I never claimed you did say that 
The “search” is trivial. To find the shortest route in a maze is very very fast on modern hardware. What takes the time is downloading and processing the tens of thousands of restrictions which you download from EC.
In the old days you submitted each iteration for validation and it would take seconds to get the result back. I think nowadays one downloads a bunch of the restrictions in one go.
Here is a nice video of AI in general, and ChatGPT in particular.
Peter wrote:
The challenge today is providing continuous support to irate users for free
Perhaps use ChatGPT to provide the support? :-)
Here is a nice video of AI in general, and ChatGPT in particular
I like her analogy to quantum physics and systems, “you don’t have to understand it just shut up and compute”
If one has not been along on what AI can achieve in language models in terms of understanding, perception & creativity, these are core stones concepts that are tightly linked when when talking Deep Neural Networks, I can suggest getting a step back and having a look at what AI did for image processing
This is non technical talk but it give a glimpse on how understanding & creativity are built out of an automated perception, I have been in courses where the maths were spelled out all over the board here ComputerVision
Has anyone noticed forums filling up with ChatGPT-generated posts?
See post(s) by yueri.1 here. This is quite a specialised forum for STM processors. You have to keep clicking on “show more answers” – that forum software is one of the crappiest ever.
ChatGPT is not “AI” in any way. It is a clever text classification algorithm which collects statistically relevant text from around the internet, and it’s amazing how much one can achieve by doing that. The last para in this guy’s post has the giveaway ChatGPT style – “sickly sweet” as it is called here:
Seek help from the STM32CubeIDE community: If the problem persists, you can visit the STM32CubeIDE support forums or the STMicroelectronics community forums to seek assistance from other users or the development team. They may have encountered similar issues or have additional insights on how to resolve them.
Why do people spend their life entering forum posts into ChatGPT and posting the text as a reply? Well, millions of people don’t have a life… and in some cases they benefit by building up posting reputation points which enable a later post on something else. For example, these days about 1/3 of EuroGA joinups are malicious (which is why I have to manually approve them) and the other day we got five (5) people from India, apparently trying to advertise some steel rolling mill products!
A few years ago we had a Greek guy here, running some gyros joint in Birmingham, who was trolling Greek forums for years, got chucked out of them all, and was trolling here. With ChatGPT he would have lasted longer, and would have lasted for ever if the moderator was just modding for policy and didn’t know anything about GA.
Since posting can be automated (as soon as somebody has posting privileges set up) I would expect FB and twatter to be full of ChatGPT posts. Same for any high-traffic forum; they cannot be modded for content.
Here is the ChatGPT-generated post in question
As a good friend of mine defined few years ago:
1. AI – a lot of ifs
2. Cloud – someone else’s computer
Nothing new in IT world, just new buzzword for lame people to bite.
I’m kind of entertained by the whole furore about AI. I’m sure there are some real issues, like kids using it to generate their homework and teachers not having either the time or the expertise to realise what’s going on. Though it probably wouldn’t be THAT hard to train an AI to recognise AI-generated text. (Hmmm, business opportunity there).
As Peter says, it’s just basically like Google finding web pages for you, except it also creates text based on the content of those web pages.
For job reasons I’ve been looking at how AI text matching works. It’s clever, but there is absolutely no comprehension of what things mean, just clever matching algorithms.
[Aviation content] Did you see the case reported yesterday about someone suing an airline? The prosecution came up with a lengthy document citing various previous relevant cases. Except none of them existed – he’d used ChatGPT which had made them all up!
johnh wrote:
I’m sure there are some real issues, like kids using it to generate their homework and teachers not having either the time or the expertise to realise what’s going on.
It was like that when any new technical solution emerged since ever. I remember professors freaking out when something, that was supposed to be 30 min calculation during the test, was solved with programmable calculator within few seconds and they were not capable stopping this because there wasn’t any rule of allowed type of calculator.
Plagiarism has been huge for many years in any education subject where “essays” are involved – basically all of humanities at degree and especially masters level. There is accordingly a big business with essays at €xxx each, slightly more if you want one urgently, and generally they are produced by retired univ lecturers. We’ve even had spammers here advertising essays.
You can spot them immediately because you see a chunk of semi-illiterate English (typical of foreign students in the UK) and then pages of prose written by Shakespeare But for PC-reasons (the higher education system is totally tied up with PC) the univ cannot act on this “evidence” unless the source is identified, and if you bought the (human-written) essay then it won’t be. That was the case at Sussex Univ, at least.
ChatGPT will presumably do these essays well but will be easy to spot, partly from a total lack of a “personal view”, but whether the univ is allowed to act is a different matter. “Obvious” is not “evidence”.
One interesting thing is that if you trawl the net (ChatGPT uses an archive from 2021 but others use current data) you will obviously pick up largely “opinionated” material, including extreme-political stuff, and ChatGPT uses a load of code (I avoid the word “engine”) which identifies this and turns it into bland prose. It must use language parsing to identify “unsuitable” material.
A while ago we were messing about at home and I asked it “how do you impersonate a frog”. It produced a load of ultra-PC prose, worthy of the BBC, warning that impersonating a frog could be harmful to the animal.
he’d used ChatGPT which had made them all up!
You don’t know whether they really existed online somewhere. They probably did
