And that's not even touching on data privacy or plagerism issues. Do you really want your entire thought process accessible, co-opted, and owned by a giant corporation with questionable ethics? Are you ok giving up the protections normally afforded to published works in order to have this ability?
Thank you for engaging constructively. 1) Are LLMs “great”? No, I specifically said “getting better”. More than a billion people use AI chatbots regularly. That’s not an argument that they are great, but clearly a lot of people find them useful (so I don’t think my “peer pressuring” is needed).
Should we “outsource our thinking”? Definitely not. LLMs still need oversight and careful review. But that’s not different to a junior employee. Data privacy? Sure, there are questions about lawful basis of training LLMs with publicly available personal information.
But that’s not too dissimilar to search results / crawling (and we are all using search engines despite that issue). You are right, the IP question is still not resolved with a lot of litigation ongoing.
Search engines aren't answering my queries the same way. Assessing whether or not a source is accurate/exact/specific enough to what I want to try to determine, or seeing that all the results are unreliable/commercialized/limited, is thinking that is being outsourced, imo.
The search engine point was regarding the legitimate basis for processing personal information, not regarding accuracy.
Boolean keywords reveal a lot less information about your thoughts than conversational queries.
So you would replace a junior employee with an LLM? Then where do new senior employees who know how to use the machine effectively come from? Or will those also eventually be replaced with LLMs? Seems to me your magic box results in humans being pushed out.
The industry shoving AI down our throats wouldn't be necessary if it were genuinely helpful.
A lot of people find illegal narcotics useful too.
No, it is a sign that propagandizing has worked. As a software engineer, I find them to be utterly useless. Even for technical guides, it sorts items into categories that aren't helpful, duplicates entries, and pads the writing with unnecessary verbiage.
It has not once caught something I neglected to enter because it's predictive, not because it actually knows anything in the sense that I would as an engineer on the project. It's guessing, and it only has success because enough guesses are for items simplistic enough that it cannot miss.
I have never tried it for use cases, so can't judge. All I am saying that's it's tremendously useful for me.
Chatbots are literally killing people. Fuck all the way off with that bullshit.
"More than a billion people use them regularly... clearly a loy of people find them useful." A lot of those users are relying on ChatGPT to do their homework for them, or as a therapist, or to provide factual verification where it cannot. We've broken our leg so we can use a defective crutch.
The difference between these LLMs and a junior employee is that users are assuming LLMs answer from a place of authority and accuracy when they do not.
I love reverse image search, it is very useful, and most people don't think of it as ai-but when it further makes the determination if an image is accurate, you sometimes end up with serious cases of mistaken facial recognition identity. Chatbots have some very shitty consequences for some people.
Or in the case of cars (esp that only use cameras to auto-drive), you get some number of crashes because the image recognition did not happen fast enough. Cases were people die, or lose their liberty or sanity, need to be solved before public consumption or widespread adoption.
*h, where
*accurate enough
If you've made specific tools, for limited use cases, like an application dedicated to a task that uses a predictive model, sell us on the ways these tools are useful.
Telling people "you guys, ai is really good now, trust me, I work in ai" without any clear cut examples (except that super nuanced editing work flow) is definitely the ticket.