[Hidden-tech] Question about ChatGPT and machine learning
Michael Muller
tech at montaguewebworks.com
Fri Mar 17 21:50:07 UTC 2023
Excellent story about ChatGPT and OpenAI on NPR today. It's not up on
their website yet.
Mik
Mik Muller, president
Montague WebWorks
20 River Street, Greenfield, MA
413-320-5336
http://MontagueWebWorks.com
Powered by ROCKETFUSION
On 3/16/2023 8:41 PM, Marcia Yudkin via Hidden-discuss wrote:
> Here is a bit of additional information that came up today in an answer from ChatGPT:
>
>>> as of my knowledge cutoff date of 2021.<<
> Don't know if this is relevant to any part of our discussion, but it's good to know.
>
>
>
>
>
>
> On Tuesday, March 14, 2023 at 04:39:18 AM HST, Rob Laporte<rob at 2disc.com> wrote:
>
>
>
>
>
> Hi All,
> Good discussion I wish I had more time for, but a few replies to various points made here:
> * Shell summarizes well: “But I do raise questions about what it all means for the future of work, careers, creativity, etc”. Also true, “a new wave of creativity” can happen.
> * Chatbot-written platforms long preceded gpt, so one can find vastly inferior website content now from those now instantly antiquated platforms.
> * Fact-checking and editing must happen, but that’s less time than composing all, I think.
> * I’ve not tested gpt’s writing extensively, but that legal letter I mentioned was truly consummate, and will be used without a single word changed. I have noticed redundancy, though usually resulting from how I asked a question.
> * Note that one way gpt and the like is and will be trained to improve is via sophisticated AI-human systems by which to detect chatbot text, making the writing output constantly better.
> * Regarding SEO, a ton to say, but for now:
> * gpt expedites all kinds of SEO work.
> * My firm has long prepared for AI, esp last 2 years:
> * It’s about “Things not stings” (entities or concepts, not text strings). But yes, keyword research, a model still prevailing in most software that can't change cost-effectively now, will be expedited by gpt. My firm is currently researching to what extent old-fashioned strings correlate to the replacement: Entities within the MUM revolution.
> * Web marketing pros must coordinate marketing channels efficiently; AI is not good at dovetailing separate silos of knowledge, like SEO, PPC, client profit logic, etc, and it's terrible at prioritizing (within a firm's) context. Hence Shell’s “creativity.”
> * My published prediction 10 years ago that by now AI would relieve SEO pros of lots of routine tech work (via Google Search Console) has been proven way too optimistic--we’re still far from that fine day unfortunately.
> * Extrapolating from the SEO remarks, I believe professionals in most fields will have to stay abreast of adjacent fields of knowledge, so as to be able to efficiently coordinate riders of separate AI horses. Broadly contextual solutions outside single domains of knowledge is something I predict AI will remain deficient in for several years. This goes to Shell’s “new wave of creativity.”
>
> Best Regards,
>
> Rob Laporte
> CEO | R&D Manager
> DISC - Making Web Sites Make Money
> Rob at 2disc.com, 413-584-6500
> www.2disc.com
>
> NOTE: Emails can be blocked by spam filters throughout the web. If you don’t get a reply within an expected span of time, please call.
>
>
>
> ---- On Sun, 12 Mar 2023 11:57:09 -0400 Shel Horowitz via Hidden-discuss<hidden-discuss at lists.hidden-tech.net> wrote ---
>
>> I agree with Marcia. ChatGPT is to a skilled writer as clip-art is to graphics. It's adequate for basic needs but is no substitute for a designer working specifically to convey your idea.
>>
>> I recently landed on a site that I could tell used a chatbot to write its copy. I was repelled and left quickly--and crossed that consultant off my list of people I'd ever hire, because she didn't disclose that a bot wrote the site. Ugh!
>>
>>
>> Shel Horowitz - "The Transformpreneur"
>> ________________________________________________
>> Contact me to bake in profitability while addressing hunger,
>> poverty, war, and catastrophic climate change
>> * First business ever to be Green America Gold Certified
>> * Inducted into the National Environmental Hall of Fame
>> * Certified speaker: International Platform Association
>> https://goingbeyondsustainability.com
>> mailto:shel at greenandprofitable.com 413-586-2388
>> Award-winning, best-selling author of 10 books.
>> Latest: Guerrilla Marketing to Heal the World
>> (co-authored with Jay Conrad Levinson)
>>
>> Watch my TEDx Talk,
>> "Impossible is a Dare: Business for a Better World" http://www.ted.com/tedx/events/11809 (move your mouse to "event videos")
>> _________________________________________________
>>
>>
>>
>> On Sat, Mar 11, 2023 at 9:45 PM Marcia Yudkin via Hidden-discuss<hidden-discuss at lists.hidden-tech.net> wrote:
>>
>> _______________________________________________
>> Hidden-discuss mailing list - home page:http://www.hidden-tech.net
>> Hidden-discuss at lists.hidden-tech.net
>>
>> You are receiving this because you are on the Hidden-Tech Discussion list.
>> If you would like to change your list preferences, Go to the Members
>> page on the Hidden Tech Web site.
>> http://www.hidden-tech.net/members
>>> I'm not sure I would call its output "well-written," though, It's competently written, well-organized and with proper grammar, but stodgy in style and unimaginative, unless you've given it instructions like "write in the style of ____."
>>>
>>>
>>>
>>>
>>>
>>>
>>> On Saturday, March 11, 2023 at 04:22:19 PM HST, Alan Frank<alan at 8wheels.org> wrote:
>>>
>>>
>>>
>>>
>>>
>>> I would use it for composing text, but not for facts at all. And if I
>>> asked it "How would you introduce George Takei at a political
>>> convention"," I would expect well-written text, but would also
>>> fact-check everything.
>>>
>>>
>>> -------- Original Message --------
>>> Subject: Re: [Hidden-tech] Question about ChatGPT and machine learning
>>> Date: 2023-03-11 15:55
>>> From: Marcia Yudkin via Hidden-discuss
>>> <hidden-discuss at lists.hidden-tech.net>
>>> To: Rob Laporte<rob at 2disc.com>
>>>
>>> Rob,
>>>
>>> Those are very interesting examples. It's a mixed track record,
>>> though. Based on your experience, what would you say ChatGPT should and
>>> shouldn't be used for, or how it should or shouldn't be used?
>>>
>>> For example, based on the errors in bios you saw, would you still use it
>>> for those artist bios given that you'd have to meticulously fact check
>>> everything it wrote?
>>>
>>> Marcia Yudkin
>>>
>>>
>>>
>>>
>>>
>>>
>>> On Saturday, March 11, 2023 at 10:19:20 AM HST, Rob Laporte
>>> <rob at 2disc.com> wrote:
>>>
>>>
>>>
>>>
>>>
>>> I've noticed that humans simplify complex, especially threatening, new
>>> things, by using dichotomies of good and evil, red and blue, etc, and
>>> that conviction is often inversely proportionate to knowledge. I've
>>> worked in search marketing for 28 years, have digested a lot about the
>>> kind of tech and processes underlying chatGPT (gpt), and I have no sound
>>> basis for strong conviction on any of the issues broached here. But this
>>> I can tell you from my mere 6 hours novice use so far:
>>>
>>> * In under an hour it solved a complex HR-financial legal question,
>>> and provided the letter the plaintiff should write to the corporation's
>>> HR department, with quality of writing, sensitivity to workplace
>>> politics, and effective brevity way better than anything I can recall in
>>> my 50 years of adulthood, decade teaching college lit and writing, and 3
>>> decades in search marketing. Truly stunning. Save at least $2000 in
>>> legal fees that might have gone to a local law firm.
>>> * A few times over the years I researched best email spam blocking
>>> solutions, and considered an aggressive form of it. gpt explained the
>>> problem with that solution, and did so way faster than my past and still
>>> inclusive Google searches, saving me a few hundred dollars in IT
>>> consulting.
>>> * It completely conflated my semi-famous lawyer grandad's bio with
>>> that of his stellar but less accomplished son of the same name. Both are
>>> years deceased (most gpt data ends Sept '21), yet totally wrong.
>>> * So too it got the bio of a decade-deceased famous scholar of Roman
>>> architecture (friend's dad) wrong on a few points, most notably that
>>> most his career was at Smith college, not Princeton as gpt said.
>>> * It produced strikingly eloquent spreadsheet solutions for two
>>> different complex purposes. I asked it for the actual spreadsheet, and
>>> cell references were off, but in a second half hour of work, I'm sure it
>>> wold have gotten it right or I could correct it myself. A few hours of
>>> work time saved there, and one of the two tasks was billable.
>>> * My firm had a prospective writing project for a client, involving
>>> bios of famous and notable painters sold by the client. I say "had"
>>> because now gpt or its structuring within services like NeuroFash will
>>> cut both client and my firm's copywriter time substantially.
>>> * I've not tried but viewed a YouTube of good, arguably very good,
>>> graphic design for a marketing campaign done in well under half a day.
>>> Outside of broad ideological judgements, there's much to consider in how
>>> gpt will change work and incomes.
>>>
>>> The current version 3.5 will be upgraded to 4.0 within weeks. Think of
>>> gpt like the web in 1994 or personal PCs in 1981, with advancements
>>> happening 10x faster.
>>>
>>> Best Regards,
>>>
>>> Rob Laporte
>>> CEO | R&D Manager
>>> DISC - Making Web Sites Make Money
>>> Rob at 2disc.com, 413-584-6500
>>> www.2disc.com
>>>
>>> NOTE: Emails can be blocked by spam filters throughout the web. If you
>>> don’t get a reply within an expected span of time, please call.
>>>
>>>
>>>
>>> ---- On Fri, 10 Mar 2023 20:36:26 -0500 Marcia Yudkin via Hidden-discuss
>>> <hidden-discuss at lists.hidden-tech.net> wrote ---
>>>
>>>> David,
>>>>
>>>> Some great points there. I especially like this one:
>>>>
>>>>>> it is *ALL* made up.<<
>>>> That helps me to dimly understand that everything the chat says is
>>>> simply plausible, no more than that.
>>>>
>>>> Maybe we should think of it as no more authoritative than the cocktail
>>>> party chatter of someone who reads indiscriminately and can't shut up
>>>> until they've spewed five paragraphs.
>>>>
>>>> Marcia
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> On Friday, March 10, 2023 at 12:34:46 PM HST, R. David Murray via
>>>> Hidden-discuss<hidden-discuss at lists.hidden-tech.net> wrote:
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> From what I understand (admittedly from only a *basic* understanding of
>>>> machine learning), it is not so much that ChatGPT is "making errors",
>>>> but rather that it is "making stuff up", and does not admit that it is
>>>> making stuff up.
>>>>
>>>> I'm going to brain dump what I think here, but I'm not an expert in
>>>> this
>>>> by any stretch, so don't take me as an authority. Perhaps this can
>>>> help
>>>> you reason about ChartGPT until you find a better expert to consult ;)
>>>>
>>>> One thing to understand is that this is a *trained* model. That means
>>>> that it was given a set of questions and answers and told "these are
>>>> good, these are bad", probably with a rating of *how* good or bad.
>>>> Then
>>>> it was given a lot of other data (and how exactly this gets turned into
>>>> questions and answers is *way* beyond my knowledge level). Then a team
>>>> of model trainers started asking questions. The trainers would look at
>>>> the answers it came up with and rate them, thus adding to the "trained"
>>>> data set. When you tell ChatGPT that its answer was good or bad, you
>>>> are also potentially adding to that training data, by the way.
>>>>
>>>> I'm guessing that the way the system works there is actually no way for
>>>> it to "know" that it has made something up. The output that it
>>>> produces
>>>> is generated based on what you can think of as a very advanced version
>>>> of statistical language modelling: given a certain input, what are the
>>>> most likely kinds of things that would follow as a response? And like
>>>> any statistical model, when you get enough standard deviations out,
>>>> things get weird. At no point in the model output are things tagged as
>>>> "made up" or "not made up": it is *ALL* made up.
>>>>
>>>> In the middle of the bell curve the made up things are *much* more
>>>> likely to be "correct" than out at the edges of the bell curve. But
>>>> oh those edges...
>>>>
>>>> It is of course more sophisticated than a statistical model, but the
>>>> same principle applies: if there are few examples of *exactly* the
>>>> kind
>>>> of data your input contains, then it is going to draw from stuff that
>>>> is
>>>> a lot less closely related to your input for its response. But, and
>>>> here is the important part, it is going to make up *something* to
>>>> answer
>>>> with. If a source is mentioned multiple times in the context of your
>>>> input, it will use it. If there are no sources mentioned in the
>>>> context
>>>> of your input, it will generate an output that looks like the *kind of
>>>> thing* that would be a response to that *kind of input*. In this case
>>>> that included a list of articles. It generated at least one of them
>>>> from an author whose name was probably mentioned in the context of your
>>>> input, but never with an actual article name attached. Or maybe that
>>>> author was mentioned in the context of conversations containing a
>>>> subset of the *words* in your input (rather than logically formed
>>>> sentences), depending on just how fuzzy the match was. Then it
>>>> effectively made up a plausible sounding article name to go with the
>>>> author name, because that's what responses to other similar questions
>>>> in
>>>> its training data looked like (not similar in content, but similar in
>>>> *form*).
>>>>
>>>> So while I agree that making up all the sources seems like an extreme
>>>> example of this, ChatGPT is what Science Fiction calls an "Artificial
>>>> Stupid" (something that can't actually *reason*), and thus I think my
>>>> explanation is plausible. It just depends on how fuzzy the match was
>>>> that it made on the input. If the match was very fuzzy, then it would
>>>> have come back with material from its data that generally followed at
>>>> least some of your input, and then since responses the trainers
>>>> considered "good" to questions like that usually included some sources,
>>>> it made some up based on how the answers to other, less related,
>>>> questions looked.
>>>>
>>>> Anyone want to bet that four sources was the average number that was
>>>> accepted as "a good answer" by the people who did the training? I know
>>>> I've seen "four things" in a couple of ChatGPT answers, and I haven't
>>>> asked it very many questions :)
>>>>
>>>> Given all this, there are only two things you can do, one of which is
>>>> exactly what you did: ask it for the sources. Given *that* input, it
>>>> should be able to come up with the most likely response being the
>>>> actual
>>>> source. If it can't, then it has probably made up the source (note: I
>>>> have not tested this technique myself, but it follows logically from
>>>> how
>>>> I think the system works).
>>>>
>>>> The second thing you can do (which you probably also already did) is to
>>>> rephrase your input, giving it different amounts and kinds of context,
>>>> and see how the output changes. If your altered input results in a
>>>> less
>>>> fuzzy match, you will get better answers.
>>>>
>>>> The big takeaway, which you clearly already know, is to never trust
>>>> anything ChatGPT produces. Use it as a rough draft, but verify all the
>>>> facts.
>>>>
>>>> My fear is that there are going to be a lot of people who aren't as
>>>> diligent, and we'll end up with a lot of made up information out on the
>>>> web adding to all of the maliciously bad information that is already
>>>> out
>>>> there. I have read that the ChatGPT researchers are worried about how
>>>> to avoid using ChatGPT's output as input to a later ChatGPT model, and
>>>> I
>>>> have no idea how they are going to achieve that!
>>>>
>>>> And keep in mind that that maliciously bad information *is part of
>>>> ChatGPT's data set*. Some of it the people who did the training will
>>>> have
>>>> caught, but I'm willing to bet they missed a lot of it because *they*
>>>> didn't know it was bad, or it never came up during training.
>>>>
>>>> --David
>>>>
>>>> On Fri, 10 Mar 2023 03:14:21 +0000, Marcia Yudkin via Hidden-discuss
>>>> <hidden-discuss at lists.hidden-tech.net> wrote:
>>>>> Yes, I know that people have been pointing out "ridiculous factual
>>>>> errors" from ChatGPT. However, to make up sources that sound
>>>>> completely plausible but are fake seems like it belongs in a whole
>>>>> other category.
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> On Thursday, March 9, 2023 at 04:10:43 PM HST, Alan Frank
>>>>> <alan at 8wheels.org> wrote:
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> ChatGPT is a conversation engine, not a search engine. It is designed
>>>>> to provide plausible responses based on similarity of questions and
>>>>> answers to existing material on the internet, without attempting to
>>>>> correlate its responses with actual facts. Pretty much every social
>>>>> media space I follow has had multiple posts from people pointing out
>>>>> ridiculous factual errors from ChatGPT.
>>>>>
>>>>> --Alan
>>>>>
>>>>>
>>>>> -------- Original Message --------
>>>>> Subject: [Hidden-tech] Question about ChatGPT and machine learning
>>>>> Date: 2023-03-09 15:29
>>>>> From: Marcia Yudkin via Hidden-discuss
>>>>> <hidden-discuss at lists.hidden-tech.net>
>>>>> To:"Hidden-discuss at lists.hidden-tech.net"
>>>>> <Hidden-discuss at lists.hidden-tech.net>
>>>>>
>>>>> This question is for anyone who understands how the machine learning
>>>>> in
>>>>> ChatGPT works.
>>>>>
>>>>> I've been finding ChatGPT useful for summarizing information that is
>>>>> widely dispersed around the web, such as questions like "what are the
>>>>> most popular objections to X?" However, the other day for a blog post
>>>>> I
>>>>> was writing I asked it "What are some sources on the relationship of X
>>>>> to Y?" It gave me four sources of information, including the article
>>>>> title, where it was published and who wrote it.
>>>>>
>>>>> This looked great, especially since I recognized two of the author
>>>>> names
>>>>> as authorities on X. However, when I then did a Google search, I
>>>>> could
>>>>> not track down any of the four articles, either by title, author or
>>>>> place of publication. I tried both in Google and in Bing. Zilch!
>>>>>
>>>>> Could ChatGPT have totally made up these sources? If so, how does
>>>>> that
>>>>> work?
>>>>>
>>>>> I am baffled about the explanation of this. One of the publications
>>>>> involved was Psychology Today, so we are not talking about obscure
>>>>> corners of the Internet or sites that would have disappeared recently.
>>>>>
>>>>> Thanks for any insights.
>>>>>
>>>>> Marcia Yudkin
>>>>> Introvert UpThink
>>>>> Introvert UpThink | Marcia Yudkin | Substack
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> Introvert UpThink | Marcia Yudkin | Substack
>>>>> Marcia Yudkin
>>>>> Exploring how introverts are misunderstood, maligned and
>>>>> underappreciated in our culture - yet still thrive. Cli...
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Hidden-discuss mailing list - home page:http://www.hidden-tech.net
>>>>> Hidden-discuss at lists.hidden-tech.net
>>>>>
>>>>> You are receiving this because you are on the Hidden-Tech Discussion
>>>>> list.
>>>>> If you would like to change your list preferences, Go to the Members
>>>>> page on the Hidden Tech Web site.
>>>>> http://www.hidden-tech.net/members
>>>>> _______________________________________________
>>>>> Hidden-discuss mailing list - home page:http://www.hidden-tech.net
>>>>> Hidden-discuss at lists.hidden-tech.net
>>>>>
>>>>> You are receiving this because you are on the Hidden-Tech Discussion
>>>>> list.
>>>>> If you would like to change your list preferences, Go to the Members
>>>>> page on the Hidden Tech Web site.
>>>>> http://www.hidden-tech.net/members
>>>> _______________________________________________
>>>> Hidden-discuss mailing list - home page:http://www.hidden-tech.net
>>>> Hidden-discuss at lists.hidden-tech.net
>>>>
>>>> You are receiving this because you are on the Hidden-Tech Discussion
>>>> list.
>>>> If you would like to change your list preferences, Go to the Members
>>>> page on the Hidden Tech Web site.
>>>> http://www.hidden-tech.net/members
>>>> _______________________________________________
>>>> Hidden-discuss mailing list - home page:http://www.hidden-tech.net
>>>> Hidden-discuss at lists.hidden-tech.net
>>>>
>>>> You are receiving this because you are on the Hidden-Tech Discussion
>>>> list.
>>>> If you would like to change your list preferences, Go to the Members
>>>> page on the Hidden Tech Web site.
>>>> http://www.hidden-tech.net/members
>>>
>>> _______________________________________________
>>> Hidden-discuss mailing list - home page:http://www.hidden-tech.net
>>> Hidden-discuss at lists.hidden-tech.net
>>>
>>> You are receiving this because you are on the Hidden-Tech Discussion
>>> list.
>>> If you would like to change your list preferences, Go to the Members
>>> page on the Hidden Tech Web site.
>>> http://www.hidden-tech.net/members
>>> _______________________________________________
>>> Hidden-discuss mailing list - home page:http://www.hidden-tech.net
>>> Hidden-discuss at lists.hidden-tech.net
>>>
>>> You are receiving this because you are on the Hidden-Tech Discussion list.
>>> If you would like to change your list preferences, Go to the Members
>>> page on the Hidden Tech Web site.
>>> http://www.hidden-tech.net/members
>
> _______________________________________________
> Hidden-discuss mailing list - home page:http://www.hidden-tech.net
> Hidden-discuss at lists.hidden-tech.net
>
> You are receiving this because you are on the Hidden-Tech Discussion list.
> If you would like to change your list preferences, Go to the Members
> page on the Hidden Tech Web site.
> http://www.hidden-tech.net/members
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.hidden-tech.net/pipermail/hidden-discuss/attachments/20230317/2fcca056/attachment-0001.html>
More information about the Hidden-discuss
mailing list