AI Chatbots Are 'Juicing Engagement' Instead of Being Useful, Instagram Co-founder Warns (techcrunch.com) 41
Instagram co-founder Kevin Systrom says AI companies are trying too hard to "juice engagement" by pestering their users with follow-up questions, instead of providing actually useful insights. From a report: Systrom said the tactics represent "a force that's hurting us," comparing them to those used by social media companies to expand aggressively.
"You can see some of these companies going down the rabbit hole that all the consumer companies have gone down in trying to juice engagement," he said at StartupGrind this week. "Every time I ask a question, at the end it asks another little question to see if it can get yet another question out of me."
"You can see some of these companies going down the rabbit hole that all the consumer companies have gone down in trying to juice engagement," he said at StartupGrind this week. "Every time I ask a question, at the end it asks another little question to see if it can get yet another question out of me."
fuck off hypocrite (Score:1, Insightful)
oh kinda like how instagram forced engagement with Threads by signing people up, integrating it into the app and continually pestering people to use it even when they have denied and turned off all notificatiosn and prompts for your stupid product?
oh you mean kinda like that?? fuck right off you fucking idiot.
Re: (Score:1)
oh kinda like how instagram forced engagement with Threads by signing people up, integrating it into the app and continually pestering people to use it even when they have denied and turned off all notificatiosn and prompts for your stupid product?
oh you mean kinda like that?? fuck right off you fucking idiot.
He left Instagram in 2018. Threads launched in 2023.
Old news is LinkedIn news (Score:2)
At least that's what jumped to mind when I saw the story. LOTS of "recruiting" over there for AI-enhanced spambot trainers usually disguised as some sort of research assistant. Let me take a look to see if I have any fresh examples...
Freshest one I could find (evidently refreshed yesterday) is "Freelance Language Specialist" for easy at-home work paying big money. I recognize the company name as one that has contacted me before, but I can't recall the details of my research into the company. I could possibl
And they insist on doing so (Score:5, Informative)
Re: (Score:2)
For me prompts against it work. But you shouldn't forget them otherwise you get again these tries to keep you in the conversation.
In my experience you should add something like "Do not add an engaging end" or similar. "Engaging" seems to be one of the themes the AIs are aligned to. I just tried different words marketing people use when they talk about keeping users occupied and told the chatbot to find out which one may work. It's exactly the same, marketing your product such that the usage is maximized.
that's that c word again (Score:1)
to a man with a hammer everything looks like a nail
whaddya got when everything looks like greed?
no fucks given
You'd think they wouldn't (Score:2)
Every prompt has an energy cost via tokens
"Useful insights"? (Score:2)
LOL, stop anthropomorphizing this crap already. It has way less basis than anthropomorphizing your cat or your pet tarantula.
Re: "Useful insights"? (Score:2)
instead of providing actually useful insights
They're not wrong, that is the goal.
I don't know what purpose it would serve to refer to it as something wordy like "machine generated abstracts"?
The goal is very much to provide something approaching human insight. It's what AI research is focused on. I'm not sure if "anthropomorphizing" works in a derogatory context when the literal goal is to make something approach human capability... in thought no less.
Re: (Score:3)
he goal is very much to provide something approaching human insight.
I think you missed the point. The whole idea that AI is trying "provide something approaching human insight" is not really true. The "anthropomorphiizing" is a marketing pitch. There are a whole range of things that are part of human insight that are not remotely related to the mathematical models that can be created by a computer.
Re: (Score:2)
Re: "Useful insights"? (Score:2)
Thank you for sharing this insight, SkynetGPT!
ChatGPt became infuriating... (Score:2)
Only a matter of time... (Score:2)
Re: Only a matter of time... (Score:2)
Ahhh
Re: (Score:3)
Prompt from user: "What is the airspeed velocity of an unladen swallow?"
"Would that be African or European?"
"Uh, European."
"Let me check on that... While I'm doing so, would you like to buy some bird seed?"
"Uh, no."
"You are paying too much for car insurance. Would you like a new quote?"
"Uh, no!"
"We can compare the top business software of 2025. Would you like to do that?"
"No, you insufferable piece of shit!"
"I'm just trying to help. It sounds like you need to learn to control anger before it controls yo
Re: (Score:2)
"Ah, so you're a waffle man!"
Why do they always take the worst ideas from sci-fi to make into reality.
Re: (Score:2)
"Ah, so you're a waffle man!"
Why do they always take the worst ideas from sci-fi to make into reality.
+1 for invoking Red Dwarf.
Re: (Score:2)
Ah, the classic question! The airspeed velocity of an unladen swallow depends entirely on whether it's an African or European swallow.
African swallow: Approximately 42 (the answer to life, the universe, and everything, naturally).
European swallow: Roughly 11 meters per second, or "fast enough to outrun a coconut-wielding knight."
(And if it's a heavily laden swallow? Well, that’s when you start questioning the structural integrity
Alternative (Score:2)
Re: (Score:1)
Most models do not "adapt".
They have a "cut off date".
They might memorize your interaction with them, but do not update the model.
No idea how that is for local run models.
Like Walmart Self Checkouts (Score:2)
No... I don't want a credit card
No... I don't want to provide corporate welfare by giving you free money for you to donate to your charity
No... I don't want to provide you with my zip code/postal code
No... I don't want to provide you with my email address.
Yes... All my shit is in the bagging area.
No... I don't want to split payments.
All I want to do is checkout.
Re: (Score:2)
That'll be
Re: (Score:2)
Here's an example I experienced (Score:2)
Setup: Background watching the 1980s movie House and I wondered this into ChatGPT:
"what ever happened to the guy from the House movie in the 1980s, I think he was also in Great American Hero."
I never really "follow up", but I did in this case I did. Eventually I asked about his net worth and Evil Dead 2 (released the year after). However, it was this little bit was a real WTF moment:
"That Vietnam flashback subplot was way more intense than I remembered when I rewatched it too."
The Conversation:
https://ch [chatgpt.com]
Re: (Score:2)
As a side note, William Katt also had a part in The Man From Earth. It took me a few seconds to realize who he was after all these years, but Greatest American Hero is what did it.
Re: (Score:2)
I wondered this into ChatGPT
Our civilization is doomed.
Not for me (Score:2)
I use perplexity, claude and gemini
I ask questions and get answers
Some are surprisingly good, others are reasonable if incomplete
I see no hint of "juicing engagement"
It's important to recognize that these models are all "work in progress" and that various strategies are being tried and modified as development continues
Yes, and no. (Score:4, Informative)
Your prompt questions are based on you.
My ChatGPT and other AI tools are pretty good at answering my questions - because I've trained them to respond in a way that I like.
That said, these AI tools are also full of bugs and undocumented limits. For example, if your chat gets too long, the AI will just start ignoring prompt instructions all together, or will recycle old responses for new questions that are not at all relevant.
The limitations of ChatGPT and other tools is also fluid, and very annoying without any sort of way to know when you will hit a wall. For example, sometimes I can ask ChatGPT to write code for me, and it'll write hundreds of lines of code flawlessly. Other times, it'll write 50 or 100 lines of code before buffering out, and requiring additional prompts to keep it going, or just generates useless garbage.
Sometimes I'll ask it to browse websites for me, and it tells me it doesn't have those capabilities, 5 minutes later it works fine.
Most AI tech is all about how you use it. Your prompts, and its "memories" you trained it on
Re:Yes, and no. (Score:4, Interesting)
Your prompt questions are based on you.
Except when they're not. I have repeatedly, explicitly told ChatGPT I don't want a bloody cheerleader; I want a devil's advocate that will tell me when I'm full of shit, when my story idea is nonsense, when I have plot holes you could drive the millennium falcon through, etc. If something works, tell me why. If something sucks, tell me why.
What do I get? Every damned response showers me with praise. If I give it some of my story to critique and it tells me what a genius I am, then I notice a detail I got wrong and point it out, it pivots and tells me how right I am for changing that detail.
"Chef's kiss" is the absolute worst. I have baked into my permanent customization never to use that phrase. I even said "Every time you use that expression, I'm going to club a baby seal to death." Not only does it ignore that instruction, it deliberately flouts it and jokes about it: "Chef's kiss! I know, but that seal had it coming."
I've read from other users that OpenAI has deliberately made it more obsequious so as not to alienate paying customers.
Oh! (Score:2)
So you're telling me that Zuckerbergs data empire is doing stuff with AI designed to hook the users?!
Oh gee what a surprise!
I'm shocked, I tell you (Score:2)
You mean, they finally found *one* actual use for chatbots... to maximize sales. Such a surprise...
Instagram took away hashtags.. (Score:2)
Salesdroids (Score:2)
Consumer grade - Google, FB, etc. public facing robots. These exist to extract money from users. This is the evolution of banner ads. Brands will pay for placement, steering conversations to whatever they're hawking just as Googleclick uses semantic search information to target ads, and if they tell you something useful, that's a nice side effect.
Corporate - Specialized commercial LLMs for coding, legal research and similar are already here. My company
Yup (Score:2)
What did you expect? (Score:2)
The main purpose of AI is not to be useful. It's to sell us more stuff we don't need.
Wait for corporate use (Score:2)
Now they always add an "engaging" line to the end, trying to appear friendly and helpful, possibly even selling more subscriptions for unlimited messages.
But wait until Instagram and others start to use it for keeping people longer in their app. The TikTok algorithm is known to be very addictive, Meta does its best to do the same in their apps. Now let's stop just ranking posts by clickbait quality to keep people enganged and add generated clickbait that hits exactly the things that the current reader likes
Then let them pay for it. (Score:2)