You can rest assured these actions will be safe...
and then the world ends, stops spinning and blows up
OpenAI this week introduced ChatGPT plugins, a way to extend the scope of its chatbot language model beyond the slurry of internet training data to bespoke business information. So wary is OpenAI of all the ways that ChatGPT and its other models can misfire that the company begins its announcement by reassuring readers that …
As for injecting ads based on keywords this is surely a non-problem for such a system. If you can ask it for a recipe and ingredients order, throwing a relevant ad retrieval call in the mix of calls is comparatively trivial, no?
Really though, this chatGPT thing can write python on-the-fly - it could code it's own ad call I think.
When I do a web search for 'Nelson' the first four hits are for Hotels, Hotels, University of Northampton(Really such a thing?) and BBC weather in Nelson. At last we get to Horatio Nelson. I expect if you type in Python you'll find something similar. If the consumers are interested in hotels and weather and discovering universities in unexpected places then that's fine. But when I want non-trending, non-dumbed-down stuff then either I have to know where to do a curated search or wade through 'Ten things you didn't know about Northampton.'
So how is ChatGPT different? Won't it be another race to the bottom?
It's quite good at bits of tech I know about but hardly ever use. eg Write a bash script to ... which does the fiddly bits. DALL-E illustrations are really handy. So long as I'm relatively clear what I want the system can 'understand' me and offer choices. But how will 'recreational' users who have vague ideas, limited literacy and just love achievement through spending money skew results. For example if BP or Shell spends a lot of money in convincing ChatGPT that they're greener than an Irish cauliflower then that PR slant will dominate.
While these models were in the lab they didn't get too much attention from public relations teams gaming the system. Now it appears YOUR MESSAGE HERE is the order of the day. Wikipedia has feedback mechanisms but there's nothing similar so far in AI. The hate still gets through in Facebook. Systems which are trained to give a certain message will be here tomorrow masquerading as sparkly hi-tech which is so much safer.
is this even a real question? It's going to be the usual race to the top, top of the revenue stream, that is. And it will be brutal get there quick, before the states and legal systems start catching up and placing restrictions on our innovative, disruptive revenue generation approach.
"It's quite good at bits of tech I know about but hardly ever use. eg Write a bash script to ... which does the fiddly bits. DALL-E illustrations are really handy. So long as I'm relatively clear what I want the system can 'understand' me and offer choices."
Exactly this. I've been saying for weeks that these things are great time savers if used with the understanding that it might return utter cobblers. Unleashing them on the masses in every instance of Office is asking for trouble.
Maybe ad-biassed results won't be such a problem though. It looks like chatGPT, Midjourney etc are having some success with getting users to actually pay to use these things. If you have to pay to use, and pay to make plugins, and pay for access to the API, maybe there will be less room for sponsorship. Of course, that still leaves Google, but by all accounts their bard thingy is well behind the competition. Google gained search dominance by initially being better than the competition. If a pay for service is better than Google, they might find themselves struggling. Probably an overly optimistic view, but you never know :)
...will be Microsoft opting you into sharing all your data within your Windows / MS cloud ecosystem. Win11 already requires enough comupting power that they should be able to train on you data locally before uploading relevant (marketing) information. Suitably anonymized, of course. Except if you reside in the EU, where they'll require proper consent and anonymization, somewhere in the 2030s.
But apart from my cynical view of the MS/ChatGPT cooperation, do you guys think the new AI engines represent a new user interface to information (+(natural language), +(compilation of multiple sources), -(reliability / reproducibility)) or does it acutally 'generate' knowledge? It seem like everybody assumes the second, but I am somewhat sceptical.
well, define 'generate'. If 'summarising' something for the first time is 'generating' , then yes, it generates a 'new value', and you might argue that 'new value' = 'knowledge'. All that talk about an 'eureka' moment, it's (only?) a combination of enough data points to come to that eureka conclusion. Maybe.
... trying to find an analogy, it'll be probably flawed, but there it goes: I'm learning a foreign language (a real case), learning various bits of it: meaning of words through various means, repetition, media, etc. Pronunciation, syntax, grammar, context, narrow and wide (all through various means, irrelevant here), and at some point, I take off, like a lame duck, using that new language, which I have learnt, with each usage, getting better and better. Nothing here is 'creative' per se, just a huge number of data points put together and applied to some learnt rules of how to apply them together. But now I have knowledge I didn't have before, and while I might confuse common meaning of 'knowledge' with something lesser, i.e. 'skill', in the long run, this skill of knowing a foreign language lets me access knowledge I wasn't able to access before so, at the end of the day, I do gain 'knowledge' in the wider sense.
It's not proper AI, I can't ask it to "shoot the breeze" with consideration to my beliefs, knowledge and general personality, it can't see me and make any decisions, it doesn't know anything about me ( yet! ) and so unlike someone I can just chat to at the bus stop all CrapGPT can spew out is mindless, cobbled together answers to questions it's able to make sense of. Useful sure but it's not suddenly going to solve every problem in existence in the next 4 months like all the morons writing for the broadsheets seem be thinking.
Bag a shite!