The GenAI / LLM Dragon eating its own tail

The global surge of data generated by Artificial Generative Intelligence (AGI) or GenAI is massive. We are already witnessing a paradigm shift in content generation, especially in text and image generation. However, this GenAI has its own consequences.

Unnoticed Consequences of GenAI

This blog uncovers the most unnoticed consequences that a layman usually does not think of, specifically the originality or liveness of the generated content itself. We are currently in the first generation of GenAI, especially for LLM. Almost all the content generated by these systems is trained and learned from publicly available internet sources, which are the work of real humans and carry a human’s touch of creativity, innovation, or thought process.

As mentioned in the first paragraph, both GenAI-assisted and pure GenAI content is circulating back to the public internet at a faster rate due to shorter content creation time compared to content generated solely by humans. With the current pace of advancements and the cutoff of training data, the LLM will be trained on or is already training on content created by itself. This can lead to a normalization of uniqueness or liveliness of the content itself.

The Effects of GenAI impacting itself

• Generative Echo Chamber: This refers to the situation where LLMs trained on data generated by themselves echo the same nature of output, potentially amplifying polarized data in further generations of LLMs.
• Model Autophagy Disorder (MAD): This phenomenon reveals how unchecked AI risks become detached from human needs through recursive self-feeding.

So What’s Next:

There are research papers emerging to mitigate this consequences, as the GenAI realm is new, so are the challenges are new and unique, but the adoption rate of GenAI can outpace the mitigations.

So what to expect is there is always a chance of bias or echoing the same context in the generated outputs. Those who aware of this gray area in GenAI, will still uphold the real value of a human curated content.

In short, want to convey the effects of GenAI, which may not behave the way you think.