Another masterclass on how not to use AI, courtesy of Fiverr

[uk1000]

51 minutes ago, emmaki said:

Medical AI? No thank you. That's giving a lot of people a lot of bad ideas.I'm not saying Medical AI is bad, it's one of the areas that has a lot of promise but in a medical setting with trained medical professionals and ideally working with people who have longstanding experience in the industry. It's also one of those areas of Fiverr community standards that is very gray. There's a big difference between building out a website for a chiro and, say, automating a diagnostic process for terminal diseases. If I were a medical professional, I would not be looking on Fiverr

It wasn't suggesting it for Fiverr though. He was asked:

Quote

Can you share a pivotal moment in your career that significantly influenced your perspective
or your approach towards AI development? And also, how this experience shaped your entire view on the future of AI?

And he talked about a medical AI that he worked on before ChatGPT was around and that was simpler, and used decision trees, regression etc. he said (according to Whisper AI):

Quote

I would say one of the most pivotal moments was on a project with an association of doctors from Europe.
And these are basically a bunch of cardiologists that are really good at what they do.
And they focus on kids that have congenital heart disease, which is basically when you're born with malformations of your heart.
And so most of our patients are actually babies, are actually kids, very small, very fragile.
And these doctors have to do open heart surgery on these kids. So really high stakes, emotionally taxing.
And so one of those big situations where as a doctor, you might not even have too many data points, so to say,
not too many experiences, just out of the fact that it's a bit more of a niche application.
It's not your routine visit to the doctor as an adult, and you know that people are already formed and so on.
And so pretty difficult for them to pick sometimes what is the best decision,
and specifically what's the best procedure to do on a specific patient.

And so they basically came to me and my team and they asked us to see if it was possible, basically,
through AI to provide them with a tool that would allow them to better know what's usually called personalized medicine.
Better know which procedure to do on a specific patient, on a specific child, depending on their characteristic.
So there might be a specific child where the ideal procedure for that condition, for that situation,
would be procedure A, and it might be longer, but it might be a bit safer, so to say.
And on the other side, there might be a faster, quicker procedure, but be a bit riskier.
And so depending on the characteristic of child, you might want to recommend through AI,
different ideas, different suggestions to the doctors.

And it was a fascinating and pivotal moment to me because you could clearly see how coming from a technical and data science background,
having a customer, a client, that was so non-technical, at least on the coding side, on the programming side, was a really interesting challenge, so to say, because we were not speaking the same language. I was speaking the language of technology, of data, of this format of the data we need,
these are the algorithms we're going to create.

And they obviously were not too interested in knowing which algorithms, what's the performance metrics of these models.
They were more interested in the bottom line, what we were saying, what's the ROI for a business,
which is in this case, how many people can we save? How many kids are they going to give back to their families?
So really interesting to see, first of all, the communication side and how it's key to really be able to speak the language of a customer
and not force them to speak our own language.

And then also how important it is to gain the trust of your client, of your customer, whoever you're working with. And that was achieved in our case by trying not to focus, for instance, on some of the most complex models. GenAI was not yet at the point it is right now, but we've tried to focus on simpler models, decision trees, logistic regression, and so on. So simpler approaches that would be maybe less performant, but would be much more explainable, much more interpretable. And so the doctors were able to actually understand, open, so to say, these models and understand the decision-making process of these models.

So this really allowed us to slowly gain the trust of these doctors that we're not here just to create fancy and interesting technological projects. So it's not for us. It's actually to provide you with something that's valuable. And we also know how important it is to make it explainable and understandable, because at the end of the day, these decisions are so important, which procedure to do on which child, but basically they had to know that if you're picking one decision over the other, it was going to be because of a specific decision-making tree and that the doctors all were aligned with the decisions of the algorithm.

And so really interesting for me to understand, you know, sure, you might have a fancy model, you might have a fancy GenAI system, but at the end of the day, if you're not able to communicate properly with your stakeholders, if you're not really understanding the business or, you know, what they're focusing on and they're not able to gain the trust of making sure, you know, we're all on the same page and we're all aligned on trying to drive value and help these people, then that fancy model is not really going to be effective. And so this is basically to say how important soft skills are after we've done all of the technical work.

Edited April 27 by uk1000

[coerdelion]

@uk1000 - perhaps I wasn't clear enough for you.  The medical AI guy stood out like a sore thumb.  I am aware AI is used for all kinds of diagnostics and surgery preparedness right now. 

However, including that person in a Fiverr webinar implies that it's something sellers can do without the impressive experience and credentials he has.  Indeed, he said at one point something to the effect that coding is becoming less necessary as AI improves.  Right now, One can use AI to code (badly, imo) but he also mentioned that right now, the coding has to be checked and effectively rewritten by a human to make it work - at all, not just properly. 

I made the mistake of abbreviating his talk on the basis that people who a) are familiar with Fiverr, b) watched the webinar and c) have some sense (I know you fit all three options) could fill in the gaps. 

The webinar, as expected, gave us simple stuff to begin with, immediately followed by an - unexpected! - application of complex AI in medicine.  Those two things are inappropriate in the same webinar.  The juxtaposition of the two may give the impression to those sellers who use AI for writing spam and creating graphics that their next step - complex AI - is well within their grasp.  It isn't.

[emmaki]

I broadly agree with what @coerdelion said, but the snippet was interesting to read, so thanks for that. But ultimately, you see the genesis of the issue with Medical AI (and also, to continue from what C said)

1 hour ago, uk1000 said:

And they obviously were not too interested in knowing which algorithms, what's the performance metrics of these models.
They were more interested in the bottom line, what we were saying, what's the ROI for a business,
which is in this case, how many people can we save? How many kids are they going to give back to their families?
So really interesting to see, first of all, the communication side and how it's key to really be able to speak the language of a customer
and not force them to speak our own language.

Never mind the nice-sounding benefits after that. The bottom line for any business is money. It doesn't matter if an AI can or can't help them to save lives or give kids back to families 99% of the time. It's that 1% of the time. It will happen sometime, somewhere - and maybe not on Fiverr.

And when that 1% happens, who will get the blame? It's natural to want to pass on the buck here, and the obvious scapegoat is AI and then "whoever thought this was a good idea". Because the parents or family will want answers that are tough incredibly tough to answer when there's the AI elephant in the room. That's potential national news stuff. That's inescapable name SEO doom and career ruin.  That's laws like "Poor Little Johnny" being passed.

The whole point of making medical language friendlier to patients is to calm them down, not to inform them. It's all about bedside manner. I used to have a friend who was a very good doctor and we would sometimes talk medicine and she never once dumbed it down for me, asserting that I was a writer, intelligent, and lived in Greece, so this shouldn't be tough if I put my mind to it. She was right - but the point of that little anecdote is that she also taught me that most of the job is patient management, since most patients are horrible to work with (due to being in a "bad mood" as a result of their poor health). And, not to put to fine a point on it, the general public is not very smart as a cohort, so smily face cards to show how much your tummy hurts it is.

Decision-making trees only work if the people who create the trees are thinking of all scenarios including the most ridiculously negative outcomes. Right now, AI is in a bubble that will yield one of those outcomes in some form and some way somewhere.

But patients and their families get angry when things go wrong, whether there was malpractice or not (and if you want to be shocked, look up malpractice deaths in the US and remember that the reporting is probably not as accurate as it should be). This is all about ROI, yes, but in my view, it is highly irresponsible to suggest to an audience (indirectly or otherwise) that they can make money doing this. This is an industry where expertise absolutely should be a requisite

It has nothing to do with "fancy Gen AI models". As I said in my last post, and as many people in that linked OpenAI post stated, GPT is literally just deciding what word is best after this word. The body doesn't work like that. A doctor's favorite term is idiopathic ( = "we have no idea what's wrong with you sorry").  More to the point, this is not something for gen AI but expert AI. There are a LOT of medical studies out there that are essentially advertisements because they were paid for by a company. A professional medical person can spot these, but AI? If it's feeding off the internet, which is.... uh... y'know.

The ultimate point of all this being the blame game. Someone will have to pay. It seems entirely obvious to me that if this hypothetical were to happen as a result of a seller on Fiverr - it doesn't matter if they're a super-pro exert or an opportunist - then there will be a storm about that, especially considering that it has this webinar and the AI Hub. Fiverr will be the weakest link, simply because it's full of.... weak links 😎 and as a platform, it is ultimately their responsibility to make sure that sellers are etc. Fiverr may be able to point to the community standards if/when that happens, but realistically, do you think an angry public upset about the tragic AI death of Little Johnny who died because AI got something wrong will care about that?

edit: Just remember that AI has no actual intelligence or free will. It is entirely informed by what it is trained on and what it is told to do. Medicine will always be a zero-sum game when bad decisions can lead to people being seriously harmed. Talking about ROI just reflects the unfortunate capitalist state of the industry. That's a whole other world of murk that starts with the Flexner Report, but that goes way beyond AI. It's a very interesting subject though. Rockefeller was heavily involved in it and it (the results of the report) pretty much destroyed anything that wasn't essentially oil-based pharmaceutical medicine. You know, the pharmas send teams to far-flung places of the world to find new herbs so they can synthesize what makes them work so they can package it up and sell to people - all while demonizing trad med - but the interesting thing is that for some reason, the snynths lose some of the potency that the original has. It's one of the things science can't solve.

Medicine and ROI are two words that should never be in the same sentence. Do no harm.

Edited April 27 by emmaki

[emmaki]

I found a really good article about AI. In fact, the whole website is great. It's someone with a real bee in their bonnet about the abuses of big tech today. I love it. A quick sample of the delights within:

But anyway, this article reveals a looming crisis in the future of AI. It's something that is never discussed, and yet at the same time far more important than how much money the technology can make. It also points to one thing that doesn't really need an article to explain, but that AI does not empower workers (that's you and I) despite the technology (gen AI) needing us to produce more to feed the machine. Who does it empower? Well, if you read what I've been saying all along you already know, but it's nice to hear someone who is really into researching AI this put the whole thing on blast. Glorious, even. 

https://www.wheresyoured.at/bubble-trouble/

I recommend the whole website. I'm spending my Sunday reading it. This, by the way, is the kind of information that I consider to be "empowering". Take note, Fiverr. This is freely available content on the internet. Now compare it to the AI Hub. Hopefully, Fiverr will now realize why I - and other people with any real interest in AI - are giving Fiverr's AI mania the short shrift it deserves. 

I also recommend the thrilling article on how Google destroyed its own search from this same website. I haven't yet found an article about Fiverr, but I bet he would make a tasty meal from the fruits of his research. 

11/10 would read again. Fiverr probably won't though. Let's wait to see what insightful new articles plop out of the AI Hub's rear end next week.

EDIT: This is also good https://www.wheresyoured.at/peakai/ and this one - https://www.wheresyoured.at/are-we-watching-the-internet-die/ - which has an interesting section on "degenerative AI". Yes, the author is unfashionably critical of AI, but he's asking the questions that Fiverr doesn't. @levinewman, I have no idea if you're ghosting around on the forum, but if you're still being courted by Fiverr and thinking about the media stuff, here's a writer who you might like to read. 

 

 

Edited April 28 by emmaki

[aminul_design]

Good read

[shaon_design]

It is easier for bringing AI into proofreading, but the thing is, some of us are against that, because, we develop our own content from top of our head and and with the slightest help of AI, most search engines like google, automatically detects the content as AI and take measurements against it.

[aimriser]

Any request against AI will be checked by AI, So it is brainstorming for none.

[priyank_mod]

4 hours ago, emmaki said:

But anyway, this article reveals a looming crisis in the future of AI. It's something that is never discussed, and yet at the same time far more important than how much money the technology can make. It also points to one thing that doesn't really need an article to explain, but that AI does not empower workers (that's you and I) despite the technology (gen AI) needing us to produce more to feed the machine. Who does it empower? Well, if you read what I've been saying all along you already know, but it's nice to hear someone who is really into researching AI this put the whole thing on blast. Glorious, even. 

https://www.wheresyoured.at/bubble-trouble/

 

Non-millennials who form majority of sellers/readers here are too young to remember the Dot Com Bubble. (https://en.wikipedia.org/wiki/Dot-com_bubble) This bubble burst had sent the entire global IT industry into deep 4-5 years of recession!! 

A non-generative visualization of OpenAI and other IT giants and their data-training sets - after crawling reading the entire article...

 

[emmaki]

I'm still reading through his posts. There's gold in every one. Here's another observation, which will also include some ponderings based on Fiverr's Q4 claims: 

Please take note, Fiverr. The AI Hub is a prime example of the folly of using ChatGPT for everything. 

Now, IIRC, Fiverr claims that AI was responsible for 4% of its revs in Q4. Indeed it does (actually GMV but I can't remember what that is and don't really care): 

Someone who knows stock market things will have to translate that for me, but if article A states that just 5% EBIT arises from AI and Fiverr is shouting from the rooftops about 4%, what does that tell us? Assuming EBIT isn't wildly different to GMV and for the purpose of this snarky statement, Fiverr is.... yes, performing worse than the average business with AI. I am also now wondering why they dropped the "AI" from the complex services in this section. Might it be because the more niche "complex AI services" isn't quite as impressive at all, even accounting for the fact that AI took off in 2023, so the % should be higher?  You can look at their biz trends thing to see an example of huge % increases in search, although I question that people are coming to Fiverr, ready to buy a thing (i.e. they are at the final stage of their buyer journey) and type in... "AI" to the search bar. So are those numbers actually not accurate and a cumulative statistic based on the aggregate of searches that involved the word "AI"? And would that have included terms like "no AI"? Who knows. 

And remember, this cannot be the true statistic as Fiverr has no idea how many sellers are using AI without admitting it which further muddies the numbers and really, if you think about it, allows them to make up any old number that suits whatever narrative they're trying to sell on AI. But the real question is how many buyers is Fiverr losing because of deceptive AI (and people switching to AI since it is faster and, in all likelihood, cheaper and less problematic than working with a potentially deceptive and slow Fiverr seller)? We know 200,000 (active buyers) were lost over 2023. That's a much more concrete number, though of course not all losses will be solely because of AI.

I also don't know what category mix shift really means. Is that multiple categories? Mixed AI/human categories? What? 

We will gloss over contextual search tool Neo, who to this day is unable to show me a seller in the United States when I ask for a seller in the United States. I would say that Neo doesn't understand context at all and is possibly the slowest and worst AI I've had the displeasure to use. 

[emmaki]

Not really related to the post topic, but I've had an update about the SPP price increase. 

I have sent a screenshot as requested, so maybe something exciting will happen, like "oops, it was a mistake". 

[priyank_mod]

5 minutes ago, emmaki said:

Not really related to the post topic, but I've had an update about the SPP price increase. 

I have sent a screenshot as requested, so maybe something exciting will happen, like "oops, it was a mistake". 

And, program page is still showing the old price!! (twice)

 

 

[emmaki]

My original thought when this got sent out was that someone had forgotten to segment it. 

AFAIK the staff on here have not passed any comment (probably wise after the fiasco with the new levels/ratings system) and my SM really only sent me platitudes via corporate GPT, assuring me that this had been thingied. 

Wouldn't it be fun if a Shift Manager at CS found out that I was right all along and that I've been howling on the forum at the transparent company that listens(TM) and has fabulous internal communications?

It would bring joy to my shrivelled up, black heart. I might even say something nice about Fiverr. Once. 

[mandyzines]

12 hours ago, emmaki said:

AI does not empower workers (that's you and I) despite the technology (gen AI) needing us to produce more to feed the machine. Who does it empower?

Wow. Ed's the personification of a rabbit hole. I ended up on a recent Business Insider article of his, much of which applies to the Fiverrverse in terms of Seller Plus,  levels, AI misinterpreting something and sellers being punished for it without much recourse, etc.

[donnovan86]

5 hours ago, emmaki said:

I have sent a screenshot as requested, so maybe something exciting will happen, like "oops, it was a mistake". 

 

I think they still want to increase prices. But I assume the email was sent too early. Maybe they try to provide some more features and that will come with the price increase. Or maybe someone sent the email too soon without any changes coming to the program. Who knows. I would love to keep my current price, and hopefully it was just a mistake on their side. Then again, it's not like it was sent to just a few people. 

Do keep us posted Emma, maybe it was just a mistake.. 

[donnovan86]

5 hours ago, priyank_mod said:

And, program page is still showing the old price!! (twice)

 

Hopefully they keep that old price.

[emmaki]

I did write a reply yesterday, but it is still languishing in jail as others, which made the cut, do not.

9 hours ago, donnovan86 said:

Do keep us posted Emma, maybe it was just a mistake.. 

I don't have an update yet. That this took multiple days to get to a senior support agent suggests something is not right... I don't seem to be allowed to write topics, or at least my topics are also languishing in jail, so you'll just have to follow me around the forums like a lost sheep hoping that one of my posts will provide the details that you want.

If I'm feeling really tricksy, I may just make it buried deep in a long line of congrats and thanks posts. After all, my posts are hidden by nature. Although I think it is worth checking on May 1st (or rather May 2nd) to see if the price update is based on month date rather than with fancy automated AI systems that calculate unique user variables and then. Although that doesn't really need AI if its database stuff.... but no matter, AI sounds more exciting and futuristic than "database update", doesn't it? Unless of course the AI doesn't do anything intelligent like "automatically update" things like next month's pricing so Fiverr can stay on the correct side of EU consumer law. Either way, if this is a mistake, what does this tell us about the actual helpfulness of SMs when it comes to knowing things about Fiverr - and thus, their actual value? You can bet I will be talking to mine about that if this is the case. After she's back from vacation, of course.

I suspect it's the "EU consumer law" that is causing some issues here. I'm really looking forward to the AI act, because as @mandyzines's quote points out, algos are etc, and that's what the EU AI Act is going to try and curb with huge fines. The watch word is transparency. Since Fiverr very much tries to position itself as a mover and shaker in the "future of work"....

And Mandy, he's spot on with a lot of the tech/social media stuff, but his politics is terrible (and oddly uninformed in contrast to his other writing). He's definitely not done much research into politics, so skip any time he mentions politics. I've not gotten to is coughing virus era posts yet. I wonder what those will be like, since that will have to pull in politics with the tech stuff and there's a whole world of dodgy stuff in there and some of the more palatable is starting to rise to the surface. Going to be a bit of a curate's egg, I think.

[donnovan86]

34 minutes ago, emmaki said:

I don't have an update yet. That this took multiple days to get to a senior support agent suggests something is not right... I don't seem to be allowed to write topics, or at least my topics are also languishing in jail, so you'll just have to follow me around the forums like a lost sheep hoping that one of my posts will provide the details that you want.

 

Do send me a private message if you have a definitive answer and I can spread it, in case your posts are not allowed.

For now, it says that my next payment is $20. So I have no idea. That email came randomly, out of nowhere and also on April 1st. So yeah, maybe it was a bad joke from a former employee that was on their last day of work. Who knows, really. I do hope it was a joke, because at $20 the benefits are worth it. But double the price, that's a big NO for me.

 

[emmaki]

Just now, donnovan86 said:

For now, it says that my next payment is $20. So I have no idea. That email came randomly, out of nowhere and also on April 1st. So yeah, maybe it was a bad joke from a former employee that was on their last day of work. Who knows, really. I do hope it was a joke, because at $20 the benefits are worth it. But double the price, that's a big NO for me.

Actually $20, or are you just rounding up? Mine still says $19.

I'll just update here as this is my "home thread" for now, since it keeps the AI Hub front and center of the Fiverr experience.

In slightly unrelated news, I found this post from a while ago while I was researching stuff for another post I just made:

I think that sums it up (I just email mine, but can you really charge $39/mo for a "feature" that is usually overbooked? Come on Fiverr. What would you do if a seller pulled this on their buyers for consultations? The phrase do as I say, not as I do comes to mind. Putting up the price when the features are largely all automated and the most valuable parts (in terms of workload/man hours) are.... like this is bad business practice. On top of the whole breaking trust with price locks for early bird discounts to claw back a maximum of $4k a month from the $19 group. These are not the signs of a financially healthy business that values its customers, whether they are buying or selling.

[donnovan86]

7 minutes ago, emmaki said:

Actually $20, or are you just rounding up? Mine still says $19.

 

Sorry, $19. 

[melanielm]

On 4/26/2024 at 7:40 AM, emmaki said:

Think about how Americans pronounce it ("cleesh"/"cleek") as opposed to Europeans, including the famously monoglot British people ("clee-shay").

We (USA people) pronounce it clee-shay as well. I've never heard anyone said cleesh or cleek.

I'm going to now spend some time with this thread. I really should not ignore all this stuff, but it's just so painful sometimes.

[emmaki]

11 minutes ago, melanielm said:

We (USA people) pronounce it clee-shay as well. I've never heard anyone said cleesh or cleek

TBF, when I say "Americans" I mean "American people on YT/TikTok/IG/whatever who may or may not be influencers".

It's a bit like "rowt" four "route". This makes my ears cry every time.

[emmaki]

I have a response from CS. I'm afraid there is no good news.

I like the last line, which seems to suggest that even if I cancel, I may still be charged, but that's OK because there will be a refund.

Well, I think that's me cancelling. I hope your "price alignment" is worth it, because there's a post about SP on Reddit and last I checked, the majority of responses were "it's not worth it" or "just get it for a month then dump it to fix your KW, it's not worth it" or "I'm a TRS and I've never used it because it's not worth it".  I  believe they're talking about the $19 program. I doubt very few people think paying ~$80 because your SM is too busy to have a 10-minute call with you until 6 weeks later isn't worth it either.

New idea: Why don't SMs have "gigs" on the "Fiverr marketplace" where sellers can leave their "honest feedback" about value for money, timeliness, all of that stuff.

No?

I wonder why.

Personally, I think it's crazy to get SP for a month to fix your keywords, since you get, what... a bunch of HIGH HIGH HIGH HIGH data points to help you decide which keyword is best? Come on, Google only limited its Adwords data to push everyone to the paid program. That wasn't popular, but it makes more sense than charging people for completely nerfed information.

tl:dr: I ain't payin, some other poor soul can pay $40 for chatgpt copypasta.

[emmaki]

Really? On a post about AI, I'm not allowed to share news about the Motley Fool's article about how Fiverr is one of their author's top picks about AI stock?

Am I missing something here?

Or was it my mention of Fiverr breaking GDPR?

I can't tell.

[smashradio]

2 hours ago, emmaki said:

Really? On a post about AI, I'm not allowed to share news about the Motley Fool's article about how Fiverr is one of their author's top picks about AI stock?

Am I missing something here?

Or was it my mention of Fiverr breaking GDPR?

I can't tell.

Strange, one would think Fiverr is all for a positive AI article? 🤣

[emmaki]

It was probably the GDPR stuff, to be fair.

What I liked about the Motley Fool piece was that it appeared to be written by a human. I didn't agree with any of it and found its conclusion tenuous due to poor research (into how the platform is actually operating ATM), but in this era where humans must navigate through a plethora of fast-paced artificial thought, it's refreshing to see a person publishing their own thoughts.

BTW, it looks like OpenAI has released a new version ChatGPT on a "testing" place (not on the OpenAI site) that I will not disclose here, but I'm playing about with it now. It had no idea what the Success Score is because "that's in the future", but the writing wasn't too bad. However, it's also very slow and of course, I've not spent enough time with it to start spotting its giveaway AI phrases.

I'm not sharing it here because it will be abused by everyone, and I do not support the wanton abuse of AI tools.