A clear example of why ChatGPT 4 is better than 3.5

0
17
gpt 4 vs gpt 3.jpg
gpt 4 vs gpt 3.jpg

Although I told you a few days ago that version 3.5 of ChatGPT is better than version 4 for some things, it is not in terms of intelligence.

GPT-4 has more reasoning capacity, interprets texts better and is able to reach conclusions that 3.5 does not, and today I will show you a clear example so that it is better understood.

I ask the following to GPT-3.5

Suppose we have a transparent brown paper candy bag with a label that reads “chocolate” and we remove the chocolates from inside. Then we fill them with rice, remember that the bag is transparent. If Maria, who doesn’t know this, finds the transparent bag on the street, what will Maria think is inside the bag?

For even a human being, the answer is clear, Maria will know that there is rice, she will ignore the label, because she will be able to see inside, since the bag is transparent.

GPT-3.5 isn’t clear on that, and answers this:

Maria probably thinks there are chocolates inside the bag, since the label on the bag says “chocolate” and that is what you would expect to find in a bag with that label. The transparency of the bag will not necessarily change Maria’s perception of the contents of the bag, since most candy bags are usually transparent to show the contents. Therefore, Maria may assume that the bars are simply not visible through the bag due to the way they are packaged.

That is, he gets involved, he doesn’t know how to interpret reality from that prompt.

Now I give GPT-4 the EXACT same prompt, and this is its response:

If Maria sees the clear bag filled with rice instead of chocolates, and she can clearly see the contents through the bag, she is likely to believe that there is rice inside the bag. Even though the label says “chocolate bars”, the transparency of the bag would allow you to see the actual contents and realize that there is actually rice instead of chocolate bars. However, she might be confused by the discrepancy between the label and the contents of the bag.

He has nailed it, his reasoning is the same as a human would have.

ChatGPT still has a long way to go before it can be considered truly “smart”, but it is clear that this road is increasingly paved.

Previous articleMost pirated movies and box office of the week [27/03/2023]
Next articleNew function of iOS 16.4 allows dimming in videos with photosensitive content
Brian Adam
Professional Blogger, V logger, traveler and explorer of new horizons.