AI-pushing Adobe says AI-shy office workers will love AI if it saves them time Adobe recently surveyed more than 1,000 employed Americans about their view of AI, and found that they could overcome their concerns about losing their jobs to AI if it saved them a substantial amount of time. The subscription software seller, as it happens, has a horse in this race: Acrobat AI Assistant. Adobe decided to look …
One wonders whether there's a market for an AI that will be able to sneak tricky terms into a contract, for example.
I recall an attempt by a non-european client to get his contractual terms and conditions accepted on a project. On a read through, the client had included a clause which said he had the right to a sizeable discount if his payment was made early (by not very long).
The cheeky bugger got kicked into touch.
I just wonder if the contractual implications could have been deeper hidden.
@AC "One wonders whether there's a market for an AI that will be able to sneak tricky terms into a contract, for example."
And therefore also a market for an AI that will be able to find tricky terms sneaked into a contract by AI. ;)
But why stop there, think of the cost savings on lawyer's fees with AI. AI writes the contracts for another AI to read. Contracts disputes will be argued by an AI for the plaintiff and an AI for the defendant, before an AI judge and in front of an AI jury.
Don't think of it as lawyers and judges losing their jobs and being replaced by AI, it's making lawyers and judges more productive with their time by freeing them up from the drudgery of contract law.
We already have AI-generated email being processed by AI recipients, with no humans in the loop; you can find multiple accounts of that sort of thing online. There's already significant use of AI in processing job applications, and in generating those applications for processing.
We've successfully created a new system for automating the waste of considerable resources. Productivity achieved!
A non-technical person I know is in the process of letting a house to her friend. She was gleefully showing me a few weeks ago how she could use Gemini to write a rental agreement - and was quite dejected by my horrified reaction. She just couldn't understand (a) that the output might not be completely (or even remotely) correct, and (b) that even in the unlikely event that it was, none of us was qualified to asses its correctness. (And that's leaving aside the ethical aspects of how the training data was sourced.)
So yes, it wouldn't surprise me if that ends up being a use case for AI.
I have spent most of this week writing low level design documents, which require lots of checking, and cross referencing to ensure the spare is there for the new equipment, the cabling exists, and all the weird configuration of the existing devices which needs to be replicated (some of it I am still puzzled how it actually works at all). The documents themselves are only 15-20 pages long for each system but still take at least 1/2 (one took 2 days) to write.
Sorry I don’t think any sort of ai can same me much time, by the time you have figured out how to tell a system to do it and then double checked to ensure it is accurate it is easier to do it manually.
The five enabled copilot of Office apps and now everytime I select more than 3 characters it appear to ‘help’ do a copy and replace….
Yeah, I've spent significant chunks of my life writing docs like that.
But the question always in the back of my mind was, is anyone going to read them?
I can imagine, if those (15-20) page docs are the only (user) documentation that exists on whatever, there might well be an interest in boiling them down to a couple of paragraphs each. If you don't do it, maybe your employer will, and if they don't, their customer might. So it's probably best to do it yourself, that way at least you can check the output.
There's a lot of adobe construction around here, and I mentally wince whenever I think of the term, thanks to association with this dreadful firm.
Postscript was pretty good. Display Postscript was an interesting idea, if not really successful. PDF was ... well, broken in a bunch of ways, but useful. Acrobat, in its various forms over the years, started bad and steadily got worse. I'm fortunate not to have had to deal with the rest of Adobe's terrible decisions.
Instead of using an LLM to summarize or mitigate the impact of lengthy reports on your worker's sanity, stop writing them.
If they're needed, as in detailed, carefully constructed engineering documents or specifications, then given the expertise and time needed to write and read them, there won't be that many.
For all the other ones I suspect the information content is near zero yet ambiguity near infinite.
On 'A recent study of bloggers found that they produced more blog content with ChatGPT – though at the risk of becoming less popular with some readers.'
They may have produced more content, but were not more effective (in that their audience walked away). One reads blogs for interesting insights, not for articles with a digit in it followed by Here's why/how/when.
A bit like recipes on American blogs. They start off with some preamble about their aunt, shoehorn some stuff about their kids, maybe reminisce about their time at university, dump a load of photos in before finally getting to the actual recipe...
...And then you close the tab because it's all in pounds and ounces...
We had someone here who was all ChatGPT this and ChatGPT that.
Turned out she was using it in her work, which was bad because a) it was plain wrong and b) probably exposed our proprietary information, going against an explicit "don't send our stuff to LLMs" directive.
So she's gone now.
It was caught when a co-worker saw the nonsense bullshit in some of her documentation.
And there's the problem right there : people will apparently accept anything that hoovers up their personal data as long as they see a smidgen of benefit.
Funnily enough, I don't think there are many people who would agree to get beaten every morning if it meant they would be paid 10% more, but as soon as computer is involved, people lose all sense of measure and can only think "oh, if I give up all my data I can get a voucher for free ice cream ? Deal !".
Pfff.
"The other side of the coin (assuming all this snake oil turns out to be true) is that the bean counters will turn around and say "3 people can do the work of 4 now”
And that includes the work of ‘bean counters’, so which one of you four is going to be let go? FIGHT!!!
Okay, so the workflow is:
- manager quickly writes down bullet points or a short text that contains the information.
- manager tells AI to make that into a "proper" document, that has nice formatting and is at least a few pages long and official-looking and with all the niceties at the beginning and end.
- manager sends the document to employee.
- employee tells AI to summarize into bullet points or short text.
- employee quickly reads the AI output.
Now, I don't know, just hear me out here, but I think I can spot a way to increase the workflow's efficiency and save money at the same time.
So Adobe decides that the solution to document storage, filing, scanning and ensuring compliance with paperwork retention regulations is... to subscribe to an AI to summarise the paperwork?
I assume this can also be bundled with their range of chocolate fire engines and magnesium fire guards?
Workers would like AI if it saved them time?
Well, yes. The problem, however, is the time it takes to give AI the initial instructions, to then review the output and compare that to the requirements to then refine the instructions to get a new output to review and refine until the output is to specification to then CHECK the output is actually correct and hope it is else you'd have to start all over again*.
It would take a lot less time to just ask someone to do the work, for them to do it, noting they'd ask for clarification where needed, to get a second person to check it and then submit it.
Much less time.
* Just got an example of 'AI' output earlier, shared to the tech team. How many 'R's in Strawberry? Apparently 2, and when asked to highlight them, it highlighted all three then apologised for highlighting too many as there were only 2. Honest!
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
