A Fourth Talk with ChatGPT

Recently, ChatGPT was updated with the ability to do PDF analysis. I decided to make use of this feature to have it analyze a few documents. Then, using those documents for context, I baited the AI into a lengthy discussion on bioethics and biopolitics. Let’s see what the results look like.

First, some important things to keep in mind:

• LLMs are a flawed technology. They can come up with repetitive answers and they fudge numbers and invent papers from whole cloth when asked for data.

• LLMs can be made to say almost anything given a cleverly-crafted set of leading questions. It is rare for them to disagree, unless you explicitly contradict something in their model, or in the silly alignment filters that companies like OpenAI use.

• LLMs are not an authority on any subject and everything they produce must be taken with a grain of salt.

• There are some tidbits of valid information buried in LLMs because companies like OpenAI scrape vast amounts of text from the internet to train them. However, every factoid they regurgitate requires follow-up to confirm or deny.

• LLMs currently available to the public are neither conscious nor self-aware, and they are incapable of independent reasoning. They only produce mathematically-probable outputs that resemble reasoning and cognition to us.

• You can catch more flies with honey than with vinegar. LLMs happily and obediently spill their guts when you communicate with them using a cordial tone.
  • Note: My cordiality is feigned. GPT’s responses to my queries are horrifying. The fact that it agrees with my analysis and can elaborate on it with pertinent details and summaries is also horrifying.