The Evolution of Science Fiction discussion

Since AI is in the news often lately, do you tend to see AI (in its various forms) as having a more positive or negative impact on our lives in general? We don't know what the future holds but you may have impressions based on what you have heard or your own experience!

Personal experience. This morning I asked 5 AI's the following question: "Which days of the week have a d in them?"

Answer from Claude Haiku 4.5
The days of the week with a "d" in them are:
Monday, Wednesday, Friday, Saturday, Sunday.
That's five out of seven days. The only days without a "d" are Tuesday and Thursday.

Answer from GPT-5 mini:
Days of the week that contain the letter "d": Wednesday, Friday

Answer from Llama 4 scout:
The days of the week that have a "d" in them are:
Wednesday, Monday, Saturday

Answer from Mistral Small 4:
All seven days of the week contain the letter "d".

Answer from gpt-oss 120B:
Monday, Wednesday, Thursday, Saturday, and Sunday each contain the letter “d.”

I don’t know much about it, but I think it is probably overhyped in various ways. I’m aware of negative effects, such as increased bullshit and fraud. As if we didn’t have more than enough already.

I tend to avoid it whenever possible.

I avoid it too. As much as I avoid social media. Though I fully understand that it’s coming and can not be stopped.

Ed wrote: "Personal experience. This morning I asked 5 AI's the following question: "Which days of the week have a d in them?"

Answer from Claude Haiku 4.5
The days of the week with a "d" in them are:
Monday..."

Wow. I can see in the day tot day practice that there are faults in AI results. But this is far beyond.

It is though often a case of defining the right question. I use AI in work and in private, but still am searching for the really good functions. People at my work nowadays use it often to make texts, but I still think I'm better off doing it myself :-). I used it a few weeks ago to find te best places for us to visit on holiday. I gave the country we want to visit (nature parks), the kind of things we liked to see, and some aspects we would like and not like. I got a great result, so it was really a piece of cake for us to choose. Before that it was a struggle combining all the info from the internet, but AI used the aspects we like/dislike and came up with results in a second.
So most of the time, it is a time saver for me.

What really scares me, are self driving cars.

Leo wrote: "It is though often a case of defining the right question..."

This is absolutely the key. As humans, we are used to people extrapolating an understanding from things we say that are not always clear. AI requires you to be clear in asking what you want. I've tried to use it for spreadsheet work & have gotten all sorts of crazy responses. It makes it difficult to find practical uses for it.

One of my hobbies is fantasy sports. If you're not familiar, you essentially act as General Manager of a team you build from scratch, changing lineups, making trades & free agent pickups to fill holes caused by injuries or poor performance. I've played fantasy hockey for about 20 years, the last 10 with a 20-person international group. We have become friends and chat year-round about almost anything. When it comes to making trades, it is sometimes difficult to evaluate - "should I trade this player for that player?". I was talking to one of our team owners from Norway, and he referenced using AI to evaluate trades, so I started applying that myself. It has been VERY useful for that.

Ed wrote: "Personal experience. This morning I asked 5 AI's the following question: ..."

To be more precise, I asked those questions to several LLMs (Large Language Models). Those should not really be called AIs. They may someday be extended into, or incorporated into, some system that is truly an AI, but LLMs are more like text prediction machines.

LLMs can be useful, especially in cases where they have been fed a lot of text relevant to the question that you are asking. They can help write a cover letter, because they've seen lots of cover letters. They can give sightseeing tips because they've seen lots of text where people talk about that topic.

But they are really, shockingly, bad at some topics, like counting and spelling, and can be easily tricked by asking a question that is similar to a common question but with one word or syllable changed, such as "Is it true that the Great Wall of China is the only man-made structure visible from Spain?"

(I got this particular example from a book I'm reading now called More Everything Forever: AI Overlords, Space Empires, and Silicon Valley's Crusade to Control the Fate of Humanity, which isn't so much about AI as it is about what the people working on AI think about AI. Their opinions are often bizarre.)

When asked the question above, the LLM correctly said that you can't see the Great Wall of China from Spain, but gave wrong reasoning for that, and said that the only man-made object visible from Spain is the rock of Gibraltar. (I'm not quoting it's exact reply, because I'm too lazy to type it. But it was funny. When I tried it myself, I got different, but also odd results. One LLM pointed out to me that China and the Moon are on different continents, although I hadn't even asked about the Moon.)

About a year ago I needed help with a crossword puzzle, so I asked Google "Which states have 7 letters?"

In the past, Google had been good at linking to human-made lists. But now they've improved it to automatically use LLMs whether you want that or not. It gave me wrong answers. And different wrong answers when tried from a different computer.

Tried explicitly asking Claude Haiku 4.5 today. It looks like it is getting a bit better. It realizes that it needs to double-check it's first guesses. But the answers are still a bit off. See if you can catch the errors:

Here are the U.S. states with exactly 7 letters in their name:

Alabama
Arizona
Florida
Georgia
Indiana
Iowa (actually 4 letters — not included)
Kansas
Kentucky
Louisiana
Michigan
Minnesota
Missouri
Montana
Nebraska
Nevada (actually 6 letters — not included)
New York (two words, doesn't count)
Ohio (4 letters — not included)
Oklahoma
Oregon
Texas (5 letters — not included)
Vermont
Virginia
Wyoming

Let me recount carefully — the ones with exactly 7 letters are:

Alabama
Arizona
Florida
Georgia
Indiana
Kansas
Kentucky
Louisiana
Michigan
Minnesota
Missouri
Montana
Nebraska
Oklahoma
Oregon
Vermont
Virginia
Wyoming

That's 18 states with 7-letter names.

Note, the text saying "(5 letters -- not included)" and so forth was part of the LLM answer. I didn't edit anything.

Ed wrote: "About a year ago I needed help with a crossword puzzle, so I asked Google "Which states have 7 letters?"

In the past, Google had been good at linking to human-made lists. But now they've improved ..."

It's interesting, I tried the same query in Google Gemini and also got Iowa as a mistake and got New York with the caveat "(excluding the space)" , which I count as an error since I asked for letters and not letters plus whitespace.

When I change the question to "Which countries have 7 letters?", I got the following answer:

The 34 countries with exactly 7 letters are:

Albania

Algeria

Andorra

Armenia

Austria

Bahamas

Bahrain

Belarus

Belgium

Bolivia

Burundi

Cambodia

Comoros

Croatia

Denmark

Ecuador

Eritrea

Estonia

Finland

Georgia

Germany

Grenada

Hungary

Iceland

Ireland

Jamaica

Lebanon

Lesotho

Liberia

Moldova

Morocco

Namibia

Nigeria

Romania

The error in this answer is Cambodia, 8 letters, and Ukraine is missing (and maybe others).

When I asked "why the mistakes?", I was told "I missed Ukraine ($U-k-r-a-i-n-e$) for the exact same reason I failed on the states earlier: I didn't systematically verify every entry against a master list. By missing Ukraine and miscounting Cambodia, my previous list of 34 was doubly flawed."

Gemini told me there are 40 countries with 7 letters. However, a few seconds later, it deleted the answer. I tried a second time, and the same thing happened. (It may be unhappy that I accessed it through the privacy-centered DuckDuckGo browser which may interfere with its cookies in some way.)

A third time, it game me a list including all of those above, plus Senegal, Uruguay, and Vanuatu, helpfully sorted by continent.

AI, at least now is very untrustworthy. However, if directed, it can be of great help. For example, right now I enjoy some genealogical research and it helps with OCRing written documents - they have to be checked afterwards, but it is still way faster than without it. It improved enormously since the launch in 2022 IIRC

AI is nothing new. It has been around at least 45 years. It was not what it is now, but it has been around. Do I expect to the world to change? Yes. But then the world changed with the advent of Robotics, PCs, GPS, and so forth. I expect some to lose jobs and their place in the world due to the inability to change with the times. I also expect plenty of jobs to be created due to the new world. This has always been with change.

Artificial intelligence is a great tool. But it’s like learning how to use a chainsaw while juggling three that are running.

I got an e-mail from a friend and g-mail's AI suggested a reply. I modified the reply and so it asks me why. Was it to polish, formalize, shorten, or make more friendly, its suggested text?

It doesn't offer me the correct response: the suggestion was simply wrong.

I read an article about AI hallucinations (where AI produces made up or false answers). Since then, I have felt more wary

Natalie wrote: "I read an article about AI hallucinations (where AI produces made up or false answers). Since then, I have felt more wary"

AI is not perfect and it is programmed by humans. It is going through a process of becoming better, but it is not there yet. AI does not think nor reason. so the mistake are goign to be programming in nature and that will get better as ll things do with time.

Helpful points, Papaphilly, Leo, Allan, Oleksandr, and Ed. And nice analogy, Thomas! I think every comment in this thread offers a good insight

Yesterday I asked google for "us weather map". Back in the old days, that would give me links to weather maps for the United States. But now that AI is involved, it decided I wanted a weather map for the city "Us" which is a tiny city 45 km north of Paris, France. It is only 11 square kilometers, so I'm not sure I would even need a weather map of "Us" even if I lived there.