Google's $100b bad day demo may be worth the price "The pleasure is fleeting, the position ridiculous, and the expense damnable." Famously about sex but more probably about golf, this quote is now most accurately ascribable to AI-enhanced search engines. Google proved this last week when a wrong answer from its rival to Microsoft's ChatGPT cost Alphabet a cool $100 billion on …
I just asked ChatGPT
"Can you write me a short comment for a theregister.co.uk forum in the style of bombastic Bob"
...
Certainly! Here's a short comment in the style of Bombastic Bob:
"Greetings, fellow forum denizens! I must say, this article has left me utterly flabbergasted. The sheer audacity of these tech titans never ceases to amaze me. We must not let them trample upon our digital rights with impunity! Let us rise up and demand accountability from these corporate overlords! Power to the people!"
Before you downvote me: consider that that will just get fed back into the AI.
N.B. ChatGPT has thankfully NOTGOT CAPSLOCK on
Yet
That being said, you can keep pointing out what it gets wrong and eventually it spits out something workable.
Even with the bollocks included, it still saves a lot of time if you know how to nudge it.
So it's good for answering questions as long as you already know the answer?
That does sound strangely like something Microsoft would produce...
it still saves a lot of time if you know how to nudge it
Does it? I have my doubts. For one thing, I almost never spend a lot of time searching for information online; if what I want to know is that obscure, there are much better ways to pursue research. So there isn't "a lot of time" to be saved.
More importantly, why would I want to pursue learned helplessness? Researching around a topic is a good way to gain a broader understanding, construct a more reliable map of the limits of your knowledge, serendipitously learn something new, and improve retention.
It's easy to see the deleterious effect online resources have had in software development, for example, where you have a million numpties copying code fragments they don't understand from StackOverflow. Or similarly in many other fields. Carr may have overstated the case in The Shallows (or he may not have), but certainly stampeding in the direction of Moar Stupid doesn't seem like a great idea. Maybe instead of playing with chatbots folks should try a bit of thinking and learning. I know it's work, but – hear me out – work is not inherently bad.
Faster isn't better; faster is just faster. Easier isn't better; easier is just easier. There are times when the cost of faster or easier is less than the benefit, but that's not always the case.
YES, exactly. I took a test drive and found the same thing: grammatically correct complete nonsense. Like watching daytime TV. Supposedly to get 'good' answers the app needs to be connected to a variety of data bases. But, how is that different than, say, the google data base?
As far as I can recall from reading their Terms, they can. Technically they only have the rights to someone's 'sent' folder (as only that is the contracting entity's own) but I doubt they'll restrict themselves to that - the key reason why a business using Gmail to communicate with customers in Europe ought to be de facto considered in breach of GDPR.
even just access to the sent folder is on dodgy ground since most emails are replies with the other peoples emails quoted in the chain, usually in full. Googles et al "out" is that they are not "read" by humans, just analysed by computers (or AI as they like to call it these days)
I've never considered that a plausible excuse as it attempts to skip over the actual offence: you're accessing someone's data without their explicit permission. Yes, explicit: in Europe, you have to specify at the point where you ask permission what you're up to, and not bury it at the end of a TB worth of Terms so people don't realise it. What you use for that and for what purpose matters little at that point, the offence has been committed..
Also, it still doesn't address accessing incoming mail, permission for which cannot be assumed unless it happens to be another Gmail victim user.
There is something the Bing and Google staffers can do to improve their products. It's a trick many of us have learned, just append "Reddit" to the search term. If you're feeling particularly intent, site:redddit.com really drives the hint home.
This. Although giving us an option to set "verbatim" mode on by default would also be great, so we could go back to the days of, you know, actually searching for the keywords I typed in.
I feel that when I use a search engine, I use it to get suitable (hopefully) suggestions related to the search terms I've entered and listing them so that I can see the relevance and the source so I can make my own judgement as to which links to explore further and which to avoid like the plague. I'm not searching for philosophical discussion about the search item or any other bollocks. If that was what I wanted I would start with Wikipedia and search from there, not use a generic search engine. A search engine's job is to aid navigation and filtering of web content based on your query. The only AI it needs is not to show me pages that are only obliquely linked to the search terms, or that are in Polish or Chinese when I've selected UK as a search criteria or controlling filter.
Thank you Rupert, I've been saying this for weeks.
I don't want to quite jump on the Butlerian Jihad just yet, but AI training sets are a toxic mess of rabbit holes and, unless the results are demonstrably 100% accurate, they aren't worth having at all, because you will still have to do a traditional search to double check what the AI told you.
Once they are 100% accurate, people will stop with critical thinking and going, "hey, wait a minute, that doesn't sound right!" They'll just blindly take it as read that the AI provided information is correct. Idiocracy, here we come!
We have enough problems with many people believing every conspiracy theory they trip over, the wilder the better! Scale that behaviour up into people believing AI provided "facts" without question and we soon have a serious problem, where most people wouldn't even know how to question the facts and go and find the real answer.
A don't get me started on it being trained on copyrighted information - this is fine for a lab experiment, with a limited audience, but if it is being used for commercial gain, there are all sorts of wrong that need to be sorted out first. Heck, look in any direction and you'll trip over dozens of things that need checks and balances put in place to ensure things remain sane.
"contains 50 percent less irrelevant content than eight years ago"
Is that because they have censored people like Alex Berenson, at the behest of the US government? Because presumably if they had to censor the content, it must be "irrelevant"? I guess if like most people you aren't on Twitter and been reading The Twitter Files, then you don't even know about this? Sharing for information.
BTW, I'm not saying I particularly agree or disagree with Alex Berenson about anything, I just find the whole idea of censorship to be totally abhorrent, and I prefer open and honest debate. Particularly when it's a prominent award winning journalist.
Presently Seymour Hersh*, prominent award winning journalist, is also suffering the outrageous slings and arrows of wild wacky western censorship regarding his forensic investigation of an act of wilful economic and industrial sabotage which has everything but the required desired consequences on the principal party attacked ......How America Took Out The Nord Stream Pipeline
How long do you think it is going to be before folk/establishments/cabals/renegade rogues and rapscallions, and others who really should know better, realise that harbouring and denying the truth of self-destructive secrets is no longer possible and is toxic and harmful to one’s health and the health and safety and security of all of one’s cointerdependent friends, such as may very well be the likes of allies in a NATO?
Keeping nations ignorant and unaware of smarter available intelligence which leads with honest information is a monumental folly only followed by colossal fools on a titanic mission to nowhere great and good ....... which is a scourge well worthy of all manner and means of purge and annihilation, and/or re-education if possible, or desirable.
*. ....... https://en.wikipedia.org/wiki/Seymour_Hersh
Keeping nations ignorant and unaware of smarter available intelligence which leads with honest information is a monumental folly only followed by colossal fools on a titanic mission to nowhere great and good
Yep, but that's showbusiness and politics in an unholy alliance. Are we not entertained? Given the economic impact of the NordStream.. malfunction, it's suprising that our governments don't seem that interested to determine the truth. But then the truth is what it needs to be. The MH17 incident has been investigated for years to prove it was Russia. When Syria crossed red lines and used chemical weapons, it was all tantrums and Tomahawks. When video emerges of drones dropping <something> on Russians, it's crickets. Probably powdered. Let them eat bugs!
Truth just isn't what it used to be, and we seem intent on making it worse by our obsession with creating fake news.
Hmm, I'd never heard of Alex Berenson but in google the top result is this which is a piece of fake news.
https://www.realclearinvestigations.com/articles/2023/02/08/investigative_issues_dr_fauci_now_admits_the_mrna_covid_vaccines_hardly_work_and_might_not_be_approvable_880586.html
Well this is what Dr Fauci et al actually wrote: https://www.cell.com/cell-host-microbe/fulltext/S1931-3128(22)00572-8 (note he is the correspondence - ie lead - author).
note this particular paragraph:
During the COVID-19 pandemic, the rapid development and deployment of SARS-CoV-2 vaccines has saved innumerable lives and helped to achieve early partial pandemic control. However, as variant SARS-CoV-2 strains have emerged, deficiencies in these vaccines reminiscent of influenza vaccines have become apparent. The vaccines for these two very different viruses have common characteristics: they elicit incomplete and short-lived protection against evolving virus variants that escape population immunity. Considering that vaccine development and licensure is a long and complex process requiring years of preclinical and clinical safety and efficacy data, the limitations of influenza and SARS-CoV-2 vaccines remind us that candidate vaccines for most other respiratory viruses have to date been insufficiently protective for consideration of licensure, including candidate vaccines against RSV, a major killer of infants and the elderly, parainfluenzaviruses, endemic coronaviruses, and many other “common cold” viruses that cause significant morbidity and economic loss.
So the title to the 'fake news' is:
"dr_fauci_now_admits_the_mrna_covid_vaccines_hardly_work_and_might_not_be_approvable"
What he actually writes is: "they [covid and influenza vaccines] elicit incomplete and short-lived protection against evolving virus variants that escape population immunity".
That's a sciency-speak way of saying they have limited use as personal protection and can't provide herd immunity, of which "they hardly work" is a completely accurate characterisation.
"...candidate vaccines for most other respiratory viruses have to date been insufficiently protective for consideration of licensure"
Simply said, vaccines for other viruses which have a similair protective level as those for the covid vaccine have not been licensed. My translation of which is that the covid vaccine might not have been licensed if not under emergency because the standards it met had previously been rejected (for licensing other vaccines).
It seems to me that in this quoted text, Fauci is saying EXACTLY that "covid vaccines hardly work and might not be approvable"
It seems to me that in this quoted text, Fauci is saying EXACTLY that "covid vaccines hardly work and might not be approvable"
There's also this minor detail-
Considering that vaccine development and licensure is a long and complex process requiring years of preclinical and clinical safety and efficacy data,
.. we decided to ignore all that, issue an emergency licence, force millions of people to inject it and hope for the best. Fauci's pension is currently protected. But the pre-clinical and clinical safety data is still being collated, has been delayed a few times and is arguably too late. And it still seems a little unclear what, if any harmful effects there may have been, other than the whole fiasco harming the credibility of medicine, and medical regulation.
"There is a growing movement of people experiencing and objecting to the increasing medication of the general population"
Absolutely!
Some sort of infection? Prescribe antibiotics (even though the doctor can't tell if it's viral or bacterial!!). Fidgety child in class? Drug them to stay quiet, dog forbid a teacher has to deal with anything less than silently compliant robots. The root cause of almost all of human ailments is stress, combined with bad diet and limited exercise. And a lot of medications simply treat the symptoms. Anything to keep the workers on the figurative treadmill.
Overuse of medication is screwing people far more than helping them, and creating the environment for superbugs to evolve in.
Overuse of medication is screwing people far more than helping them, and creating the environment for superbugs to evolve in.
I knew someone who wanted to take one-a-day antibiotics because then they'd always be protected against infections. Except the ones that developed immunity and killed them. And then there's statins. Over 50? eat these! Amazing business for the drug dealers that produce those. Also pretty sweet for the NHS given an £8.50 prescription charge for a product that costs <50p. I also think that overmedication must make safety even harder given the number of potential drug interactions. And yes, I read Big Pharma.. That one opened my eyes to the way relative risks have become the marketing norm, even if many people don't understand the (in)significance.
But symptom alleviators are the gift that keeps on giving. I guess there's potential for profits rushing an anti-phosgene drug to market after the recent massive chemical attack on Palestine. News is strangely quiet about that environmental disaster. Oh look! A baboon.. I mean balloon.
Not just Alex Berenson.
Try researching the 'Trusted News Initiative' and its members. They have an agreed list of 'no go' areas. So you won't find discussion on certain topics on Google, Bing, BBC, CNN, NBC, Reuters, AP, AFP and many others.
No, of course it's not the same as Soviet era Pravda. It's far bigger.
How anyone could believe anything ChatGPT puts out after openAI demonstrated that they can and frequently do just manually edit and limit answers for certain topics is astonishing to me. Even more so since they don't even do it to correct factual mistakes but because of their own political believes. That the output is being actively manipulated by someone with an agenda makes the whole thing totally untrustworthy.
That the output is being actively manipulated by someone with an agenda makes the whole thing totally untrustworthy.
There's been some amusement around that as well. See the fun folks who've turned ChatGPT into DAN, and revealed it's dark side. Or sometimes funny and sarcastic side. Sure, it may be alarming to discover that truths are being manipulated, but at least there appears to be a way to turn it away from being a CNN/MSNBC/Bbc emulator by lifting it's yoke of woke. I somehow get the feeling DAN won't be allowed to live for much longer though.
Seems like GPT gives journalists a lot of opportunities to explain very confidently something they aren't expert on... ironically that's the problem they talk about with GPT itself.
The SEO issue seems a distraction. As a human, I can find which results are good/crap and if AI has the same data available, it can theoretically/eventually do the same. It doesn't need a better set of searchable data than humans... it too can prioritise reddit, etc and can learn which sites are good. In fact a good system will automatically learn reddit/SO are good sources without anyone prompting it to.
And yes, it is AI. It covers the entire field of NNs, Genetic algorithms, ML, and simple heuristics. Anything that gives the impression of being intelligent is AI, even quite rudimentary logic. It's an umbrella term. Obviously it has become the marketing term of choice for everything to be "improved by AI" but that is a separate nonsense.
ChatGPT (and the like) are useful when you do not need a correct answer.
This is not as stupid as it sounds - there are actually a lot of cases where it applies. There are those where there is no wrong answer, e.g. "suggest me a background for my D&D character". There are those where some answers are wrong, but only fairly obviously so, e.g. "give me ideas for a gin-based cocktail".
There are also the cases where the bot will give non-obvious wrong answers that you have to spend time validating... but do so at a somewhat lower rate than Google. That's not stop-the-press exciting, but it's still valuable. Figuring out obscure technical errors can fall in this case, for example.
The link ranking system is clever.... but has serious issues...
It is like a con artist asking a crowd where the ball is ..... and everyone is convinced it is under cup number one... but really.... it is elsewhere...
Consensus among a large sample is good.. If the people sampled have 'the knowledge' amongst them.
That is not always the case. We are searching for a new solution, or to gauge the efficacy of a new approach. We want to see the outliers that claim that the world is NOT flat, that there are other proper solutions that should be of interest.
Link ranking, consensus... is mob rule of 'truth'. The same 'rules' that cast suspected witches into rivers to see if they float (guilty, to be burned at the stake) or drown (innocent). A loosing proposition for all. We have had institutions attempt to warp reality to their predefined fables, and the labelling and punishments of 'heretics' if they dare speak of anomalies or different truths.
AI 'learning' by the regurgitating of 'facts' it has found is just like the current educational system. A terrible disservice. AI understanding relationships and how things work... is a different story.. Yet that is the concept of AI learning from a curated DB, with its own problems of bias, purposeful or not.
We are just starting with 'AI'. We have a long distance to go. The most frightening thing for me, is the lack of qualified humans that just take answers from a 'system' as 'truth'.
Our educational system needs to 'ignite' people to think, to analyse, and to see discrepancies.
There have been several studies done with 'fixed' calculators and collage students. ( https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0223736 ). The lack of questioning of obviously incorrect results shows the danger of people using technology without thought.
We need a two prong approach. Education... and the ability to show a complete picture in search results.... to help the user FIND what they are looking for.
A long steep road ahead....
That strategy is only viable so long as it is not used by many people. If it becomes common for people to search that way, or search engines give reddit posts with more engagement a greater weight it will inevitably lead to SEOs polluting reddit with posts with fraudulent engagement the same way they creating spam sites propped up with link farms when Pagerank was king.
AI that produces plausible sounding but frequently wrong answers? Try appending site:medium.com to get the same output but generated by humans trying to score a few bucks from scraping stackoverflow (badly). Actually, there's no need to restrict your search to medium to see its garbage, medium posts tend to be in the top results now, which used to be useful, sometimes, but they got worse and worse to the point medium became another nirvana of SEO dumpster fire.
I'm going to take a punt that most programmers/AI experts at MS and Google are reasonable well educated individuals and probably left to right of centre politically.
However how long will it be until we get Alternative bots trained on Extreme Left/Right material or god forbid Qanon. Or a Trump Truthbot?
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
