How many of us got their workflow disrupted by AI? It's been a while since I've googled a "how do I..." type thing, because of how much I appreciate Bing Copilot or Phind
Safe to say my life did not change with the arrival of Web3!
Single digit percentage?
For sure more than a single-digit percentage of people have tried a Q&A bot by now. Shoot, Windows just updated to insert Copilot into users' machines, with the icon right next to the Windows button. It can't not continue to grow. It's such an amazingly effective way to get information versus scouring through web pages.
Tried is doing a lot of heavy lifting. https://meow.social/@crumbcake/113156927685392932
> Shoot, Windows just updated to insert Copilot into users' machines, with the icon right next to the Windows button.
This goes back to my point about consent and opt-in vs opt-out.
> It can't not continue to grow.
"If we shove this down people's throats enough maybe their gag reflex will adapt."
> It's such an amazingly effective way to get information versus scouring through web pages.
The only reason I stopped using Google isn't to choose AI instead. It's because the obtrusive "AI Overview" pissed me off.
Some recent examples:
* "I installed windows and now I can't access my linux partition" * "I want to add an `in` variable to an existing pl/sql function, can it go on the end of the parameter list? (Answer: No, because...)" * "What size soccer ball does AYSO U10 use?" * "what is the name of that python library that handles command line arguments, and could you provide an example that demonstrates required flags?"
nothing earth shaking. not things that're impossible to find, but... instead of reading someone's blog post in which they work as hard as possible to belabor the point so you'll scroll through all 14... this thing just gives you the answer.
Yep, this was one of those tries:
Me: "I was billed twice for the same storage unit"
Highly Intelligent Chat Bot: "Let me check on that...you have two storage units so you receive 2 charges each month"
Me: "I was charged two times in the same day for unit #X"
HICB: "Unit #X is charged Y amount on date Z each month"
Me: "I received two charges for amount Y on the same day for unit #X"
HICB: "That's correct, you have two units and get charged Y amount for each one. Please let me know if you have any other questions"
Me: "Is it possible for you to be any less helpful?"
There being negative cases don't mean the prolific amount of positive cases don't tell me the world has shifted.
Seriously I was very skeptical of using ChatGPT, worried it'd give me bad advice but I keep not finding quality content in google searches ("Hey googz, what's the flag in a bash if statement that verifies a file does not exist? --> oh here's a page full of every possible thing bash can do") and i ask ChatGPT and it's just bam, the answer.
It helped me recover a crashed harddrive, verify how `in` and `out` variables work in pl/sql functions ("yes, all `in`s have to come before any `out`s")... it's just so right so concisely that, like it or not, it's here to stay.
There are some reports of people hired to write low quality spam articles that fit this. There isn't much else.
There exist lots of people that claim some small gains, mostly because they use shitty tools that can't do what some tools could do 15 years ago (often the same tools). I'm afraid that's not enough to keep our current LLMs alive.
Yep, a large part is Google. But the most popular IDEs seem to be less capable of providing useful information nowadays too. (Yet, there exist good ones out there, and they haven't been explicitly enshitified, it's the languages and ecosystems that changed.)
Realistically how many people are currently employed "writing low quality spam articles"
That's not reality. That's just reddit crapping on strawmen because a lot of the visible results from AI is 1) Crap creative work and 2) Polluted search results/comment sections
These are red herrings
There's real people doing real productive things with AIs that do boring things right now.
The ability to search for things that are still fuzzy in my mind, ask open-ended questions and get an answer back, or find something when you can't come up with a good keyword
I've never had the google-fu that some other developers have so this was something that often required asking a peer before. Sometimes a forum or some Discord, even.
Now you just feed your garbage question to a large language model that swallowed the entirety of the internet and a few seconds later it's spitting back something useful.
It's the first time I have access to a robot that's on stand-by with surface level knowledge of everything.
I suspect people who don't find this extremely valuable in their day-to-day life are just not catching the moments when they could've used it -- while on the other side of the spectrum the value it removes from comment sections/image search results is impossible to ignore.
Up until LLMs started eating their lunch, a lot. I knew several people who wrote softball slop articles for no name tech blogs. Lots of generic how-tos in basic Java or Python questions, etc.
Now that's all automated and its mostly slop being used by AI to generate more slop.
> There's real people doing real productive things with AIs that do boring things right now.
elaborate.
I can think of at least a couple, but most of their approaches are only "AI" in a marketing sense, like Moderna using data driven, highly automated testing to create vaccine candidates. Absolutely nothing to do with LLMs as they exist in most places.
> Now you just feed your garbage question to a large language model that swallowed the entirety of the internet and a few seconds later it's spitting back something useful.
citation needed. I'm about 4 for 8 in terms of real technical questions and real, correct answers. Feeding Copilot and ChatGPT some questions about CVEs exploited in the wild got me wildly wrong answers until I started doing a lot of prompt work, but by that point I could have just googled the CVEs myself and got an answer I knew was accurate.
Ditto with fixing some MongoDB calls that look correct but didn't work, and took me longer to understand than just trying random shit, or asking on reddit.
Maybe it will get there, eventually, but outside of mass content generation I can't see a reason to trust the results.
My guess is the GP is talking about things like Perplexity. What is indeed ChatGPT, but it's set so that it can point you to useful sites instead of just giving you an answer. (I don't know why so many people think they answer itself will be useful, it can only be good if it was taken verbatim from some site.)
And yeah, those things are useful. But to say they'll disrupt one's workflow is a bit too much.
I did about an hour ago. AI answers run the gamut from boilerplate that's useful that still ends up needing tweaking to utterly making shit up that doesn't exist. I did a copilot search once that referred to an entire framework that it just made the fuck up, and you know it would've been great if it did exist, the framework would've suited my needs perfectly, it was just complicated by the fact that it didn't.
I'm off that train, I hate having my time wasted and until they can guarantee veracity in answers it's just a google search that's even less reliable than google, which is saying something lately.
I've asked it pretty niche questions and I've never seen it make up anything, much less a whole framework
Plus it gives you its sources so I don't see how this could've happened. A whole framework?
I dunno maybe that would help, but it seems like a deeply un-serious product then.
If you're asking it to write a story about a camel in a hat you probably shouldn't choose "precise" mode
Bing Copilot is your old regular ChatGPT applied to searching the web.
* Generating Bing queries based on what you're asking
* Reading the web results
* Giving you an answer based on those (highlighting in the answer where it got the facts from and linking to it latex-style at the end)
It has a Creative/Balanced/Precise slider to inform how close it should stay to the info it just read online
"Copilot" alone is the coding companion that's integrated in your IDE
It does not have that slider
Same, but you’re missing planet-sized elephant in the room here. This use-case has been technically solved for 20 years. What’s happened is that web search was ripe for consumer value extraction (aka enshittification). Have you ever seen a market more saturated with perverse incentives, from ad delivery, content farms, data mining, and of course the very search engine that prints money from it?
Now, if you wanna do an apples-apples comparison, imagine popups, consent forms, hidden influences from advertisers, recommendation engines, arbitrary lock-outs from community guidelines violations, ads while waiting for responses, JS bloat at recipe-website levels, and so on and so forth.
By all means, enjoy ChatGPT or whatever in its naked current form. But make no mistake that this is the honeymoon phase. Potential doesn’t govern the direction of new tech. Incentive does.