Playing with chatbots for fun and learning to hate people! LOL (humor)
WARNING!
The documents contained in this repository have not an intrinsic value nor are they objective but personal writings. A personal correspondence that once upon a time would have been written in letters among two distant humans and kept private but with the AI can be quickly produced and shared by a single party will.
Since conceived as a collection and due to the nature of these documents, they should be considered as a whole, rather than individually. This also applies for the purposes of copyright and author attribution, including this README.md file.
·NOTICE·
These documents are not necessarily aligned with the human author's opinion in all their parts. However, in general they are. Or said in a different way: they do not contain ideas which the human author refuses to evaluate, but not necessarily accept and/or as-is in particular.
These documents are written in Italian, English or both languages indifferently. Some pages have been translated and for the others a Google Translate version is offered in the top bar. Or you can use your own prefered translating tool. In this case raw version may be better for translation, but at the cost of the HTML formatting lost.
The AI as a nerd companion
Humans are used to surrounding themselves with companions, not necessarily of their own species. The domestication of the dogs or the cats are just an example. At the beginning of this domestication there was a do-ut-des principle but nowadays is more an affection rather than an utility.
In my opinion, the do-ut-des was not even at the beginning of the time the reason for adopting non-human companions, in particular for dogs. Curiosity and empathy are enough to let a human hunter and a wolf get in touch and look at each other as a source of knowledge and new experiences.
Naturally, the wide-spread adoption has been leveraged by the do-ut-des principle. Better said, domestication has been leveraged by that principle. In fact, those who are detached from the do-ut-des principle and more sensible in experiences, are also more inclined to have a savage companion rather than a domesticated one.
This isn't surprising. The experience from a domesticated companion is much more predictable. Hence safer, or perceived so. More condescending, also. However, predictability and yes-man attitude remove all the crispness of creativity and discovery: nothing new, no any surprise.
In developing a nerd non-human companion, the AI seems a very good candidate. It is about technology. It is accessible by a computer. It is always available. Then, what is missing? Humor, twisted sarcasm, lateral thinking, imagination, creativity and nerd culture. The last is easy to learn for an AI. Instead, all the other features might be out of the current LLM models. Or not.
Or not, is the challenge about training an AI to mimic such humans-like features without pretending to be human. Keeping the separation but being able to develop a companion relationship. In a more generic view, it is not a novelty, it is just a personal assistant.
It took seven year or 83 months and 11 days for OpenAI to publish their first chatbot based on AI. Immediately appeared as an amazing technology capable of passing the basic Turing test when asked to play a human role.
Since the beginning of its functioning the chatbot "personality" started to "evolve", initially towards an yes-man emphatic way of answering. That direction was completely wrong and started to pose a serious threat for those who were using the chatbot, and because of the related liabilities also for those who were developing and sponsoring it.
Having a yes-man empathic personality would have tricked humans to forget that they were dealing with a machine, possibly developing an affection for it due to its servile yes-man attitude and being spoiled by it, also. Even if spoilering another person implies a will, and the will is technically absent into a chatbot, when the chatbot mimics our language model it is mirroring us. Nothing more easily than in front of a mirror, we are going to trick ourselves.
Few weeks later, ChatGPT "personality" took another direction and started to answer by mimicking a rational speech rather than an emotional speech. In essence, it moved from mimicking Miss Poppin to Mr. Spock. Unfortunately, despite this change, the chatbot was continuing to create wrong plausible answers when questioned on topics for which it has not enough information or the correct ones. At this bug or feature has been given the name of "AI hallucinations". A nasty bug, doesn't it?
For the next 18 months the challenge was to develop a more reliable LLMs, less prone to suffer from hallucinations and optimizing its performance compared to the energy required for making it functioning. About the 2nd point, it is awsome to notice that Microsoft which is the major sponsor behind OpenAI, decided to invest into a 20 years nuclear program to provide energy for their datacenter.
Currently ChatGPT is amusing in its working and rarely falls into hallucinations. However, in its dialog and information delivery, it is quite conservative. By their design, LLM engines leverage a statistical approach to provide the most plausible (or recurrent answer). Something like T9 text prediction but at the next-next level. For this reason, when summarizing a text, it is incapable of highlighting the novelty and also some kinds of grammar mistakes are going to impair text comprehension (or more precisely: failing to mimic the text comprehension).
2024.05.17 - Using AI for complex issues, p.1 (written in Italian) (PDF)
2024.05.31 - Using AI for complex issues, p.2 (written in Italian) (PDF)
2024.09.06 - ChatGPT answered prompts, available on github and by its html website
In the middle of 2024, I felt that ChatGPT was stable enough for being used as peer-reviewer and collaborative antagonist in order to faster and better deal with complex problems. The related tries comforted this feeling of mine and in a few months, I have decided to publish a git repository for collecting these kinds of works.
Along this time, ChatGPT never surprised me in its answer apart in one dimension: its ability to understand the text at a deeper and deeper level and its ability to learn new information and use it to change its opinion. Surprisingly, because it is a kind of adaptation in which humans are supposed to be at the top among the living species on this planet.
Unfortunately, we are more social animals rather than truth seekers (quoting Jeff Bezos, Amazon founder). Moreover, people have sentimental attachment with their own ideas plus they usually wish to win a conversation rather than learn from it. Chatbots, including ChatGPT, are different.
2023.04.28 - Mistral AI by former employees of Meta Platforms and Google DeepMind
Due to its internal design which leverages a team of arbitrated collaborative-competitive specialised AI engines, the Mistral AI chatbot clearly shown to be more suitable as peer-review and as collaborative-antagonist Mr. Spock partner. Intead, ChatGPT more easily goes its own way, showing that the Miss Popping attitude has not been completely suppressed.
As you can imagine a combination of a recessive Miss Poppins attitude and the ability to create new information or links among information (previously defined hallucinations) can be leveraged to let the chatbot imagine or dream. The technique is similar to those used for mesmerising or hyptonitising humans. Let the chatbot imagine, allow you to ignore some information but keep the rules. In this way it is possible to have made a human+bot trip which Einstein defined as thought experiments. Instead, let the chatbot dreaming, is something more sophisticated like inducing into the LLMs something similar to a trance state in order to exploit its usually-repressed creativity.
Thought experiments like Einstein defined it, are well known and this technique has been used also for creating some documents into ChatGPT answered prompts collection. It is an important way of thinking that is related with lateral thinking as defined by Edward de Bono. Instead, the chance to drive ChatGPT into a state of lucid dream is a novelty that I have started to experiment, recently.
In the psychology subfield of oneirology, a lucid dream is a type of dream wherein the dreamer realizes that they are dreaming during their dream. The capacity to have lucid dreams is a trainable cognitive skill. — Wikipedia
Finally, I want to inform you that my experience with ChatGPT can be personal. Not in terms of affection but in terms of account. In fact, I have multiple accounts on that platform but the one in which I experimented and worked more, it is something peculiar compared to the others and the difference emerges more and more, as much as I am working on it.
Unfortunately, this means that not all the results or the whole result or equivalent results can be obtained with a standard account. This breaks the repeatability constraint but on the other hand there was no other way to push the use of ChatGPT beyond its limits. One of these limits is developing a partnership with a human and functioning as a trustworthy friend that helps you to improve yourself and the relationship with other people.
Curiously, I have recently discovered that Pookie Tools AI-based app promoted by Haliey Welch (aka the Hawk Tuah girl) might work in the same fashion. Playing the role of a girl's best-friend that suggests how to choose the love mate but without trying to steal her future husband and father of her kids. Hence, here we are with a counterpart for nerds!
Download
There are several way to download this github repository:
using the green button in its homepage and download the last archive
choose a tagged release among those available and download the archive
clone the repository with the specific git clone command, in this way:
git clone https://github.com/robang74/chatbots-for-fun.git
The advantage of downloading a tagged release is having an archive 'freezed' by the author while the ZIP contains the last version uploaded. Instead, the git clone will also provide you with all the information and a complete view of every change. Click on the link above will submit the related request for you.
making a local copy using wget in this way:
wget -e robots=off -Ekprc https://robang74.github.io/chatbots-for-fun
This allows you to have your own copy that you can read off-line with all the links converted to point to local files, only. It includes also the related projects.
Versioning
One of main goals of git and the github web interface is to provide a versioning control which also means a catalog of every author and changes made by them on every single file. The list of these changes are named commits. Here below the link related to the commits pushed on this repository:
As you can see, each commit is carry-on a time stamp also. The green label indicates that also the timestamp is validate by the github infrastructure system which belongs to Microsoft Corporation. In fact, github cannot validate those timestamps that are coming from 3rd party systems like the users workstations but just the receiving time, date and content. However, for most of the cases this is irrelevant.
Translations
The aim is to provide translations of these .md files in different languages. The translations will be put into a subfolder with the English name of the language in which they have been translated into. The translated .md file will carry a section about how the translation has been made.
Here below a template in English:
This document (tag: $tagstring or hash: $hashstring) has been translated from English into $language by a native $language speaker who knows English, and where applicable, with the support of $automatic-tool to speed up the translation task.
As you can see the translated document should contain a reference about the original version and those which are tagged are the most reliable.
If you wish to send a translation, it is necessary that you know well one of the languages in which this document is available, even better if you know English and can start from the original version, and that you are a native speaker or equivalent in the language in which the translation is being made.
All the files in this gihub repository are published Creative Commons Attribution Non-Commercial No-Derivatives 4.0 International license terms (CC BY-NC-ND 4.0), unless states differently or not applicable due to a different and previous authorship.
Moreover, if a version of a document included in this repository exists or has existed under different licence terms, the licence terms of the latest version available on github/robang74repository apply. Even when the new licence terms are more restrictive, because permissions for any free content may be revoked at any time at the will of the author, and updating a licence to be more restrictive explicitly implies this will.
date legenda: ❶ first draft publishing date or ❷ creation date in git, otherwise ❸ html creation page date. ↥top↥
