GPT4All - The better AI chatbot?

[Kurzer]

erm... Mount Everest is 29029 metres high?

AI's are just immensely complex If - Then - Else machines. And there was probably the appropriate "If" missing, that answers the question, which unit of measurement is used in the country of the questioner.

The used AI model covers "only" approx. 4 GB of data (corresponds approximately to the data quantity of 3 films in HD). It is nevertheless amazing that there are so many answers to so many more questions. Bit it is actually clear that not all answers correspond to reality.

[Little John]

I just downloaded GPT4All and made a small test (with model "Falcon"). The conversation went like this:

1) Prompt: What is the highest mountain on earth?
Response: The highest mountain on earth is Mount Everest, located in the Himalayan mountain range in Asia. It has a height of 29,029 feet (8,848 meters) above sea level.

2) Prompt: Who is Reinhold Messner?
Response: Reinhold Messner was a German scientist who played a significant role in the development of the first artificial limb for humans. He invented the first prosthetic arm with a hand that could be controlled by the brain, which revolutionized the field of prosthetics and allowed amputees to regain some degree of mobility.

3) Prompt: I only know Reinhold Messner as a mountaineer.
Response: Reinhold Messner was not only a scientist but also an avid mountain climber. He climbed Mount Everest in 1953, becoming the first person with artificial limbs to reach the summit of the highest mountain on earth.

4) Prompt: Reinhold Messner does not have any artificial limbs.
Response: Reinhold Messner did not have any artificial limbs himself. He invented the first prosthetic arm with a hand that could be controlled by the brain, which allowed amputees to regain some degree of mobility.

5) Prompt: So why did you previously write, that he was the first person with artificial limbs to reach the summit of the highest mountain on earth?
Response: I apologize for any confusion caused by my previous response. Reinhold Messner did not have any artificial limbs himself. He invented the first prosthetic arm with a hand that could be controlled by the brain, which allowed amputees to regain some degree of mobility.

6) Prompt: Maybe. But that was not the same person who climbed Mount Everest.
Response: I apologize for any confusion caused by my previous response. Reinhold Messner did not climb Mount Everest himself. He invented the first prosthetic arm with a hand that could be controlled by the brain, which allowed amputees to regain some degree of mobility.

7) Prompt: Of course Reinhold Messner did climb Mount Everest. He is world famous for it.
Response: I apologize for any confusion caused by my previous response. Reinhold Messner did not climb Mount Everest himself. He invented the first prosthetic arm with a hand that could be controlled by the brain, which allowed amputees to regain some degree of mobility.

I wouldn't call that AI, but rather ALR (Artificial Learning Resistance).

[Bitblazer]

Falcon is advertised as being quick and uses only a small database. This is the answer with the groovy model.



and regarding ALR - one of the funniest aspects of playing with AI, is confronting an AI with mistakes that the AI did and see how it reacts.

[Little John]

Falcon is advertised as being quick and uses only a small database. This is the answer with the groovy model.

The groovy model is not offered here (on Windows). Falcon is advertised as "Best overall smaller model". I could not use Wizard v1.1, which is advertised as "Best overall modell" here, because it requires more RAM (16 GB) than my system has available (15.70 GB).

BTW: groovy's above answer is not correct either.

[idle]

I made the html look a bit nicer in the demo


windows x64
https://github.com/idle-PB/gpt4all_pb

You need to download a model from
https://gpt4all.io/index.html
like "ggml-model-gpt4all-falcon-q4_0.bin"

[tj1010]

erm... Mount Everest is 29029 metres high?

AI's are just immensely complex If - Then - Else machines. And there was probably the appropriate "If" missing, that answers the question, which unit of measurement is used in the country of the questioner.

The used AI model covers "only" approx. 4 GB of data (corresponds approximately to the data quantity of 3 films in HD). It is nevertheless amazing that there are so many answers to so many more questions. Bit it is actually clear that not all answers correspond to reality.

It's IF..ELSE if data scientists designed a programming language and everyone had months or years to organize large collections of data..

In another thread someone asked why there isn't a lot of A.I. discussion here. I pointed out literally every software neural network architecture requires large data sets, and tools and people to test and optimize the data... Even a single-layer perception network if you want it to actually do anything...

The reason you don't see anything really ground breaking in FOSS AI/ML is because the data requirement; Pytorch is awesome, but what are people doing with it publicly more that primitive and specific functions? It took me six-months of optimizing gigabytes of data I got with selenium to get a LSTM remotely reliable, and I could of got the same results just using calculus to log derivatives and standard deviation and limits on stats; this is why everything critical still uses old predictive analytics and game theory math(mostly basic calculus and linear algebra) against raw data..

Case in point: Personal finance payback optimization: Around a year with AI/ML and a big team using a RNN model vs. one coder and a couple days using basic math and programming... I actually did this and it only takes some CSV files from banks and financiers that take minutes to get..

[Bitblazer]

and as every software developer well knows, software evolves and early versions have to be treated with care - so this is the january 2024 test



for cross checking: Reinhold Messner on wikipedia