Minor things that bother you

[kphoger]

You know, if sniffing my nose a couple dozen times makes my ears clogged, then blowing it a couple thousand times really should unclog them.  But alas, no.

[Molandfreak]

https://www.today.com/parents/family/ai-written-essays-one-word-professor-rcna265106

Great, so it's no longer just the em dash that sets off AI alarm bells. I love using transition words. 

[Max Rockatansky]

One of our employees filed a safety complaint about having to drink tap sourced water from a dispenser. 

[TheHighwayMan3561]

https://www.today.com/parents/family/ai-written-essays-one-word-professor-rcna265106

Great, so it's no longer just the em dash that sets off AI alarm bells. I love using transition words. 

I feel like we're going to swing so hard on this where we're so paranoid about AI (and amateur detectives trying to snuff it out) that writing anything is just going to be pointless.

[Molandfreak]

https://www.today.com/parents/family/ai-written-essays-one-word-professor-rcna265106

Great, so it's no longer just the em dash that sets off AI alarm bells. I love using transition words. 

I feel like we're going to swing so hard on this where we're so paranoid about AI (and amateur detectives trying to snuff it out) that writing anything is just going to be pointless.

Which really sucks because I actually enjoy writing. Since I was a geezer completing my master's, I didn't understand what ChatGPT was capable of, and I still haven't used it apart from playing around with it to see what it could do once.

With the Janesville comprehensive plan, I absolutely could have outsourced some of the writing to AI just as I could have simply copied and pasted the data I gathered from the Census Bureau and Bureau of Labor Statistics. But I wanted to understand the data on a deeper level, so I performed each calculation with the data in Excel. Is this level of intimate attention to detail, striving to truly understand, and having someone truly become an authoritative source about a subject just going to go away completely?

I suppose someone also asked that question when computers and calculators became available, but this is fundamentally different since at least the equations are actually being inputted by a human with an end goal in mind applying the calculations.

[Scott5114]

With the Janesville comprehensive plan, I absolutely could have outsourced some of the writing to AI just as I could have simply copied and pasted the data I gathered from the Census Bureau and Bureau of Labor Statistics. But I wanted to understand the data on a deeper level, so I performed each calculation with the data in Excel. Is this level of intimate attention to detail, striving to truly understand, and having someone truly become an authoritative source about a subject just going to go away completely?

No. People who have outsourced this level of analysis to ChatGPT have sorely regretted it, because rather than using the numbers provided, it fabricates them, so the results end up being totally wrong. More than one company has gone under by this point because they trusted ChatGPT with an analysis, didn't bother to check the numbers themselves, made decisions based on what ChatGPT said, and then reality came calling.

If anything is likely to go away, it's ChatGPT, because all of the sunny projections of market share and future profits from the AI companies are just as baloney as what ChatGPT craps out. The AI companies put out these fraudulent numbers to trick investors into providing the fuel keeping the merry-go-round going, which requires everyone to believe that AI is going to take over and continuously grow. The media is complicit because either they don't want to lose access to Jensen Huang and Sam Altman or they don't have a clue why all of this is a bill of goods, so they don't ask any of the tough questions someone should be asking.

[The_Ginger]

If anything is likely to go away, it's ChatGPT[.]

We're already one step closer to that. Sora, the OpenAI platform that made the viral short AI videos has shut down.

https://www.nbcnews.com/tech/tech-news/openai-shuttering-sora-video-generating-service-rcna264989

[Rothman]

With the Janesville comprehensive plan, I absolutely could have outsourced some of the writing to AI just as I could have simply copied and pasted the data I gathered from the Census Bureau and Bureau of Labor Statistics. But I wanted to understand the data on a deeper level, so I performed each calculation with the data in Excel. Is this level of intimate attention to detail, striving to truly understand, and having someone truly become an authoritative source about a subject just going to go away completely?

No. People who have outsourced this level of analysis to ChatGPT have sorely regretted it, because rather than using the numbers provided, it fabricates them, so the results end up being totally wrong. More than one company has gone under by this point because they trusted ChatGPT with an analysis, didn't bother to check the numbers themselves, made decisions based on what ChatGPT said, and then reality came calling.

If anything is likely to go away, it's ChatGPT, because all of the sunny projections of market share and future profits from the AI companies are just as baloney as what ChatGPT craps out. The AI companies put out these fraudulent numbers to trick investors into providing the fuel keeping the merry-go-round going, which requires everyone to believe that AI is going to take over and continuously grow. The media is complicit because either they don't want to lose access to Jensen Huang and Sam Altman or they don't have a clue why all of this is a bill of goods, so they don't ask any of the tough questions someone should be asking.

I still find your assessment way too severe.  Plenty of companies are using whichever AI agents as tools to make things easier for themselves without causing Armageddon.  I suppose certain AI tools may go away as they are outcompeted by others, but some.form of AI will be here to stay, especially as their results continue to be honed.

[formulanone]

https://www.today.com/parents/family/ai-written-essays-one-word-professor-rcna265106

Great, so it's no longer just the em dash that sets off AI alarm bells. I love using transition words. 

I feel like we're going to swing so hard on this where we're so paranoid about AI (and amateur detectives trying to snuff it out) that writing anything is just going to be pointless.

"Moreover" (saved you four clicks)

It came from a TikTok video by an adjunct professor of "social media and personal branding", which means the entire article has the substance of unsweetened gelatin.

If they start challenging the ellipsis and semicolon, I think we're doomed.

[kphoger]

https://www.today.com/parents/family/ai-written-essays-one-word-professor-rcna265106

Great, so it's no longer just the em dash that sets off AI alarm bells. I love using transition words. 

In even that short article, I count two em-dashes.  Was it written by A.I.?

[Scott5114]

I still find your assessment way too severe.  Plenty of companies are using whichever AI agents as tools to make things easier for themselves without causing Armageddon.  I suppose certain AI tools may go away as they are outcompeted by others, but some.form of AI will be here to stay, especially as their results continue to be honed.

These things tend to look like they work until they don't. Moreover, (sorry, had to) in most companies management has yet to investigate how many man-hours are actually saved once you account for the time a human spends verifying the AI is correct and cleaning up when it's not. I would imagine—when supervised correctly—it's a modest improvement, but not anywhere near as earth-shattering as people tend to think it is. AI is nowhere near good enough to actually run as an autonomous agent, and trying to do that is folly.

Another point on the "it looks like it works until it doesn't" thing is when people use it to automatically generate code. It may save time by generating code that works. However, if there isn't a human writing that code, that means that if it has a weird bug that isn't obvious until later, there's nobody around who actually understands what it wrote. You can feed it back into the AI and try to get it to fix it, but there's no guarantee that will actually work. And if it just breaks more stuff trying to fix it, you're down a bunch of man hours when someone has to go in and rip out the defective code and fix it.

And none of this addresses the elephant in the room—financials. Even if all this stuff worked, there is no path to profitability on any of these companies, because the amount they would have to charge to break even on operating expenses is a non-starter for their customers. (We know this because when they raise prices, people cancel their subscriptions!) And since this is operating expenses it goes up as they get more customers. There is no economy of scale to be had here; more users means you need more computing power, not less.

I think you're correct that AI is here to stay, but I think the future of AI is in functions baked into other programs. It would be really cool if there were an AI routine that automatically vectorized a raster image that you could call as a click of a button inside Inkscape. It would be neat if you could use AI upscale functions inside Photoshop or whatever. But agents? Nah, that's sci-fi shit that won't work.

[kphoger]

My wife uses ChatGPT for her Scentsy gig.  She's not a very artistic/creative person, so coming up with names for product bundles and stuff like that is a time sink for her.  With ChatGPT, she can generate a list in a few seconds and just pick the ones that look the best, and maybe tweak a couple of them.  She's also used ChatGPT to create product bundles in the first place:  she plunks in the names of all the products she's looking to sell, and in no time at all ChatGPT digests the scent descriptions of all of them and churns out a list of bundling ideas;  most of them end up being pretty good, with just some easy substitutions or swaps between bundles.  Saves her a whole bunch of time, with no ill effect at all.

[Molandfreak]

https://www.today.com/parents/family/ai-written-essays-one-word-professor-rcna265106

Great, so it's no longer just the em dash that sets off AI alarm bells. I love using transition words. 

I feel like we're going to swing so hard on this where we're so paranoid about AI (and amateur detectives trying to snuff it out) that writing anything is just going to be pointless.

"Moreover" (saved you four clicks)

Sorry. I thought "transition words" was enough for get the point since the article goes on to explain that "furthermore" and "thus" also fit into the category.

[kphoger]

"Moreover" (saved you four clicks)

Sorry.

I don't understand the "four clicks" thing anyway.  It only took me one click to get to the article, and the word was identified in the third sentence.

[Molandfreak]

I also wonder if this is something unique to the nature of the "social media and personal branding" courses that particular professor teaches. Unless they are writing an academic paper about marketing or something, it would be odd to see these words in a sample marketing piece.

[JayhawkCO]

Saves her a whole bunch of time, with no ill effect at all.

Other than boiling the oceans.

[kphoger]

Saves her a whole bunch of time, with no ill effect at all.

Other than boiling the oceans.

Well, there's that...

[Scott5114]

My wife uses ChatGPT for her Scentsy gig.  She's not a very artistic/creative person, so coming up with names for product bundles and stuff like that is a time sink for her.  With ChatGPT, she can generate a list in a few seconds and just pick the ones that look the best, and maybe tweak a couple of them.  She's also used ChatGPT to create product bundles in the first place:  she plunks in the names of all the products she's looking to sell, and in no time at all ChatGPT digests the scent descriptions of all of them and churns out a list of bundling ideas;  most of them end up being pretty good, with just some easy substitutions or swaps between bundles.  Saves her a whole bunch of time, with no ill effect at all.

How much would she be willing to pay for that? $500/month? $1000? Those are the sorts of levels that would be required to make the AI companies viable. One heavy user of Claude managed to cost Anthropic $150,000 all by himself, just by using the product as intended.

One start-up founder told me that he had discovered an A.I. tool made by Figma, a design start-up, that allowed him to use the equivalent of $70,000 in Claude tokens through an account that costs him $20 a month. The founder, who asked to remain anonymous to avoid tipping off Figma, said he had used the loophole to build six software projects at the same time.

I assume since you said since "she picks the names that look the best" and "most of them end up being pretty good" that at least a few of them are mediocre or stink and aren't used. That means she's exercising some degree of oversight over the thing and not just blindly accepting everything it does. When people talk about "agentic AI" they're talking about doing stuff like telling it to generate the names and bundles, having it write the product descriptions, giving it the password to their account, and telling it to do the listing and then never checking what it did. Then they're astonished to find that they somehow have a live listing for a bundle called "Poopoo Potpourri" that consists of a Citronella candle, a hunk of beeswax, and a new car, with a total list price of $3.99.

[formulanone]

"Moreover" (saved you four clicks)

Sorry.

I don't understand the "four clicks" thing anyway.  It only took me one click to get to the article, and the word was identified in the third sentence.

Two ads immediately got the way and third was a reminder to add my email, which blocked my screen.

Your clickage may vary.

[LilianaUwU]

My wife uses ChatGPT for her Scentsy gig.

Two red flags in the first sentence.

[kphoger]

Then they're astonished to find that they somehow have a live listing for a bundle called "Poopoo Potpourri"

https://pourri.com/collections/all-products?srsltid=AfmBOoqiD_zUfZyMu8T89eIhGB900LaLLQjhtGtQxiywvDbvfDaNAYHn

[hotdogPi]

The "worst answer" thread gets a lot of likes in the answers but nobody ever seems to "like" the questions themselves.

[kphoger]

My wife uses ChatGPT for her Scentsy gig.

Two red flags in the first sentence.

Sorry if you don't think I deserve a wife.

[kphoger]

Weather Underground:  "Tomorrow's temperature is forecast to be MUCH WARMER than today."

It's only forecast to get 9°F warmer than the current temperature as I type this.  Seriously?  That's "MUCH WARMER"?

[Molandfreak]

Weather Underground:  "Tomorrow's temperature is forecast to be MUCH WARMER than today."

It's only forecast to get 9°F warmer than the current temperature as I type this.  Seriously?  That's "MUCH WARMER"?

If it's 46°, overcast, and windy one day and then 55° and sunny the next day, that is the difference between windbreaker weather and T-shirt weather.