February, innit.

[Dave Miller]

Mmmm. I think the authors are pushing it a bit, there, Vv!

In most cases, there are two quite independent meanings and therefore a fascinating situation. Including (2), though, raised my eyebrows: surely “blue” as a depressed feeling is just a metaphorical use. Seeing red with anger and green with envy still need the explanatory extra words, and blue doesn’t, but the background meaning is still the colour.

(10) in the second list seems weak, too: a present (gift) is surely something which you present (bring forward). Noun and verb forms from the same root - and still bearing clear connection - are plentiful and not of particular interest. Better perhaps to have contrasted gift with current time?

[Little Jack Horner]

My son has recently introduced me to ChatGPT, a “platform” that demonstrates artificial intelligence and which is currently the subject of much discussion. I asked him to present a challenge to it. My challenge was: “Write a story about a sailing ship crashing onto rocks with just a few survivors”. Here is the story that ChatGPT produced “in a few seconds”. It may not be worth a Booker Prize but I was impressed. Perhaps it does not demonstrate genuine imagination but I think it is a good counterfeit.

The wind was howling and the waves were crashing against the hull of the sailing ship as it made its way through the stormy sea. The crew was on high alert, trying their best to steer the ship to safety. But despite their efforts, the ship was no match for the powerful storm.
Suddenly, there was a loud crash as the ship hit a group of rocks just off the coast. The impact was so great that the ship began to take on water rapidly. Panic set in among the crew as they scrambled to save themselves.
A few of the sailors managed to make it into a lifeboat and rowed away from the sinking ship. They watched in horror as their home for the past months went down with a final gurgle. The sailors were tossed around in the rough seas as they rowed, struggling to make it to shore.
When they finally reached the beach, they were exhausted and traumatized. They had lost nearly all of their crewmates and their ship. They were the only survivors of the tragic accident.
As they sat on the beach, looking out at the sea, they couldn't help but wonder what the future held for them. They knew they had to return to civilization and start over, but the memory of the ship crashing onto the rocks and the loss of their crewmates would haunt them forever.

[Verbivore]

I've long been a sceptic when it comes to 'news' reports (which is why I have about 20 'fact-check' websites bookmarked under News). Now, thanks to the emergence of AI-based programs such as ChatGPT, I am becoming even more sceptical.

I fail to be (yet) convinced that these programs will in any worthwhile manner benefit humankind.

[Dave Miller]

I think we are being distracted at the moment into examining how well the machines replicate how a human would write - just as, in the 50s and 60, all robots were depicted in human form, and assessed by how well they pretended to be human.

We nowadays (eg in the car industry) have robots which bear no resemblance at all and are all the more effective for that.

The AI will be useful when it can be told, in plain English, to do something useful (rather than to do something human). I’m not sure what that is, but examples might be writing computer code, setting machines to manufacture specific shapes, or carrying out research to answer questions?

[Verbivore]

Some relevant articles on AI / ChatGPT:

OpenAI’s ChatGPT has a major political bias, and it could ruin artificial intelligence

Exploring the potential for bias in ChatGPT

How OpenAI is trying to make ChatGPT safer and less biased

Buzzy ChatGPT chatbot is so error-prone that its maker just publicly promised to fix the tech’s ‘glaring and subtle biases’

OpenAI Chatbot Spits Out Biased Musings, Despite Guardrails

ChatGPT could be used for good, but like many other AI models, it’s rife with racist and discriminatory bias

And in response to Dave’s point about doing “something human”, here is an example of a clumsy attempt by ChatGPT to craft a poem (atrocious metre! – perhaps the programmer’s definition of poetry was merely something that contains line-end rhymes):

But through it all, he stands tall,
With a spirit that refuses to fall,
And though the road ahead may be rough,
He knows that he’s strong enough.

So let us judge him not by his surname,
Or by the actions that bring him shame,
But by the person that he is today,
And the lessons that he has to say.

Had I submitted that piece of drivel for a school exercise, I’d have scored a Fail.

[Verbivore]

Doctor Dolittle rises again?

[Verbivore]

I've been playing with ChatGPT.

It gets things wrong! (No surprises there.)

First, I asked it questions about a particular model of Mercedes-Benz. It told me that there was a V12 engine option. WRONG! There were only V8s in that series.

Then I asked who designed that series. Again, WRONG answer!

Next, I asked it about another series of Benzes; again, it got that WRONG on numerous details!

In the second round, I asked it about the village where I live. It mis-named four of the six streets (tiny settlement), so again WRONG! It also WRONGly gave the population as 200, when according to the 2021 census, the population was 80.

For the third round, I asked it about an AIDS-prevention program that I ran in the '80s and '90s. ChatGPT gave WRONG answers to who were the original staff, where it was based, and who founded it. In its emended response, it correctly named yours truly (a detail I had supplied when I called it out for the error) but included a statement from its previous wrong answer – a statement that could be regarded as libellous were I of a litigious mind.

In round 4, I asked it about a university where I once worked. It completely screwed up the institution's history, so again, multiple times, WRONG!

For each wrong answer, I challenged the thing. Each time it started its response with:

"I apologize for the mistake in my previous response. I was mistaken in stating … . Upon further research …" (Its 'further research' was likely gleaned from my response wherein I called bullshit.)

"You are correct. I apologize for the mistake in my previous response."

"I apologize for the mistake in my previous response. You are correct …"

At least it's programmed to apologise, but I doubt its sincerity (can machines be sincere?). Rather, I suspect it was using my 'corrections' to update its knowledge base.

ChatGPT has a long way to go before I'll be convinced to use it for anything serious.

[Little Jack Horner]

I think it needs to be remembered that ChatGPT is not a search engine, nor an alternative to Google or Wikipedia or printed books. Of course it makes mistakes. It claims to be artificial intelligence and there is no particular reason why it should be better than human intelligence. Humans make mistakes all the time and we need to apply the checks and analyses we would apply to any other source material. It is still work in progress. We need to be patient and wait for a few years to be sure whether it is useful and in what circumstances. In the meantime, I think its performance is astonishing. The international baccalaureate is allowing for its use, providing students cite properly in any material they produce in just the same way as they would for referencing printed material. 

At a discussion group I am attending this Sunday we are going to talk about assisted dying. A friend of mine in the group posted the following to ChatGPT:
I suggest we talk about the choice we should (or should not) get about when and how we die. Euthanasia, suicide, assisted death, mercy killing or any other form of intentional end of life is a difficult subject to think about and discuss, but a vital one nonetheless.

He received the following. I think it is amazing. The paragraph breaks are my own.  

Yes, the topic of end-of-life choices is certainly a sensitive and complex issue. It raises a variety of ethical, moral, legal, and religious questions that don't have easy answers. Nevertheless, it's a topic that needs to be discussed openly and honestly, especially as medical technology continues to advance, and people are living longer lives with more complex medical needs.

Euthanasia, suicide, assisted death, mercy killing, and other forms of intentional end-of-life choices all have their own sets of implications and consequences, both for the individual making the choice and for society as a whole.

Some people argue that everyone has the right to choose when and how they die, and that it is a fundamental human right to have control over one's own life and death. Others argue that these choices are morally wrong, and that intentionally ending a life is never justifiable.



There are also practical considerations to take into account. For example, if euthanasia or assisted death were legal, how would we ensure that the process is carried out safely and ethically? How would we prevent abuses, such as pressure on vulnerable people to end their lives prematurely? How would we ensure that people have access to palliative care and other forms of support to alleviate suffering and improve quality of life?



It's important to note that the availability and legality of end-of-life choices can vary greatly depending on the country and culture. Some countries, such as Belgium, the Netherlands, and Canada, have legalized euthanasia or assisted death under certain circumstances, while others, such as the United States, prohibit it under all circumstances.



Overall, it's clear that the topic of end-of-life choices is complex and multifaceted, and there are no easy answers. However, by discussing these issues openly and respectfully, we can hopefully find ways to ensure that everyone has access to compassionate care and support at the end of life, regardless of their personal beliefs or circumstances.

[Dave Miller]

That is, as you say, LJH, amazing.

It’s important to note, though, that the artificial intelligence is processing words and sentences that have already been produced by humans. On that subject, things have been well expressed by careful thinkers. When asked a different kind of question - such as those asked by Vv - the AI might need to trawl through snippets of exchange on social media groups or related forums, and they are not necessarily carefully (or knowledgeably) written.

The useful task the AI is performing is to do all that trawling. Whether what it finds is true is not up to the AI. It does the trawling much faster than we could, but if we found it slowly ourselves, we’d need to assess it for truth. We need to do the same thing for the AI’s output and must not be distracted by how well it is expressed.

[Dave]

Feb 21, 2023 21:15:30 GMT Little Jack Horner said:

My son has recently introduced me to ChatGPT, a “platform” that demonstrates artificial intelligence and which is currently the subject of much discussion. I asked him to present a challenge to it. My challenge was: “Write a story about a sailing ship crashing onto rocks with just a few survivors”. Here is the story that ChatGPT produced “in a few seconds”. It may not be worth a Booker Prize but I was impressed. Perhaps it does not demonstrate genuine imagination but I think it is a good counterfeit.

....

Is there repeatability? If you again or several others all submit the exact same request, will the output for each submission be identical?

[Verbivore]

Mar 10, 2023 2:03:16 GMT Dave said:

Feb 21, 2023 21:15:30 GMT Little Jack Horner said:

My son has recently introduced me to ChatGPT, a “platform” that demonstrates artificial intelligence and which is currently the subject of much discussion. I asked him to present a challenge to it. My challenge was: “Write a story about a sailing ship crashing onto rocks with just a few survivors”. Here is the story that ChatGPT produced “in a few seconds”. It may not be worth a Booker Prize but I was impressed. Perhaps it does not demonstrate genuine imagination but I think it is a good counterfeit.

....

Is there repeatability? If you again or several others all submit the exact same request, will the output for each submission be identical?

G'day, Dave!

According to my readings of the instructions, responses are tailored to the particular questioner. I suppose that means it must build a profile of each user.