Example Dialogs of Serge - a LLaMa based Chat Bot

He seems to have signs of common sence

Also he has some skills in generating marketing content

As I am mostly uninspired for writing song texts, I asked him to write a text for me - so I can focus on the other tasks

listen to the resulting song here:

His logic / math skills though seem to show some weaknesses (unless AI ages different from humans)

His programming skills are even worse. Not allone that, he lies to me! The result from the program is actually around 3.9, no way that's "rounding errors".

When you call him out for an answer that is obviously wrong, he corrects himself

Be aware you are not talking to a real human, this is what could happen if you forget...

He even does not bother about criminal intent (I like the subtle irony in the response)

No, he will not sell his grandmother, but...

If you do not use the web interface you can drive the model directly which enables more complex tasks like letting LLaMA write a complete book for you

It just takes a bit of structuring and pandoc, latex and linux shell magic to generate PDFs that appear as propper books. Here is an example of a completely AI written book. And here you can find out what queries where done in order to create the book. Even most of that queries where generated automatically by the structure programm. The code of the structure program is now also included in the document showing the queries.

After simplifiyng the structure code in a generic way you can even fire up a whole series of (small) books that are equal in structure but can be very different in content in a single weekend.

Here are all my example "books" including queries and structure codes. By comparing them (especially the last 5 that were done after generification) you can see, that there is nearly no change in the structure code necessary. In most cases just interchanging the "TECHNOLOGY" terms did everything.

• About Programming, Game Theory, Numbers and AI (Queries)
• About Interdisciplinary Research and Intercultural Cooperation (Queries)
• About Cats and Dogs (Queries)
• About Flowers and Bees (Queries)
• About Beer and Chips (Queries)
• About Sex, Drugs and Rock'n'Roll (Queries)

While the topic of the first and second book matches up very well with the structure code (because the combination of terms overall is sane) the latter four book topics looked like they would force the AI into struggle. But as you can see, it will write about anything you instruct it to. I laughed a lot about what it finally came up with. It was far beyond my expectations.

There are some really nice cherries like: "In mathematics, sex refers to the process of finding solutions to equations or systems of equations." which sounds appropriate.

Or did you know about our "PowerPets" Project? No? Well here is the news:
"The ACS has been conducting extensive studies on using animals as renewables sources since 2013 with their 'PowerPets' project. This includes investigations into how different types of pets generate electricity through various forms of exercise such as running wheel toys, treadmills, swimming pools or even just playing fetch. Additionally they have also conducted research into harnesses that could be used by cows to produce power from walking around fields."

There is so much we can learn from LLaMA 7B. In her book "About Sex, Drugs and Rock'n'Roll" she shows us new ways of how to dramatically reduce our institutes power comsumption:
"Sex, drugs and rock'n'roll are often associated with partying and wild nights out on the town. However, these activities have a more practical application in terms of improving the efficiency of power distribution networks. Studies have shown that when people are given the opportunity to party or take drugs, they tend to be less productive at work. This means that there is less energy being used for productive purposes and more for leisurely activities. By providing employees with an incentive to party or take drugs, companies can reduce their overall energy consumption."

What is all this good for?

As you might have noticed in the last section, results from the Language Model are not always (or maybe even never) appropriate for scientific writing. But never the less can be of value for your writing process. Of course the Language Model is not intended for doing your work writing research papers. But as we all know there is often something called "writer's block". It's like you looking at an empty paper and suddenly you realize there is nothing, just like that's in your mind: emptyness.

Now imagine you can generate a raw structure, with some content that closely relates to what you want to write. But it may be complete nonsense. However, in that moment there is no longer any emptyness. You say: "Hey, that's wrong! I know better!". And there you are. Or you say: "Hey that's someting I never thought about! Let's check it's potential". Of course you have to really go for profound references or experiment yourself. But you would not be able to do so without being inspired in the first place.

At this point I hope you had fun with the collection on this web page and found something you could get out of it as a take away.

Have a nice and productive day,
Nicolas