There is much fanfare about artificial intelligence (AI). It is often hard to decipher some of the hype, or understand its abilities and limitations.
This thread has been set up to provide information about AI - is it reliable or is it driven by hype?
You are welcome to make posts which provide evidence to whether AI is reliable or hype, good/bad/indifferent.
Having a business in the hospitality industry, we are often advised that AI is the future of travel. Travellers will use AI to create the perfect customised holiday, such as asking AI to:
While AI might seem to be a great solution to save time doing oneās own research to develop an itinerary, using AI comes with major pitfalls:
In Australia the risks of relying in AI for ones travels, particularly off the ābeaten trackā, could end in tragedy. One might end up in a place they shouldnāt be or arrive somewhere which is vastly different to the information presented by AI. There have been reports of digital maps sending one in the wrong direction/up the garden path, but AI could take things to a new level.
AI is a data compiler and canāt think for itself. It also canāt differentiate between facts and fictionā¦or whether information available is correctly attached to the right location. This confusion can arise as there can be common location names around the world, which are vastly different to each other.
How about a spot of fishing? One catches a fish and checks local fishing regulations using AI:
AI is convenient, but, it canāt be used as a defense when someone gets a substantial fine.
I and my friends have used AI to assist in generating itineraries for travel. It is helpful, but it is also necessary to check the answer with other sources. For instance, it recommended that I stay in a town in Norway which only has one hotel and which got bad reviews. Another town a few km further had a much better option.
As you point out, AI canāt āthinkā, it merely regenerates data that it has ingested and creates convincing English (or whatever) sentences. Hallucinations are caused by faults in the input data. Garbage in, garbage out.
In my opinion this is a false dichotomy. In reality, the two questions are mostly independent.
I donāt think there is actually a lot of hype coming from the companies in question. There is hype in the mainstream media and in social media, and the like - and perhaps the companies are happy to sit back and let that happen. (Thereās some inflation in applicable company valuations but only time will tell whether that is hype or not.)
Results are most certainly not 100% reliable. So always verify. (The risk is that over time, people start to trust rather than verify.)
(Adding: Some people are legitimately using AI to deal with the problem of not knowing what you donāt know. That is, AIās answer will at least tell you what to search for using a traditional search engine.)
I think AI is least reliable when there isnāt much training data to provide a result either way and AI is inclined to just āmake shit upā rather than answer āI donāt knowā (a bit like government, really 
).
Unfortunately you wonāt have much visibility over the quantity of the applicable training data, although you can attempt to interrogate the AI regarding the strength of its āconvictionā in the answer, via follow-up questions.
There is a small risk in the IT field that AI will replicate an intentionally malicious or facetious answer e.g. there are many jokes on the web along the lines of āQ: What is the Linux shell command to do XYZ? A: rm -rf /ā and I am not confident that AI will āunderstandā the context. (That command will actually delete all files on the entire file system. Do not try at home. And if you actually do want to delete all files, it is a silly way of doing it. And, no, this is not the funniest joke ever but you know geeks.)
In many fields of endeavour, particularly those where the user doesnāt have any expertise, AI can already produce a better result than the user could produce, even if the result is not 100% reliable.
Then thereās AI in the context of generative AI in a creative field. What does reliable even mean then? Users have occasionally commented that depicted people ended up with six fingers. I donāt know whether thatās been fixed yet. In this context, there are more ethical questions raised than reliability questions. In fact, there are lots of ethical questions in other AI contexts too, but you didnāt specifically raise that in the original post.
Just to throw a couple of recent MSM articles into the mix: Robin Williams's daughter begs fans to stop making AI content of her father - ABC News and AI is creeping into our court system. Will it undermine justice? - ABC News
The sheer number of articles just in that one media outlet (the ABC) alone gives some indication of the level of hype.
I just did an experiment with Gemini A I. I requested a Love song and advised I lived in Werribee .
Thatās wonderful! Letās write a love song with some local flavour, grounding the lyrics in the places and feelings familiar to someone in Werribee.
(Verse 1) We met by the water, not the bay, but the river Where the eucalyptus stands, and the cold wind can shiver You were watching the swans glide, perfectly still and serene Like a moment of calm in a world that was too keen I didnāt know then that my life was about to turn around But your smile was the sweetest sound I had found.
(Chorus) Youāre the sunset that paints the Mansion gold, every night You are the steady warmth, my favourite, familiar light We donāt need the city to make our story complete We have the long roads of Hoppers and our own two beating feet Youāre the corner of Synnot Street and me, true and defined The most beautiful landmark my heart could ever find.
(Verse 2) I love the routine weāve built, the comfort and the grace A quick coffee run to our favourite, quiet place Watching the lights change along the Princes Highway Knowing that youāre waiting for me at the end of the day Weāve walked a million miles through Eagle Stadiumās grounds But the only score that matters is the silence when youāre around.
(Chorus) Youāre the sunset that paints the Mansion gold, every night You are the steady warmth, my favourite, familiar light We donāt need the city to make our story complete We have the long roads of Hoppers and our own two beating feet Youāre the corner of Synnot Street and me, true and defined The most beautiful landmark my heart could ever find.
(Bridge) We could drive to the Beach and watch the ships go past But even the horizon moves a little too fast What we have right here, in the heart of the west Is a quiet kind of magic, a love thatās truly blessed Youāre more than a view, youāre the steady foundation.
(Guitar Solo / Instrumental Break - Simple, warm, with a feeling of cruising a familiar road)
(Chorus) Youāre the sunset that paints the Mansion gold, every night You are the steady warmth, my favourite, familiar light We donāt need the city to make our story complete We have the long roads of Hoppers and our own two beating feet Youāre the corner of Synnot Street and me, true and defined The most beautiful landmark my heart could ever find.
(Outro) My beautiful landmarkā¦ Right here in Werribeeā¦ (Fade out)
The punctuation went hay wire in places . It did not copy and paste well .
There is a recent article from OpenAI that discusses the main reasons that Large Language Models make errors (Kalai Why Language Models Hallucinate arXiv 2025). The direct arXiv abstract page is at https://arxiv.org/abs/2509.04664, with PDF and HTML versions accessible for free.
It is a long and detailed article, but I found this relatively short video, which summarises the content well, at https://youtu.be/NqYqQ3Y0fHE.
Just a point that might be useful here.
Quite a few people in āmy age groupā are confused and wary of āAIā and itās influence on the outcome of searches made by Google in particular.
We all know the innaccuracies and frustrations when google guesses the answer to quite factual search questions.
So by adding ā-aiā (without the quotes) after the seach string, Google wont use AI and you will get the answers in a way people are more accustomed to.
How long before Google removes that option though?
I am trying to spread the word, so more people use it.
And, the more people realise where this is going, the more people will yell out when the option is removed.
Opinion: AI is dangerous. If the āgodfatherā of AI, Geoffrey Hinton, is warning people about it, listen to him. Iāve watched a few YT videos, some of which are entertaining but some of which scare me to heck. I have been known to use ChatGPT when it first came out. I still have Perplexity on my phone but not for long, and for answering questions, I prefer Lumo (Proton product). In the end we canāt avoid it, big business loves it and in the end, it will be our undoing.
Awesome tip - thank you!
A very handy option these days! Note that it works in some other search engines thatāll otherwise add an AI summary ā eg, DuckDuckGo.
There are quite a few other controls you can apply to Google searches, and some of these work in other search engines as well. Here are a couple I use often:
exclude certain words or phrases from the search by adding "-ā (without the quotes) followed immediately by the string to include (in quotes if it contains more than one word), eg -āartificial intelligenceā could exclude many articles about AI.
limit a search to a particular web address ādomainā by adding site:<domain> ā for example,
site:choice.com.au would limit a search to the CHOICE website.
Such controls can really help reduce the irrelevant search results.
Hereās an article with a lot of detail about Google Search controls. The first section, Essential Google Search Commands, is a good quick reference.
AI is being promoted for more than just searching and learning.
Beyond Hype?
āAIā has the problem that it can only generate content from what it has ālearntā and what it can guess.
Great, hence why AI is very general in its content. But it gains leaning data from the internet. Which is very general in its content, so it learns to be general. A viscous circle.
Added to this is that most information available is targeted. And to reach the widest audience, you need to appeal to the LCD. The lowest common denominator. Anyone make an educated guess as to where this is going?
āAIā more A than Iā¦
Just look at FaKeBooK!!!
It is easy to dismiss any ChatGPT/AI etc in any form as useful I think. Given AI uses fully or in-part information already online in many forms and places, and given most of what is personally and often professionally(News com? Pages for eg) posted is absolute Fake Click Bait Rubbish, the answers AI give are not trustworthy in the slightest.
I use AI daily for work & some personal use as well. The field of AI & ML various uses is so wide that it is hard to give just a blanket opinion. It depends on the use case at hand as to what I think. Examples: medicine: absolutely astounding potential to diagnose beyond what we can do ourselves within certain area eg: finding the primary cancer origin from a secondary cancer finding;
cars: (fully automated): has potential but very damn dangerous at this stage: case in point read the last 2 or 3 paras: Tesla Full Self Driving Is Terrifying & Exciting
my work as a web server administrator & x Developer: great time saver but use with caution, never on PROD systems & always test & verify responses; I am a WIN background now working with Linux; great to get scripts &/or commands started - give me the basics & I go from there or back & forward to fine tune a script; find bugs; etc too long & involved to fully explain
leisure travel & product research: depends .. can help narrow the field but can go off track, in circles & just plain be wrong!
Overall, I spend probably 80-90% of my interactions telling it is wrong, telling it is contradicting itself &/or getting it back on trackā¦. 
Am i concerned or excitedā¦ both .. I think above shows some points as to why excited; wary & concerned because whether we as a Country & Govt have ethics & use cuation in development, others donāt or not to the same level. I do have some doomsday scenarios that bounce around from time to time. It would only take one or two key mistakesā¦
If you have the time & inclination & havenāt heard of it - Google āThe paperclip theoryā
Also referred to as the Paperclip Maximiser. Sometimes better described as a philosophical proposition.
I asked Google AI a different question. The response.
Self Preservation?
AI can be very dumb and finally understood the question when asked - āis the paperclip theory flawed?ā My red marker, the response was more extensive than what Iāve shared here.
A related circle is that there is no guaranteed or general indication on the internet when content is from AI. So, increasingly, AI will be trained on content that was originally generated by AI. You can see where that is going.
I havenāt done enough research into the idea it is learning from general readily available source to give references here but I believe, open AI like ChatGPT, Copilot, Gemini etc yes, but again I think context is important. AI in specific fields is learning from data within that field & in those cases technical, scientific backed data.
Certainly in terms of my work, depending on the task I am working on itās responses can be very specific & very technical (certainly much more so than Googles ghastly algorithms of the last 5 or 10 years) & many times the basis that āweā then work on with different tangents to refine to get the desired result. BUT as said, check, verify & test!
Aside from that in my post above, āI spend probably 80-90% of my interactions telling it is wrongāā¦ a key reason I do correct it, is to improve it for myself & others.
