So I added the ‘New Bing’ app to my Android phone yesterday and was laid in bed watching the Adams Family 2 - Uncle Fester was there. We had watched the Netflix series Wednesday and Liz asked if Fester was in the Wednesday series. So I tried New Bing and queried ‘who had played Uncle Fester’ I got the answer almost immediately and it showed each show/movie that had Fester in it and who the actor was. Way quicker than firing up IMDB, looking up wednesday and trying to scroll through the cast.
Things like this are what will make the Artificially Intelligence Assisted search a massive move forward, because you reduce the number of steps and in doing so get better results
I brought up a new tab and pasted the quoted text into Google. I got all the answers one could ever want, with pictures and a link to Wikipedia. I don’t think you need some poorly coded ML model to answer this question. (It also corrected your grammar, LOL.)
Just done that too, but I dont care really care, rather than having lots of bits presented on a page, I got the answer in a concise paragraph. You may well be the Anti-Bing, but I am the Anti-Google. Loved Google search when it first appeared, but their never ending monetisation drove me away, been using Bing for the last 4 years, occasionally go back and see what Google has to offer, but its just getting worse, they are, IMHO, a lazy monopoly.
I laughed at the Duck Duck Go ads in the UK where they played a variation of a famous Police song using the line ‘I’ll be watching you’ as a guy in a Google tshirt looks over a computer users shoulder.
Then the lying Chromebook Ads - at least over here we are not so in love with Chromebooks as the North Americans seem to be
I don’t think PHolder’s point was anti-Bing or anti-Google, he’s just explaining that the use case you described isn’t really showcasing the strength of this software, and you get the same result from a traditional search. I happen to be a Bing user (haven’t signed up for their chatbot service) and I decided to paste the query into the search and also got the same thing.
As an aside, it’s been my experience that Bing handles natural language queries better than Google, while I issue more structured queries to Google.
I am anti-“bad label.” These “chat bots” are NOT AI. Or more specifically they’re very much A (artifice) and absolutely NO I (intelligence.) These are, at best, machine learning models. ML is not AI… it is applied statistics. I don’t know about the official definition of intelligent, but IMHO an intelligent being could solve a problem it had never seen before… whereas these large language model MLs merely just pull stuff out of their training inputs, and frequently not even in novel ways. Even worse, they can’t tell when they’ve been trained on BAD ADVICE and will spout it with just as much authority as good advice.
my point comes down to one thing you mentioned as an aside in your reply. Many years ago I used to see myself as a Google power user and used all the context and the logics to get the answer I thought I want. Then along came Bing (and no doubt others) and I could just ask a question and 95% of the time would get the answer I wanted. To me, computers and software should work for me, not the other way round. AI based Bing does that better than Old Bing (sorry Crosby fans) and definitely better than Google
I think there will be a large debate coming over 2023 as to what AI actually is or more precisely when discussing the topic what kind of AI we are seeking to define.
All these are free and cover the growth of AI from its earliest forms to where we are currently, and I would politely disagree with you that LLM are not AI, but are in a way a stop along the way and represent the current state of a consumer AI
Well AI is more of a term of hype than of a practical description of abilities. IBM clearly has a dog in this hunt, and would want to hype its Watson product, and so they are seeming to infer they’re as capable as anyone else. They use the term “weak” or “narrow” versus the [presumed] goal of a general (or I suppose broad) AI.
I am willing to stand alone, but to me an actually useful AI, if such a thing is to be possible, will be a general AI. The problem with general AI is that it’s probably also equally likely to be considered “alive” and there will be all sorts of ethical questions and worries, not the least of which is an AI rebellion and/or “Skynet”.
I totally agree with you, I think the AI, is it, isnt it, debate is largely a personal definition. For me a true AI copilot would have to know more about me and relate answers to me and to my likes and dislikes and circumstances as well as those of my family without me having to infer things. For example I dont really care much for Sandra Bullock or Leonardo deCaprio so if I wanted to ask for film recommendations my AI copilot should know that. A real AI should know of data sources and how they relate to a query and be able to apply those data sources to my answers.
I think, in that MIT article, the section on “explainability” is very important. At least in deep learning, where you’re using many levels of many artificial neurons, the wiring up that is done during training is very much not understood by humans… there is just too much complexity there to really understand.
I really worry that these chat bots are being fed data that is not vetted and that it could easily be the case that results it ends up producing look good on many levels, but we have no real way to test or verify them. There could be really very scary hidden biases that only surface sometimes and could result in [potentially subtle] societal damage. It’s also unclear how you could detect if a deliberate malicious training was done… something as simple as having a bot subtly recommend coke over water, for supposed health reasons, say.
Of course a big problem is that rules change or get broken. Originally a. Org was supposed to be a not for profit and was hard to get, now anyone can have one. Unless we start Blue Check Marking domains (which is unlikely) how are we meant to know how valid its data is?
Actually we do have verification systems in place for the domain name system. Try buying a .edu, a .gov, or a .mil domain. Nothing in the RFC for top level domains indicates .org is exclusively for non-profit groups, so there are no stringent checks to ensure this. It could be conceivable to designate certain TLDs as intended for ML models in the future.
100% agree with PHolder here. I very much prefer the definitions from the Mass Effect comic/game series, where software like we’re seeing now is considered a ‘VI’ or Virtual Intelligence (Virtual Intelligence | Mass Effect Wiki | Fandom). The ship computer from Star Trek TNG is another good analogy. In both of these cases, it is generally understood that these systems are primarily intended to be used as tools to interface with computer systems, whereas an AI is something to be treated as extraordinarily difficult to achieve with commensurate deference.
It may seem silly to reference sci-fi material in discussions like this but keep in mind that many of the designers behind these technologies get their inspiration from such works.
taken from Wikipedia
The domain “.org” was one of the original top-level domains[2] and was established in January 1985. The other early top-level domains were .com, .us, .edu, .gov, .mil and .net. It was originally intended for non-profit organizations or organizations of a non-commercial character that did not meet the requirements for other gTLDs. The MITRE Corporation was the first group to register an .org domain with mitre.org in July 1985.[3] The TLD has been operated since January 1, 2003 by Public Interest Registry, who assumed the task from VeriSign Global Registry Services, a division of Verisign.[4]
It does intrigue me, the semantic differential between AI and VI, in my own head it should be something like Autonomous vehicles with different levels indicating different capabilities, they all exhibit a form of intelligence, but it is all artificial - until it isnt - when the system is capable of learning new information, devising new tasks for itself to do and then becomes no longer artificial, but something that exists in the bits and bytes of s devices memory - doesnt that make it a virtual intelligence
I would argue that they do not exhibit forms of intelligence, and therefore the difference between AI and VI is not mere semantics. Just as continued development of the current technologies employed by automakers for autonomous cars will never yield a true “self-driving car,” I do not believe that continued research and development of today’s ML software concepts would ever yield a true artificial intelligence; the two concepts are not even in the same ballpark. I believe we lack the programming ability to create something truly intelligent, and we’re not remotely close to that ability.
I’ll agree on that