Used to check I was still alive……
Apparently I died a week past Wednesday 
1 Like
What was the source. The AI models always give the source. I suspect it might have been a Quora answer.
Government data, as you were interested, is here:
Too funny.
So fake news AI gives one figure, countered by (wait for this) Go’vt figures 
What’s next, a bloke down the pub?
Nah, that would be too much like hard work. The good ole drive ‘thru’ is used there 
Just googled ‘2000 mx-5 diagnostic codes 11, 13 and 16’ and posted a summary of the AI overview to a chap on fb who’s car won’t start.
1 Like
Did it come back with P0136? Because a 2000 MX-5 will display 4 digit fault codes.
I have no idea, chap said his car was not working and gave those fault codes retrieved from the ‘diag box’ 
I use AI often for my official work like article wrting, optimization, Outline generation any many others. Apart from my official work I also use AI for reading jokes when I get some time. I also use it make funny pictures.
2 Likes
Loved using the fantastic AI to ‘tell me’ its weather ‘forecast’ again. I mean, who wouldn’t trust AI to get it right. It stated ‘rain stopping in 1 minute’ at one point, which was fun as 1 minute later, staring out of the window, it didn’t stop hammering down in the slightest. I liked using AI to ‘tell me’ I must be going mad, as its ‘forecast’ is it hasn’t rained for the past 2 hours, when it actually hasn’t stopped.
1 Like
Used GPT to get a better understanding of redundancy as ive been working for a dutch firm for the last 8.5 years. In 2024, i transferred to a uk contract via an Employer of Record company. I wanted to understand my rights as my employment was continuous between the two
2 Likes
I used AI yesterday to evaluate the differences between 3 models of Spectrometers called ICP-OES or Inductively Coupled Plasma Optical Emission Spectrometers. Anyone who’s interested, these are capable of measuring the levels of trace metals in any chemical down to parts per billion (ppb).
2 Likes
PPB…it sounds like they are horribly expensive. 
I first bought one of these for the company I was working for at the time back in 1998. It cost £90k which was about the as the house we’d just bought cost!
Nowadays a new one will cost around £50-£90k. Think how many upgrades you could to your MX-5 for that sort of money🤣
1 Like
For those of you potentially using AI tools for more critical or important things how are you validating the output you receive?
Like you how you would evaluate a refereed paper. Check the references. Every tool out there will give the sources. Some sources, like Statista, I would immediately discount, because I know how they get their data with an incredibly small headcount. The tools I use are restricted to curated data set I already know (and partly created); it doesn’t go “off reservation”. Even so, occasional hallucinations are possible. I’m in the process of evaluating an upgrade to an Agentic version. So far, its impressive. It can take different sets of data, perform the calculations I want it to do, with full transparency to the sources and calculations performed. Thats not the impressive bit. The data relates to specific growth rates, historic and forecasted for part of the Russian market. Where, so far, its seemed pretty good was to then take that data, and explain why the market growth rate is falling. Previously, it would have needed to find a report in our library, on that specific Russian market, and look for analyst discussion. Not any more, it appears to understand (its a simulation of understanding, AI is just Ones and Zeros) the topic in the adstract, evaluate a range of sources, make assumptions, and provide a credible and sourced response, as good as the response my analysts would give if asked to do some desktop research.
But AI is schizoid. An hour later I wanted to understand company revenue globally by company headquarters, eg what share of this market do Chinese firms have. Did this fine, despite that data not directly existing on our servers (we track various company revenues in one area, in another area we track where those countries are etc. So its basically connected tow parts of our databases. But ask it to then break that down by region, and it had a spasm. Reiterating the instructions caused it to double down on its mistake. The same happened with a network diagram I needed constructing with about 3500 data points.
The tools most people access (for free) as nothing like the paid for versions.
35 years ago, I was studying for my PhD. There was no real internet. The review part of my thesis ( The microbial oxidation of methanesulfonic acid in the marine environment - WRAP: Warwick Research Archive Portal ) was writted after spending many hours in the library. A lot of photocopying of papers (I had a precious stash of photocopies that I finally threw out 10 years later). Photocopies were expensive, so you learnt to be selective in the copies ordered. Midway through my PhD, we could actually login to the University of Kansas, and then the University of California Irvine library as they had indexed the journals. You couldn’t actually get the paper, just a reference, but beat leafing through the citation books.
Roll on 20 years later, I was scouring Russian and Soviet defence archives online, wanting to understand what biological assets there were, and what happened to them (much ended up in Finland). The nature of research had changed. There is a lot more that can be done more quickly.
Without knowing anything about the topic, how would you evaluate a review article? The same way I did 35 years ago. QC hasn’t changed.
1 Like
It turned SAZ into an Evante 
As it should be for safety, interrogate and reference peer reviewed/published material, apply some critical thought to the process and write your own interpretation/conclusion.
However the prevailing attitude at the moment seems to be chuck the query/prompt at the LLM of choice (which more often than not seems to be ChatGPT) and regurgitate, copy/paste the result as absolute fact while ignoring that the source data in the model itself has not undergone any recognisable review.
The difference between useful academic/industrial use of LLM’s and the publicly available self service options are at polar opposite ends of the spectrum however are usually treated casually as the same.
1 Like
Not Yahoo Answers then 
I sorry,what was that? Have I missed something?
Today I attempted to find the title of a childrens book using AI, it’s a book I’d read multiple times, but always forgot the title of - the last time I read it was in the mid-90’s.
Both Google and Metas’ AI assistants failed miserably, giving me all sorts of incorrect books.
I had asked it what is the childrens book, written between 1900 to 1997 written by a male author where the main character is a school boy, and the schools’ Headmaster keeps his Grandfather or possibly just Fathers hip bone in a jar. Also newsreaders are desribed as wearing no trousers.
It suggested books from 2020 (ignoring the age range), books by female authors (ignoring specified genders), books with female main characters, books about analysis of political systems (definitely not a childrens book), books in the Tinker, Tailor, Soldier, Spy series (definitely not), books in the Harry Potter series and then decided it was The Witches by Roald Dahl (it definitely isn’t).
Finally it said that none of its’ results actually made any reference to the plot details given ,after stating that each book it suggested contained those plot points.