tailored to their corporation’s needs
No these models are out there now. The more specialized, the less likely you need huge computing or graphical power. The corporations have realized bigger isn’t better.
I think you did not get the point.
Do I not? People are framing this as a bad thing, and it is, but not the way people think. The smart corporations won’t hoard away the tech to only do what they want with it. The smart companies will use the users as the product and harvest incredible amounts of information. Not only do you get user information, marketing data, human data, you get people to create models for free. Sure they get some use out of it for whatever project, but they might have just cooked up a million dollar idea that they can leverage and steal as well. I think “AI” is going to be a lot less scary when it’s integrated everywhere and user dependence is the biggest problem.
Yep, makes sense. You don’t want a researcher using the same tool as a lawyer or fiction writer. The researcher needs AI to summarize existing literate in a factual way, the lawyer needs to source actual cases, while the writer needs novel combinations of existing literary ideas. A single tool isn’t going to meet all those needs.
A single tool isn’t going to meet all those needs yet
Why would we want one? We don’t have a single social media tool: forums, link aggregators, micro blogging, networking, etc. are all separate tools. We wouldn’t want to do all of those on Facebook.
ChatGPT is just a demo of a technology that can be used for all sorts of cool things. Trying to make ChatGPT do it all isn’t really needed nor desirable.
I do think it’s desirable. It’s unnecessary for users to keep track of which tool is best for which purpose if one tool can do it all. There’s no reason why one tool wouldn’t be able to; even in the worst case it could just automatically choose the best tool to answer your prompt, saving you the trouble of doing so.
The tools would be integrated into things we already use.
I’m a doctor, and our EMR is planning to start piloting generative text for replies to patient messages later this year. These would be fairly informal and don’t need to be super medically rigorous, needing just a quick review to make sure the AI doesn’t give dangerous advice.
However, at some point AI may be used in clinical support, where it may offer suggestions on diagnoses, tests, and/or medications. And here, we would need a much higher standard of evidence and reproducibility of results as relying on a bad medical decision here could lead to serious harm.
These are already in two different sections of the medical chart (inbox vs. encounter, respectively) and these would likely be two separate tools with two separate contexts. I would not need to memorize two tools to use the software: in my inbox, I’ll have my inbox tools, and in my encounter, I’ll have my encounter tools, without worrying about exactly what AI implementation is being used in the background.
That’s literally their point. You have a specialised tool for each.
Being general makes them much harder to train and worse at each individual task.
And they meant that in the further future, even that might stop being a problem.
My main issue with using the general chatbot is that it’s an incredibly inefficient way to convey information. For writing tasks I essentially need to type most of the answer first to get reasonable outputs when considering my actual constraints.
More specialized tooling will have these constraints built-in, which will increase productivity.
Even if we have the perfect general chatbot, it’s still a lot of work to concisely describe your requirements to it.
Chatbots are only useful for displaying FAQs answers. In all other situations getting in contact with a real human is the only way to attempt to get a resolution
This is where I think AI has the power to really increase productivity. General-purpose AIs are nice, but having one that knows the current status of projects, learns the people you talk to frequently and your tone, and has access to all your company’s internal documentation would
Instead of digging around in Confluence for a document, you just ask the AI. You ask the AI what meetings you have and what they’re about. You ask the AI to write an email to Joe about the contract renewal and it spits one out for you to proofread and send.
It would be like everyone having their own personal secretary, but one that works seamlessly with everyone else’s and never takes a sick day.
I imagine there will eventually be businesses that aggregate data specifically to sell to LLM businesses. Like photostock but with a bunch of LLM conversational stuff.
