AI fails

[Psamathe]

Yes, indeed re the rainfall % probability, but it's more the 'confidence index' that would be helpful in assessing whether or not the various models are converging.

I'm pretty sure I've mentioned it before, but Meteociel is a really useful: you can Google "Meteociel xxx" for UK cities, and it gives you the raw output from five or six different models (including ICON) without human interpretation, and if you look at the different models you can see the degree of convergence (or not) for yourself, so giving a fair picture of 'confidence'.

I use the FlowX app (iPhone, iPad, Android, Mac) which gives access to loads of different forecast models (including ICON, KNMI, Meteo France, CMC, ECMWF, RMI, NOAA in the UK, other forecasts for other areas of the world).

It presents the data on a map you can time scroll. Couple of features I find really useful are that it gives up to 14 dys forecast and eg select ICON-D2 (best in my opinion) when the forecast time ends it automatically switches to ICON-EU, etc. ie the 14 days forward includes several forecasts depending on how long they forecast. Plus you can display several different layers at the same time eg wind streamlines and rain and maybe could, etc. Vast numbers of parameters and very configurable eg in example I have sun, cloud and wind on upper graph, temperatures on lower though for some locations have winds (speed and gust plots) on lower graph. I also have the Sun & Moon positions displayed. Just slide finger vertically (fine) or horizontally (course) to scroll through time.

Line almost vertical to beft is because I've set it to ICON-D2 and that line is the limit of the D2 area so west of the line is showing ICON-EU.

It's a free app BUT optional subscription and most of the good stuff is subscription only. But as I do a lot of outdoors I find it worth the subscription (for myself).

 

[briantrumpet]

I use the FlowX app (iPhone, iPad, Android, Mac) which gives access to loads of different forecast models (including ICON, KNMI, Meteo France, CMC, ECMWF, RMI, NOAA in the UK, other forecasts for other areas of the world).

It presents the data on a map you can time scroll. Couple of features I find really useful are that it gives up to 14 dys forecast and eg select ICON-D2 (best in my opinion) when the forecast time ends it automatically switches to ICON-EU, etc. ie the 14 days forward includes several forecasts depending on how long they forecast. Plus you can display several different layers at the same time eg wind streamlines and rain and maybe could, etc. Vast numbers of parameters and very configurable eg in example I have sun, cloud and wind on upper graph, temperatures on lower though for some locations have winds (speed and gust plots) on lower graph. I also have the Sun & Moon positions displayed. Just slide finger vertically (fine) or horizontally (course) to scroll through time.

Line almost vertical to beft is because I've set it to ICON-D2 and that line is the limit of the D2 area so west of the line is showing ICON-EU.

It's a free app BUT optional subscription and most of the good stuff is subscription only. But as I do a lot of outdoors I find it worth the subscription (for myself).

View attachment 9940

Thanks.

 

[Psamathe]

Moment ago listening to The Naked Scientists Podcast and they. accounted the case of a person who went to hospital in US claiming they were being poisoned by a neighbour. Turned out they'd asked ChatGPT for salt alternatives to reduce salt intake and help blood pressure. ChatGPT recommended they switch to Sodium Bromide on food (rather than Sodium Chloride). After switching for the last 3 months this caused him to suffer bromism which causes psychoses and hallucinations. Took several weeks to get the bromide flushed out of his body and recover.

 

[Rusty Nails]

I've been mulling for a while what the Briantrumpet School of Excellence's motto would be,

"ventus meus melodicus est"

 

[briantrumpet]

Oops. I guess he didn't follow the homework instructions carefully enough.

 

[briantrumpet]

 

[PurplePenguin]

View attachment 10115

Still just reminds me of the time when some people considered Wikipedia unusable because some things could be wrong.

 

[PurplePenguin]

Disappointed in AI today. Asked for a list of test match results at Lords between England and Australia. Everything prior to 2001 was just a made up response. This info is on Wikipedia albeit on different pages.

 

[Pblakeney]

Disappointed in AI today. Asked for a list of test match results at Lords between England and Australia. Everything prior to 2001 was just a made up response. This info is on Wikipedia albeit on different pages.

Prime example of AI being useless today. It might be great in the future but it is just a little bit sh!t today.

 

[briantrumpet]

Disappointed in AI today. Asked for a list of test match results at Lords between England and Australia. Everything prior to 2001 was just a made up response. This info is on Wikipedia albeit on different pages.

To be fair, that kind of thing ought to be fairly easily fixable, though it would be interesting both to quiz AI now - "Why have you invented the results prior to 2001?" - and to ask real humans why.

 

[PurplePenguin]

To be fair, that kind of thing ought to be fairly easily fixable, though it would be interesting both to quiz AI now - "Why have you invented the results prior to 2001?" - and to ask real humans why.

My assumption is that after 2001 there will be lots of internet noise about it e.g. BBC news reports whereas there is far less noise about the 1934 result. If I ask explicitly about the 1934 result, it responds correctly, but when producing a list it just wings it.

I still think it is better than Google. I did tell me I should go to Cricinfo for an accurate answer.

 

[briantrumpet]

My assumption is that after 2001 there will be lots of internet noise about it e.g. BBC news reports whereas there is far less noise about the 1934 result. If I ask explicitly about the 1934 result, it responds correctly, but when producing a list it just wings it.

I still think it is better than Google. I did tell me I should go to Cricinfo for an accurate answer.

Rather undermines the current claims then, given the hype about how it can find and collate data. Sounds more like me doing school homework on something I had no interest in and couldn't be bothered to do properly.

 

[icowden]

To be fair, that kind of thing ought to be fairly easily fixable, though it would be interesting both to quiz AI now - "Why have you invented the results prior to 2001?" - and to ask real humans why.

In Google you just click on the paperclip next to the paragraph you are interested in. It will take you to the website that it harvested the info from.

 

[Rusty Nails]

At the moment I believe AI is more useful for scientific and medical research purposes than the rather trivial purpose of saving people the time of searching on Google or other sources for occasionally subjective and ‘nice to know’ stuff.

 

[C R]

At the moment I believe AI is more useful for scientific and medical research purposes than the rather trivial purpose of saving people the time of searching on Google or other sources for occasionally subjective and ‘nice to know’ stuff.

Not really, you see the same problems with technical questions, and code produced by code assistants tends to need a lot of vetting. For the time being I remain skeptical about LLMs usefulness.