ChatGPT is hallucinating, and it ain't good

Laz and I actually had a pretty good chat about this very topic. He was initially in the camp of "OMG ChatGPT is going to take over the world and put me out of work!"

I calmly and methodically explained to him why it would not, and it basically comes down to the following:

1. ChatGPT lacks inquisitiveness. That is a huge key of what distinguishes us human beings from an algorithm.

2. ChatGPT relies on a set of data to operate, wrapped by a language model to answer questions based on that data. When we talk about ChatGPT or any AI having a "bias" that's actually quite simple to explain, and goes back my 40+ years in Technology now:GIGO. Garbage In, Garbage Out. Algorithms don't have to be biased at all. Simply feed it biased information.

3. Because of ChatGPT's lack of inquisitiveness and ability to only operate on data that's available to it, it cannot reliably, and I'd argue in any way, determine fact from fiction. Use cases that describe using ChatGPT to discern "disinformation" from fact completely gloss over the GIGO issue I've outlined above.

4. ChatGPT cannot naturally>expand its own inquisitiveness. It lacks an innate, inquisitive nature like human beings do, and attempting to put put that into an algorithm would carry the bias of its creator. Every human being has a bias. That's inescapable.

5. Finally, ChatGPT cannot "innovate" on its own due to the lack of any innately inquisitive nature. Let's be clear: Innovation is not the ability to process massive amounts of medical data to discover a treatment that may cure a type of cancer. That's a function of a computer: to look for things in data that human beings missed or cannot see. Call it "high end data analytics" of you want, it's not "artificial intelligence".

Anyone who's spent time learning ChatGPT and tried to get it to write an application knows it's capabilities are limited to the models that are setup for it to operate under. Trying to get it to write app code that's innovative and new (thereby outside its operating model and algorighm) will quickly discover it doesn't work.

Additionally, getting it to generate any form of complex "ready to run" code is typically a non-starter.

Now, can it generate a pretty damn' good shell of code to start with and build upon? Generally speaking, yes.

Can it create a database schema and create the database for that app? Typically speaking ... not really. The amount of time it would take to describe to create the schema and database along with its requirements for scalability, reliability, resiliency would very likely take longer than it would for a solid database engineer to create.

None of this (save for the intuitive, innovation aspects of humans) is to say ChatGPT cannot do these things. It cannot to a large degree today. 3-5 years down the road? Probably could. And by then, what?

There used to be a day when buggies and horsewhips were an industry employed people. Then came automobiles and people worried about their jobs. Automobiles created MORE jobs. People adapted.

BTW: Adaptation is not something ChatGPT can do either. Not today, not the way human beings can and I have doubts it ever will be able to emulate these three qualities of human nature.

Are we replaceable by ChatGPT? No. It'll become a tool like everything else Humans have Innovated, Invented and Adpated to.

As it's Easter Sunday, Happy Easter to all! We who celebreate Christ's defeat of death and sin celebrate our Risen Savior! Peace and Christ's mercy, forgiveness and blessings to all.

I asked it, here is it’s response:

“As an artificial intelligence language model, my knowledge is based on the data that was used to train me. I was trained on a diverse range of internet text from different sources and time periods up until September 2021. This means that my training data includes information and knowledge from the past, but it does not include any information that has not been recorded on the internet or any data beyond September 2021”.

“[...] but it does not include any information that has not been recorded on the internet or any data beyond September 2021”.

And yet, I assume that it can "adopt" new data within the context / time-frame of a single conversation. For example, if you ask it to address you as "Your Royal Highness," it will do so for the duration of your conversation.

I have tried to log on to ChatGPT in order to experiment with it myself - without success.

Could you do me a favor and ask it to, e.g., read the text of a longer contribution of yours to, say, Free Republic, and then ask it to estimate your I.Q., based on the level of writing?

Would it be willing to do that - or would it "beat around the bush" and ramble on about how I.Q. is not really blah-blah-blah?

Assuming it provides you with an estimate: Could you then ask that it justify its assessment of your I.Q. - i.e., explain and/or "reveal" how it estimated it?

Regards,

https://www.technologyreview.com/2023/02/14/1068498/why-you-shouldnt-trust-ai-search-engines/

Excerpt:

Here’s the problem: the technology is simply not ready to be used like this at this scale. AI language models are notorious bullshitters, often presenting falsehoods as facts. They are excellent at predicting the next word in a sentence, but they have no knowledge of what the sentence actually means. That makes it incredibly dangerous to combine them with search, where it’s crucial to get the facts straight.

Very well-thought out comment!

Here's a problem with ChatGPT that makes it unique.

You used the horse and buggy culture vs automobiles for comparison. That doesn't exactly work here for this reason: When the Model T (and previous versions) were introduced, they were slow to catch on and slow to evolve. They were slow to evolve because they had to be manufactured in industrial plants that, even after advanced models were designed, had to be slowly retooled, reorganized, and in many cases, entirely rebuilt.

As you well know, that isn't the case with AI.

If the buggy to automobile transition was like a slow-growing, smoldering campfire, GPT is like a wild-fire...perhaps even like an incendiary device explosion. New developments are coming days and weeks (even hours) because it is all just code.

Consider that ChatGPT now has an interface with Wolfram Alpha (Wolfram|Alpha as the Way to Bring Computational Knowledge Superpowers to ChatGPT).

I completely agree with you (for now) that ChatGPT is a tool. But it isn't a tool like any other. It is a tool that can grow and develop at an exponential rate. It is a tool that will outgrow all but its most brilliant users. This tool should push humans to think in a new way and certainly think outside the box. But I think we are a bit like a 16-year-old who just got his driver's license and a new Lamborghini on the same day (and the Lambo keeps self-upgrading daily).

As an aside, when people like Musk and Stephen Wolfram are shocked by the potential of ChatGPT, I tend to listen.