Bizare 3rd person texts when generating articles

For some reason recently I very very frequently start getting these article that are written in 3rd person as if talking about what is going to be written in the article instead of the actual article like so, the following is a mild version of it but sometimes its just unreadable because every part does this weird “The article will be talking about blabla” over and over throughout the text basically in every paragraph:

Foundations of Evolutionary Computation in Engineering

The segment on the foundations of evolutionary computation in engineering delves into the core principles and methodologies

that underpin the application of evolutionary algorithms in engineering systems. This part of the article elucidates how the conceptual frameworks derived from biological evolution are transposed into computational algorithms to solve complex engineering problems. It is a journey that begins with the fundamental concepts and traverses through the intricacies of applying these principles to practical engineering scenarios.

This exploration starts with a deep dive into genetic programming, a type of evolutionary algorithm where computer programs are evolved to solve, or approximately solve, problems. Genetic programming serves as a primary tool in evolving complex systems, enabling engineers to automate the design process in ways previously unattainable.

The article then shifts focus

to evolutionary dynamics, discussing how the principles of mutation, selection, and inheritance are integrated into engineering algorithms to iteratively improve and optimize system designs.

The narrative also addresses

multi-objective optimization, emphasizing how evolutionary computation can tackle problems with multiple competing objectives, a common scenario in engineering design. This approach allows for the exploration of a wide range of possible solutions, each balancing the trade-offs between different objectives in unique ways.

The concept of differential evolution, a method that optimizes a problem by iteratively improving candidate solutions concerning a given measure of quality

, is then introduced

. Differential evolution exemplifies the versatility and robustness of evolutionary algorithms in solving complex, non-linear, and multi-dimensional problems often encountered in engineering.

As the section progresses,

the role of bio-inspired computing is highlighted, showcasing how strategies and processes observed in nature are mimicked in computational algorithms to solve complex engineering problems. This approach extends beyond mere optimization; it’s about creating systems that can adapt, learn, and evolve within their operational environments.

In synthesizing these foundational concepts,

the article lays the groundwork

for understanding how the principles of evolutionary computation are not just theoretical constructs but practical tools that can drive significant advancements in engineering. This section sets the stage for deeper explorations into specific applications and innovations in the field, illustrating how evolutionary computation is not only reshaping engineering practices but also redefining the boundaries of what can be achieved in design and optimization.

Even though it is often surprising to get very different results from one day to the next it is also possible that some tweaks to your prompt can lead to more stable results.

What is the current version of the prompt and what have your tried so far?

This is a recent development I have been using this same prompt for varieties of articles for 2-3 months, this comes with a variety of new bizarre behaviour like being unwilling to create code “because I am unable to”, or “because I cannot execute it in this environment” along with getting triggered saying “I am not able to generate that” where a simple “why?” makes it run again.

Usually happens after the first starting the second or third article in the chat. This did not happen before, this doesn’t happen really if I continually start new chats (as far as I have been able to tell).

1 Like

Here, this is from the docs:

It implies that positive or negative results may be random if the sample size is small. And from personal experience I have been there as well.

Like I said before, if you are willing to share your prompt the community can take a look and find out if there are some low hanging fruit to improve on the output.

It is not the prompt as stated and I will not be sharing it, in API the prompt performs 5 x better (as it has better understanding and capability I have seen so far) in ChatGPT it happens after subsequent article generations meaning it is deriving that it should be 3rd person from previous interactions, this is not behaviour this prompt has displayed in the past for the months that I have been using it.

As an FYI similar behaviour can be observed in different prompts with multiple articles and generation, where it starts to refuse formatting in the prompt and defaults and does not deviate from formatting it introduced itself going through the conversation even though the prompt explicitly mentions not to do things it does.

All my stated behaviour is observable in many of the chats I have (and I go through a lot of them in a given week)

After just starting to open new chats for a single article, I can say it is now also doing this 1 or two times out of 10 chats without any previous conversation. so it’s one of the parameters that creates variations in chats that is causing this