It’ll just been an evolution of job creation. Same way we don’t have car parking attendants anymore, lift operators in shops / hotels etc etc
Got something to say or just want fewer pesky ads? Join us... 😊
[Technology] What is the most impressive thing you have seen AI do?
- Thread starter Dick Swiveller
- Start date
Green Cross Code Man
Wunt be druv
That is interesting. Maybe they haven't looked at/generated enough of their own AI content to notice the style. I say style with more than a hint of generosity. AI is so formulaic in the way it writes, it seems, at the moment at least, really obvious. Just lacks humanity.
To be fair, in my final year teaching I mix up conventional facts about current medicines, with mechanistic detail from animal research, and insights about how to make better drugs (e.g., via engineering greater disease selectivity). In other words I expect my final year students to use facts to create a speculative narrative that weighs up the strengths and weaknesses of competing ideas and information. Using judgement. If I asked them to write an essay about treatment of heart failure I'm sure ChatGPT could knock up a first class effort.
- May 8, 2018
- 9,297
Once saw him do an all dayer at Cheltenham and still turn up for the 6am TalkSport Breakfast show
When I was at uni I sat a third year exam in which we were given four questions and had to answer three of them.To be fair, in my final year teaching I mix up conventional facts about current medicines, with mechanistic detail from animal research, and insights about how to make better drugs (e.g., via engineering greater disease selectivity). In other words I expect my final year students to use facts to create a speculative narrative that weighs up the strengths and weaknesses of competing ideas and information. Using judgement. If I asked them to write an essay about treatment of heart failure I'm sure ChatGPT could knock up a first class effort.
Two of the questions were the same. Like, the numbers were slightly different, but the problem and solution were identical.
Even at the time as students we thought that was taking things a little too far. Not enough to question it or object to our own marks or anything though. Obviously.
We’ve been looking at this in library circles as we have found that it makes references up. It goes by what is plausible, rather than what is actually true, and in doing that it’s creating articles and books that don’t exist in reality. It’ll often even show PIDs but they won’t link to the correct resource or or go to an empty page.
If you have the option, RLUK did a really interesting talk on inbuilt bias in Ai as one of the keynote talks at their conference. It was excellent, really interesting and shows a lot of the issues with AI.
Green Cross Code Man
Wunt be druv
Yes, it seems the algorithm seems to necessitate giving a plausible answer at the expense of accuracy and even to resort to fiction as a last resort. I tested chatgtp and Google bard on a load of 60's and 70's type rock track lyrics. They both may have improved in the past 6 months but they both attributed lyricists incorrectly and also mismatched bands with tracks.
Sorrel
Well-known member
In the field I work in, if you ask AI for a high level summary or to do reading comprehension it's great. If you ask specifics and details it is only any good if it has been specifically training on the right level of depth on the correct data. What matters is that training.
While there is no reason that an AI can't be an expert on lots of topics, in practice there are likely to be specialised AIs for specific areas, to work on a clear set of task: generalised AI is more like a toy or Wikipedia in terms of detail. For the next few years there are going to be a lot of humans checking what AI comes out with. After that, though, there are likely to be fewer in some areas and there will be an interesting set of mistakes.
While there is no reason that an AI can't be an expert on lots of topics, in practice there are likely to be specialised AIs for specific areas, to work on a clear set of task: generalised AI is more like a toy or Wikipedia in terms of detail. For the next few years there are going to be a lot of humans checking what AI comes out with. After that, though, there are likely to be fewer in some areas and there will be an interesting set of mistakes.
After playing a lot with the AI image generation in Copilot. I have spent today playing with AI song generation. The lyrics are poor as I used Copilot rather than writing my own but this is the result of asking Copilot for a song about NSC in the style of 1920s jazz with a solo singer and then asking Suno to generate the performance from just those lyrics. Obviously the style is wrong, it cuts off early and the aforementioned lyrics are poor but the way it creates this in 5 minutes blows my mind. Also playing with software that lets you download voice models and make it change the vocals on a tune and whilst it only works well with parody voices like Cartman, it is good with similar genre voices.
NSC by @wrychoirs227 | Suno
1920s jazz quintet with solo singer song. Listen and make your own with Suno.app.suno.ai
Any other mind blowing/scary uses of AI out there?
I've been playing with this. It's ace.
From Withdean to Wembley | Suno
anthemic indie rock song. Listen and make your own with Suno.
Dick Swiveller
Well-known member
- Sep 9, 2011
- 9,160
- Thread starter
- #31
https://sonauto.ai/Home is good as well. You can search for an artist and it will give you a load of keywords to make the song. I gave it Bonnie Tyler and it got the gravelly part great in the chorusI've been playing with this. It's ace.
From Withdean to Wembley | Suno
anthemic indie rock song. Listen and make your own with Suno.