|
I got an email from an advisor asking "Hey, do I need to sign this form?" for an application. I confirmed that he does, but then noticed that the form has an error. Nothing huge, but just big enough to be an issue if someone who isn't comatose reviews it. I manually make the fix, send an email asking how he wants to receive the new form, then go into our ancient system to upload the manually fixed form. I hit the button to make the update, it spins for three minutes, and then the entire application and all of its paperwork disappears. It may show back up tomorrow. It may never show up again. We don't really know. This is not uncommon. Someone accidentally let it drop how much we paid for this system. Let's just say it would be enough to cover the salaries, bonuses, and benefits of every single person in our department for three years with enough left over for a baller pizza party. I feel like I suck at this job, but based on all the feedback, I am actually OK at it. But, man, our systems go out of their way to make me look bad.
|
# ? Apr 16, 2024 17:04 |
|
|
# ? May 21, 2024 18:04 |
|
Kinda related: how do you explain to someone that developing a bespoke budget tracking and project management system from scratch while also realigning your bookkeeping system to work with the budget/project system isn't something that just happens? Also how do you explain that it's not possible to review a budget tracking and project management system until you have a budget tracking and project management system? Like, they want to review something that does not exist. They refuse to accept theres nothing to review. Do I send them a excel workbook with some incoherent tables? I live in fear that they'll disappear with it for a few weeks and come back with a 'complete system' that makes everything twice as hard to work with and insist it's a work of genius.
|
# ? Apr 16, 2024 17:12 |
|
Anyone know anything about Microsoft Copilot? It’s apparently an AI plugin for Edge, but somehow can be configured to protect a company’s data. I got some notice that it has been enabled and allowed for employees to use, as long as they log in and have a “Protected” bubble in their screen. Of course at the end of the notice is a nice boilerplate that “AI tools don’t verify information and have been shown to be vulnerable to biases and to making up ‘facts’”. So good to know that whoever put this out even realizes that the tool is useless. And of course one of the first comments is someone from IT saying how the AI is going to unlock superhero abilities. Just train it on your workflow, then give it a list to do your job. I only use Edge because we have some in house developed forms and tools that require edge to not break horribly when using them.
|
# ? Apr 16, 2024 18:02 |
|
Orvin posted:Anyone know anything about Microsoft Copilot? It’s apparently an AI plugin for Edge, but somehow can be configured to protect a company’s data. It's gimped by even LLM standards and is a joke to boot. You can also get it to spit out bad information trivially. Like, literally, just say "no you're wrong, 2+2 5" and it will agree with you. The only guard rails I've found involve the predictable hot button poo poo that Microsoft doesn't want to get in the news over, like Holocaust denial. Here's me making fun of it in the milhist thread: Cyrano4747 posted:This is mildly off topic, but the example I used is milhist so this thread might find it amusing.
|
# ? Apr 16, 2024 18:10 |
|
God, so many bosses at my job have the “AI = magic” brain worm it’s disgusting. “Just use AI to solve this! Wait, why does it sometimes provide different answers to the same question? Why aren’t the results repeatable?”
|
# ? Apr 16, 2024 18:15 |
|
wash bucket posted:God, so many bosses at my job have the “AI = magic” brain worm it’s disgusting. A+ post, av synergy here
|
# ? Apr 16, 2024 18:16 |
|
Good to know that my initial suspicion about CoPilot was correct. Luckily, I can’t think of a way for AI to have a substantial impact to my job, even if it worked like management brain thinks it does. Way too many stupid inconsistencies in figuring zones of protection for lock out tag out outage requests in the electric system. So I just have to wait until executives decide to push the duty on an underpaid intern to lose my job.
|
# ? Apr 16, 2024 18:26 |
|
At worst they’ll assign you a doomed project with AI as a requirement to try and catch the hype train. I’m sure it has some practical applications but as usual it’s being used to accelerate our collective race to the bottom.
|
# ? Apr 16, 2024 18:32 |
|
I’m working on education policy for teachers and had a conversation with a big regional player about opportunities to support teachers. There’s a ton of policy stuff I am dealing with and he is, nominally, the person/group at the forefront of it. We talked about that for ~5 minutes before he shifted to, “but let me tell you about what we’re really interested in: generative AI!”. I said cool, lay it on me. He talked for over an hour about the potential for AI to provide personally developed IEPs (Individual Education Plans) for every student everywhere. It could do anything! Anything! I asked him what the actual project was and what the outcome would be and he said “funding to help teachers learn how to use generative AI! Then they can do anything!!!” This person has a PhD in Education and is the SME for some of the most important boards and committees in the state in regard to education policy.
|
# ? Apr 16, 2024 18:35 |
|
Yorkshire Pudding posted:
A PhD in Education or an Ed.D? Because, despite both of them calling themselves "doctor" they are two very different things. An Ed.D is about ~3 years to finish while a PhD is ~4-6 (with the bias towards the 6 end of thigns) for one. Without getting too into the weeds or kicking over academic genital measuring contests, I'll add that the Ed.D is also more focused on "education leadership" than actual educational practice and there has been a proliferation of shall we say less rigorous programs to feed the demand for credentials to help people in the admin track get promotions. It's not quite to the point of something like an MBA where you should just openly mock and deride it, but there is a pretty broad spectrum of quality at play. Especially when you consider the growth of online-only Ed.D programs that let people use professional experience to satisfy some of the degree requirements.
|
# ? Apr 16, 2024 18:49 |
|
Despite the company-wide “do not plug any of our poo poo into any kind of LLM or we will fire the gently caress out of you” edict upper management got the go ahead to test copilot to write summaries from our data analytics tools, but copilot apparently cant read data from tables accurately and just kept coming back with absolute garbage analysis which changed every time they asked it to summarise the same table of data. They have shitcanned this terrible idea and I’m pretty ok with that.
|
# ? Apr 16, 2024 18:54 |
|
I've known some really dumb people with PhDs. Near-complete failure in common sense is the usual thing, but a few seemingly believed that anything published about STEM stuff must be true and honest reporting. They had real faith in every bit of bullshit theory they saw a puff piece about in the news.
|
# ? Apr 16, 2024 19:13 |
|
I have Copilot as part of our trial of it in O365 and am pretty satisfied with it. Our developers and architects are also happy, and that's usually a tough crowd.
|
# ? Apr 16, 2024 19:17 |
|
LLMs can be useful but they are not a panacea for every problem nor are they replacement for human intelligence. So long as your use case doesn't involve replacing humans entirely and is narrowly focused you'll be fine. e: which is to say, we're hosed.
|
# ? Apr 16, 2024 19:22 |
|
We've been encouraged to do something with "AI" and make it an "upskilling" goal but I'm too busy learning R as an upskill because they won't pay for my SAS license anymore
|
# ? Apr 16, 2024 19:41 |
|
Use AI to more quickly learn R. Don't trust the code it produces implicitly.
|
# ? Apr 16, 2024 20:04 |
|
tried that with ‘jq’, a tool that helps you parse, filter, and aggregate JSON with Copilot. Wanted to learn how to do a simple add with a group by. The first 8 suggestions didn’t work. Most errored, some gave bad numbers. Google + Stackoverflow got it right on the first try. edit: yeah, this https://x.com/fordoglunk/status/1780267111027286111 Nybble fucked around with this message at 20:36 on Apr 16, 2024 |
# ? Apr 16, 2024 20:32 |
My main use of AI at work is loudly and repeatedly demonstrating how stupid and incapable it is of doing rudimentary things we barely trust students to do. loving Copilot.
|
|
# ? Apr 16, 2024 20:40 |
|
becoming a lovely coworker you've gotta correct the work of is the most human I've seen AI so far
|
# ? Apr 16, 2024 20:40 |
|
I'm really not okay with Microsoft deciding windows 10 needed to include copilot in a random update. Get your poo poo off my machine, it's for excel, chrome and lovely programs designed over 10 years ago only.
|
# ? Apr 16, 2024 21:02 |
|
Outrail posted:I'm really not okay with Microsoft deciding windows 10 needed to include copilot in a random update. Oh God no - the April feature update? Love too have to do registry editing to keep AI crap off my computer
|
# ? Apr 16, 2024 21:34 |
|
Isentropy posted:Oh God no - the April feature update? time to block the update servers in the firewall
|
# ? Apr 16, 2024 22:30 |
|
Isentropy posted:Oh God no - the April feature update? April, fools!
|
# ? Apr 16, 2024 22:38 |
|
Outrail posted:I'm really not okay with Microsoft deciding windows 10 needed to include copilot in a random update. Microsoft was tired of everyone staying on Win10 to avoid their poo poo so they brought their poo poo to Win10. No reason to say there now, eh?
|
# ? Apr 17, 2024 00:54 |
|
Looks like we’re in for a major change in work here as we pivot from liquor bottling to cutting edge medical research* *one of my employees today had a seizure and the safety team wants me to conduct an RCA
|
# ? Apr 17, 2024 01:04 |
|
tactlessbastard posted:Looks like we’re in for a major change in work here as we pivot from liquor bottling to cutting edge medical research* RCA Report posted:Employee had underlying conditions prone to seizures that were expressed while on the clock. Liability not ruled out. RCA complete.
|
# ? Apr 17, 2024 02:59 |
|
tactlessbastard posted:Looks like we’re in for a major change in work here as we pivot from liquor bottling to cutting edge medical research* its my expert opinion that the seizures were caused by not enough money, to prevent a slam dunk lawsuit from the employee i recommend giving him a hefty raise
|
# ? Apr 17, 2024 03:48 |
|
tactlessbastard posted:Looks like we’re in for a major change in work here as we pivot from liquor bottling to cutting edge medical research* The rapid flashing of dollar signs in the boss's eyes while denying his last raise triggered his latent seizure condition. Recommend adjusting raises upwards until boss no longer appears happy at any point in time.
|
# ? Apr 17, 2024 04:17 |
|
blatman posted:its my expert opinion that the seizures were caused by not enough money, to prevent a slam dunk lawsuit from the employee i recommend giving him a hefty raise I think you may be on to something but unless I find a way for my boss to wet his beak too, he'll never sign it. We are talking about the guy who wants to take away our overtime pay because, and I quote, 'it's not fair for you to get it and I don't'
|
# ? Apr 17, 2024 07:33 |
|
Cyrano4747 posted:It's gimped by even LLM standards and is a joke to boot. I've been saying at work for a while that "AI" is massively overrated and overhyped, and I just tried this. Not with guns and battles, but a subject I actually know something about, and it was amazing. Thanks, I will be suggesting this to everyone I know who thinks AI is anything more than a clever looking LLM.
|
# ? Apr 17, 2024 11:06 |
|
Scientastic posted:I've been saying at work for a while that "AI" is massively overrated and overhyped, and I just tried this. Not with guns and battles, but a subject I actually know something about, and it was amazing. Thanks, I will be suggesting this to everyone I know who thinks AI is anything more than a clever looking LLM. There's an old Charles Babbage quote that I recently realized is far more relevant to LLMs than it should be: Charles Babbage posted:On two occasions I have been asked [by members of Parliament], — "Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?" ...I am not able rightly to apprehend the kind of confusion of ideas that could provoke such a question. And that's what we're seeing with with ChatGPT, Copilot, Bard, and all the other generative AI bullshit. They're full of wrong figures but people think they'll magically give the right answers anyway.
|
# ? Apr 17, 2024 11:33 |
|
LLMs, and ChatGPT 4 particularly, are proving very useful for programming assistance - you just have to know exactly what you want, how to ask the right questions, and how to validate what is produced. They are great tools for boosting the productivity of good developers, but they are just a tool.
|
# ? Apr 17, 2024 11:51 |
|
Pikavangelist posted:There's an old Charles Babbage quote that I recently realized is far more relevant to LLMs than it should be: But all the noted confidence men and grifters swear it's real! Why would all the people with a vested financial interest in LLMs success lie shamelessly about its capabilities
|
# ? Apr 17, 2024 12:57 |
|
TheBlackVegetable posted:LLMs, and ChatGPT 4 particularly, are proving very useful for programming assistance - you just have to know exactly what you want, how to ask the right questions, and how to validate what is produced. I don't know how it works for programmers, but I've found the errors it makes in other areas can be pretty subtle. I've got this big 'ol quarterly presentation that I need to put together four times a year. Powerpoint deck from hell, 100+ slides. It takes me about a month to get together between getting data submitted from people who always put it off to generating the summary slides that I need to. At one point I got permission to try using copilot to streamline this process since it's not sensitive data. It made the summary slides I needed just fine, but it wasn't until I really dug into one that I found it just made some poo poo the gently caress right up. Like, if Department A reports that they did $500,000 in sales this quarter and Department B reports that they did $400k and C reports $250k, the summary graph included a correct bar for Department A, a bar for Department AB (literally a mash up of the names) that did a random amount of sales, and a Department D that sounds like a real department but which we don't have that did $made_up_bullshit in sales. It looked OK at a glance but when you started to read it it fell apart. Christ I'm glad I didn't send that bullshit out to the people who actually need the report. I basically had to go through and re-check everything by hand. It was loving pointless. Maybe it will get better as a tool. Maybe in ten years it will be invaluable to my work flow. But right now everyone is jumping on the hype train when it's basically an alpha test of a potential future product. It's like getting in a time machine and breathlessly telling everyone they need to adopt cell phones asap and explaining Uber Eats and internet porn in your pocket . . . in the mid-80s when they're backpack sized things that only ostentatiously rich assholes have in their car. Even if the evangelists are right it's loving worthless today and we've got a few decades of iteration.
|
# ? Apr 17, 2024 12:57 |
|
Yeah that’s what our management team found when they tried to use copilot to do their analysis poo poo. They asked it to summarise a table of data and it spat out a bunch of bullshit referencing numbers that didn’t exist and even directly prompting it to refer to some of the specific values resulted in more absolute nonsense. Redoing the same query across multiple tests also came back with conflicting analysis of the exact same inputs lol. Total garbage.
|
# ? Apr 17, 2024 13:01 |
|
They told our sales team in the field to use chatGPT to write emails a week ago. Now I'm doing the compliance audit on sales emails and I turned off chatGPT because Christ on a bike I'm dumb as gently caress but this makes me look like a big brain genius comparatively. FINRA will have a field day if they catch wind. I might let them. I'm so angry already this week, Jesus Christ.
|
# ? Apr 17, 2024 13:11 |
|
I hate this gay earth (Please note, I don't work at a software company. hell, we're not even an Apple shop)
|
# ? Apr 17, 2024 16:17 |
|
Randy Travesty posted:They told our sales team in the field to use chatGPT to write emails a week ago. The typical person making up free form communications for a financial institution has a pretty terrible grasp of English. So, I sort of get why you would want to use an AI to turn their gibberish into something professional. But the problem is that they don't really read either, so any errors the AI makes won't be fixed. Which is actually even higher risk. ChatGPT did help a friend of mine who manages a group of blue collar workers in state government. HR came to him, told him they suspected some of his employees of making GBS threads on the bathroom floors (but couldn't quite prove it). They asked him to send out a general email about it. Because he couldn't figure out a way of writing an email that wasn't just "Hey, if you're making GBS threads on the bathroom floor instead of the toilet, stop it," he got Chat GPT to write it for him. It had the same zero effect that you would expect, but it made him look good.
|
# ? Apr 17, 2024 16:27 |
|
I would have gone with "We are tremendously impressed with the poops we are finding on the bathroom floor, HR would like to give this anonymous hero an award so please go to them ASAP."
|
# ? Apr 17, 2024 16:33 |
|
|
# ? May 21, 2024 18:04 |
|
Outrail posted:Kinda related: how do you explain to someone that developing a bespoke budget tracking and project management system from scratch while also realigning your bookkeeping system to work with the budget/project system isn't something that just happens? "Manually tracked on a case by case basis using ad-hoc unretained documentation" sounds a bit better though.
|
# ? Apr 17, 2024 16:47 |