I think we have an easier job determining right away if humans are lying about something
I think we are equally bad at determining lying between humans and AI. The people who are getting fooled by AI answers would click the top result on google and get fooled there as well so I dont see it as a massive decrease in info quality even though I can admit it is a decrease.
So it sounds like AI is only really useful for your line wider area of work
There are plenty of jobs where it will be more and less useful but my claim is that its generally still useful for majority of professions/people as a quick way to retrieve info via natural language querying. The results are mostly accurate and can include sources if you ask. Thats good enough for most people and most questions. Sure if you need to dig through docs or reference the exact paper then you can search google and get it yourself.
I dont think Teachers is a good example, teachers use it all the time and seem to really like it, idk what structural engineers do so i cant really comment. But even if the engineer has no use I still say its a good feature because not every feature of an operating system has to be used by 100% of people. Windows screen read is a good feature but I dont use it. The share button is a good feature even if I dont use it. Carplay is a good feature even if I dont use it etc.
I agree for excel, I think copilot in spreadsheets is using the wrong tool for the job. Its just a bad use of the tech. I also think the price for an enterprise sub is way to expensive for what you get.
So you agree that pressing a button to bring up a box that you can query with natural language is a good feature you just think the LLM part is slower and computationally inefficient? I could agree with that if there was something better proposed. I just see an LLM being a good tech for this because of how dynamic it is and with the addition of tools to do specific tasks in a determinism fashion its a powerful tool for the users.
Not sure. I think the right will but I also think if they dont then its the better time for a bloody conflict.
You'll have a lot more people on your side after an election loss because trump is already unpopular and people will feel cheated out of democracy. Whereas if you start a conflict now your side will look like the ones trying to overthrow democracy since he was the legitimate winner.
I think that might be a bug. I tested it now by opening a spreadsheet pressing the copilot button and asking something about the sheet and it returned the answer correctly and didnt ask for it to be uploaded.
ah yes i forgot to list the 100 edge cases that cover 0.0001% of the phone market but you know what I mean.
I think we will get something similar to copilot on linux, obviously there will be distros that dont ship it but people are working on projects and once the open source models improve and the UI/UX workflows get polished up I believe major distro will ship them. The linux foundation owns the MCP protocol which is the goto protocol for AI's to interact with things.
I also doubt they provide a huge increase in productivity. I'd say between 0% and 10% productivty increase and its more of a quality of life and small time saver than a powerhouse productivty tool like its marketing suggests.
Obviously I am not talking about copilot being built in to other OSs. I am talking about an OS assistant that uses LLMs to handle natural language queries being built into OSs.
Yes it has, IOS and Android have had siri, bixby and google assistant for over a decade.
In copilots current form its chatgpt with some basic functionality added on. Its a baby compared to siri, bixby and google assistant. I think its a good addtition to windows and we will see it grow to be a more powerful version of google assistant and we will see this develop and be implemented on the other major desktop OSs.
I agree that LLMs can erodes critical thinking skills and can be unreliable but I think they ARE fit for purpose for majority of searches and queries people have day to day and people are figuring out what is and isnt a question for an LLM. Like im not going to ask an LLM how to configure some piece of software I'll go to the docs and read it because i need this to be configured correctly. I wouldnt ask an LLM if I can eat this weird mushroom because i might die if its wrong.
But I would ask an LLM what tech I can use if I want to get X result and then look through the summaries of each suggestion. I would ask for a report or document template to be generated because im proof reading the document anyway. I would ask for help automating a task. I would ask for help writing random low effort slop posts that I have to do for office stuff, like marketing emails, event announcements etc.
My reasoning for this post is that even though I dont like copilot currently I can see that at its core its a good feature and with the right polish it can be a great improvement for users. A big gripe i have is that marketers have way overpromised what assistants like copilot can do. When i speak with other people I can see they have already been leaning heavily on natural language queries for over a decade now and having this built into the OS would be a huge quality of life improvement and would improve what the tool can do. People already been outsourcing their thinking to google many years ago so I cant pearl clutch over doing the same with chatgpt and we can put our heads in the sand(like most people in this thread) and pretend people arent using these LLMs for information but the reality is that they are and we need to accept it and be involved in building the software that people want.
I'm a linux user and I think it would be very useful to be able to click one button and say "Set a calander event for the 25th my dads birthday and set a reminder a week earlier" and have it set that. It works on mobile just fine.
Ok so your main complaint is that its to energy intensive? Would you concede that its an OS assistant is a good feature if the query computation cost was lowered? Because I'd argue it already is and the cost of an LLM query isnt unreasonable. The large power costs come from model training and per query cost is negligible.
Also I wont make an argument on the copyright for training data because i dont respect copyright.
Do you think majority of people disable their phone assistant or care about data privacy? You've raised extreme minority positions and expect that to be a convincing counter argument to a generalized statement. I'm sorry but thats just out of touch and even though i hold the same minority positions I still acknowledge the realities of what average users want from technology.
It could be traffic from outside this community, so they havent read the rules in the sidebar. Maybe people just hate copilot so much its a subconscious reaction lol.
Ok thats cool but how many people do you think disable it? If you think its a bad feature thats fine but im trying to make a general point about assistants being considered a good feature and a single example does nothing for me.
If its a situation where bad info can cause you to waste significant time and lose money then Copilot is not the right tool for the job. But thats not every query and I wouldnt even say its more than half our search queries.
Its boring that people are completely unwilling to engage at all with the arugment. Comparing a situation where you get the wrong info from an LLM with a situation where you get the correct info from google is pointless and shows nothing. Ignoring that people get the wrong info from google all the time is so disingenuous. Everyone would acknowledge google is still a useful resource even if there is the chance to come away with the wrong answer.
So the real comparison to make isnt correct answer from google 100% of the time vs correct answer from copilot 60% of the time. Its correct answer from google 90% of the time vs copilots 70%. So you can weigh up the options and use the right tool for the job.
We can acknowledge that assistants like siri are good even though they have had data privacy issues. Youtube has privacy issues but I still use it and consider it a good service. Lemmy has privacy issues and I still use it and consider it a good service.
Overall Siri is a good feature is good and a majority of the users use it.
I think we are equally bad at determining lying between humans and AI. The people who are getting fooled by AI answers would click the top result on google and get fooled there as well so I dont see it as a massive decrease in info quality even though I can admit it is a decrease.
There are plenty of jobs where it will be more and less useful but my claim is that its generally still useful for majority of professions/people as a quick way to retrieve info via natural language querying. The results are mostly accurate and can include sources if you ask. Thats good enough for most people and most questions. Sure if you need to dig through docs or reference the exact paper then you can search google and get it yourself.
I dont think Teachers is a good example, teachers use it all the time and seem to really like it, idk what structural engineers do so i cant really comment. But even if the engineer has no use I still say its a good feature because not every feature of an operating system has to be used by 100% of people. Windows screen read is a good feature but I dont use it. The share button is a good feature even if I dont use it. Carplay is a good feature even if I dont use it etc.