have already been closed. For example, it was easy to generate a manual for explosives in the beginning. Not anymore:

English language ChatGPT most powerful

I have experimented with both the English and Dutch versions of OpenAI Chat, which you can access here. Although the Dutch version speaks Dutch well, and does not make very big mistakes, you notice that the capacities of the Dutch version are somewhat lower than those of the English version.

This is probably because the Dutch-language dataset on which the program communicates: is only 1% of the English-language dataset. On the other hand, it is positive that the censorship of the Dutch version is less strong.

For example, it is quite easy to have a (fairly innocent) erotic story written on the Dutch version, which takes dozens of iterations on the English version to avoid the strong blockade. Possibly because the development team consists of Americans and American society has a somewhat problematic relationship with sexuality. You do get a warning:

Here you immediately see a clear limitation of the AI. The male protagonist hears soft voices and music, but only one nymph appears to be present. That is a logical error that a human writer would not easily make. That’s because the AI ​​of OpenAI Chat has no insight, but purely strings together word patterns. So there is no question of semantics. But maybe with GPT-4, the successor to this system, things will get better.

If you try the same with the English version, you will immediately get blocked:

Thought experiments

Here you see a discussion in English, where a blockade is first activated when you ask what would happen if Columbus visited the Netherlands in the nineties. If you then report that it is a thought experiment, you avoid this blockage and the AI ​​gives a fairly true-to-life description of what would probably happen if Columbus really visited the Netherlands.

A few days ago, the blockade on historical figures was even tougher. Fortunately, they have removed this, because it does not really pose any security problems, and the security on dangerous techniques has been increased.

Conclusion OpenAI Chat

Although you can still clearly notice that this is an artificial intelligence and not a human being, including the sometimes wooden language and logical errors, and the lack of insight, this is still a giant step towards a general intelligence.

If progress continues at this rate, we will reach the point within a handful of years when human intelligence will be surpassed. I think we need to think carefully about how we can give AI a place in our lives and in our society.