My team and I have been using Copilot for over a year. We find it useful for technology related bugs and answering specific algorithmic questions. When I write new code, however, I find myself turning Copilot off as its suggestions are not only useless, but also way too distracting.
As someone who is part of my company’s AI initiative, I have tested the code generation abilitiy of Copilot and have found it pretty much useless. Just asking for a simple ‘create a vite react application that shows a single page view’ will create a whole lot of code that just does not work. Not only does it not work, but the code is garbage.
I am mentoring a young college student who is interning for a summer. I cannot tell you how many times Copilot has taken him down the merry trail of ‘hallucination’; leading this poor kid on wild goose code-chases of function that is simply wrong. After a while I told him to turn it off. Not only did he start to actually learn; his questions to me decreased noticeably .
I use Copilot mostly for summarizing meetings that I do not have time to attend. For that it seems pretty useful. It is also really good at summarizing information scraped from sites like stack-overflow or reddit, and answering specific questions. Other than tasks like that - the only people using AI seem to companies who stand to make money from its adoption. These stories that AI will put developers out of work are usually coming from companies who are trying to sell me products that promise to do this miraculous transition. to automation for me.
Is anyone getting better results? What is your experience using AI at work?
Ngl, it’s an amazing help, I use ChatGPT plus, and it able to cut through the weeds on dry spec document, prints, analysis, and knocks out meeting minutes for me. Instead of wasting hours cutting through 2k pages of specs to find a certain thing I just ask it where it’s at, and if a particular product will work for it etc etc.
If you set the right preliminary prompt for it, it can work wonders. I wouldn’t use it as an end-all-be-all, it’s just a tool. Like a saw in a carpenters hand or a brush to a painter.
I have a friend who was using paid ChatGPT for a year before his company adopted copilot, and I just asked him a few months ago about it and he said copilot isn’t at the same level as ChatGPT. Just hearsay, I only use ChatGPT as an advanced search engine and as an experiment. After I wrote my “influence of the cascade” doc, I asked it to turn it into an excel sheet with the categories I made to better visualize the data.
AI hallucinates too much too provide good feedback for my work which is cybersecurity governance. Next thing I know the SSP reads like the environment is bullet proof on prem when in reality it’s just a cloud app….
I don’t use it in my tech support job regardless of how much the industry and my employer keeps trying to convince me how great it is. I don’t have any interest in teaching it how to answer my questions. And I have to fact check anything it spits out anyway, so I nope. I’ll just reseach myself. I don’t need the extra tasks. If it provided dead accurate answers to my queries, I’d find it useful. It doesn’t.
I try but have no luck. No matter how I phrase my prompts I, for the life of me, cannot get ChatGPT to change the timing belt and water pump on this Honda.
Yes one of the many gripes I have with AI is that it is 100% confident all the time regardless of how stupid and useless answers can be at times. Here is your answer I have saved you shouts the AI handing me a regex pattern that will not do anything remotely like what I requested. Here is some code that will wipe your database, you’re welcome!
I have found it helpful for hyping me up in one pagers I have to write to make the doc cleaner and make me sound amazing.
That being said, we recently went through a surprise second annual review and one of the requirements in order to be scored as remotely employable was you have to be using AI and be creating projects that implement AI solutions. So I guess I’m gonna be using AI to code more AI regardless of the quality because I like keeping my job. So it’s AI all the way down from here.
I’m sure im about to lose my job any second because I won’t conform and also expect that any job I apply to will not care about the 20 years experience I have supporting Cisco Fibre channel switches. The only question i will be asked in an interview will be how do you use Ai, etc..
We use AI a little, but as mentioned it is often very confidently very wrong, so much care needs to be taken when using it.
We have a bunch of paper engineering reports dating back to the late 1970’s, and we’re nearly done digitizing them all. Once they’re all digitized, we are planning to try using a sandboxed AI to help us use all of those reports like a giant searchable database. I think that might be the most effective use of AI for my work, but at this point that’s just a guess, it might end up being a total boondoggle.
We have developed several extremely useful and powerful software programs that utilize machine learning, which is often considered a subset of AI.
Yeah honestly I’m morally opposed to AI in theory in that it often steals the work of others.
Where did it learn that horrible code? It mined the hard work of thousands of crappy interns. But while they weren’t good at their job that was their work they worked hard on and now an algorithm is gonna take all their spaghetti code and serve it to me.
All that being said in this case my morals are for sale. If I have to pretend that stuffing AI into toasters or whatever betters humanity in order to give health insurance and food for my family then aye aye captain.
Yes I was lazy and wanted chat gpt to write me a quick and dirty python script to act as an auto clicker and the code it made was absolutely worthless. I ended up writing it myself… this was for personal use to deal with a dumb issue I had on my old laptop.
I work in a fire engineering consultancy for the construction industry, so given that context I’m very averse to any use of AI to come up with technical solutions. I’m dubious on how effective it even would be at coming up with reasonable solutions to common/basic technical issues let alone figuring out a robust fire engineered solution for a complex building. I think its use for that would also be rather limited inherently given a big part of our role is presenting and arguing the technical justification for our advice to design teams and approving authorities (so parroting a solution I don’t understand from a chatbot would fall apart quickly lol).
My company has been using ChatAdvisor to build up an internal database of past bids and various guidance documents (part of the idea being that limiting its database in this way should reduce hallucinations). So far I’ve only really used it myself to scan large guidance docs and point me to where a topic of interest is (so basically a fancy ctrl+F) or compare iterations of documents to get a quick and basic idea of the differences between them and where the changes are so I can then read myself. I know others have used it for preparing bids for work based on that internal database or to use it to search and see if we have past experience in a type of building. I’ve not had much cause to do either of those yet since for the bids I prepare it’s usually easier and quicker to just modify a template.
I could see using it perhaps to scan technical submittals to summarise relevant info to fire given these are sometimes hundreds of pages long with very little relevant info to our role and it should then be able to point to where in the document this info is. I could also see using an LLM to figure out how to phrase email responses to difficult clients perhaps, though anything like that would need some thorough checking to make sure the technical info isn’t lost as a result.
That’s the one thing I want to use AI for because I don’t wanna write unit tests. I don’t wanna do it.
and maybe if it could add comments and documentation to the 20k lines of Java I inherited recently that only has documentation in the form of a video of the developer talking to the camera before he left.
fradiger
(the world is a beautiful cat and i must meow meow meow)
20
I currently do research on how ChatGPT and other LLMs impact and can be integrated into education (university-level coding courses) and the amount of different ways to use these tools is resulting in them being an extreme force multiplier. These companies that think LLMs will turn all of their workers into hyper efficient task robots (or ..replace all of their employees with “hyper efficient task robots”) are mistaken. Instead, the users ‘properly’ integrating these tools into their workflows are the ones reaping the benefits (note: an active research field is figuring out what ‘properly’ actually looks like).
We used to see students modeled on a classic normal distribution (bell curve). However, as LLMs and AI become pervasive in a student’s day to day, you will start to notice an interesting phenomenon. When the material gets harder and these students get into more technically challenging courses, performance of students actually starts to look more binomial in nature because the group of students that has been relying on the LLM to do their work starts to face significant challenges, while the other groups have less trouble adapting to the harder material.
