Nobody cares about "AI" (Chatbot: I disagree.)

[Innula Zenovka]

 

[Dakota Tebaldi]

I don't post in the NFT/Crypto Nobody Cares thread as much as I used to, but of course there's still all kinds of scammery going on and one of my favorite news sources is Molly White's blog Web 3 is Going Just Great.

But I wanted to let you know that there is a blog just like that for AI fail-fun, called Pivot to AI, that serves all of your AI FUD needs. The blog is oppositional to AI of course but it does not focus on theory or thinkpieces about why AI is bad or how it will destroy us all like the Terminator or anything like that, the sort of thing traditionally associated with "antis". Like Molly White's blog, it's just a running list of news articles and other reports about AI that all just show how bad AI is at doing literally anything actually serious or important that people, companies, or government agencies try to do with it.

 

[Argent Stonecutter]

Aha, David Gerard (Red Drag Diva of Usenet fame).

 

[Innula Zenovka]

 

[Argent Stonecutter]

OpenAI’s “Whisper” transcription tool is prone to “hallucinations”, whereby it makes up chunks of text, according to experts – including racial commentary, violent rhetoric and imagined medical treatments. One machine learning engineer said he found hallucinations in around half of all transcriptions he analysed

I hate that word.

Those aren't "hallucinations". What people call "hallucinations" are what these tools are fundamentally designed to generate. It's just that often the hallucinations happen to match reality so people are fooled into thinking they're actually doing something useful.

 

[Free]

Elon Musk Recommends Feeding Your Medical Scans Into His Grok AI

Grok now has image understanding capabilities, and to test and improve, Elon Musk wants you to feed it medical documents.

futurism.com

X-formerly-Twitter owner and xAI CEO Elon Musk claims that his foul-mouthed AI chatbot Grok is now capable of understanding images. And what does its creator want you to do? Feed it private medical documents, of course.

"Try submitting x-ray, PET, MRI or other medical images to Grok for analysis," Musk wrote in a tweet on Tuesday. "This is still early stage, but it is already quite accurate and will become extremely good."

"Let us know where Grok gets it right or needs work," he added, presuming that his audience will be willing to freely test his chatbot for him with their possibly compromising medical information — and maybe even to trust its analysis.

Sadly, he was right. Many users have already replied sharing what the Magic Grok Ball had to say about their medical documents, ranging from brain and bone scans to blood charts.

Being Musk fans, they were unsurprisingly enthusiastic about the chatbot's results. A few celebrated the fact that they would no longer need to see a specialist.

ffs...

 

[Dakota Tebaldi]

It's not a HIPAA violation if you start a cult and get your members to upload their own medical data willingly!

 

[Innula Zenovka]

Ooops

Chaos as thousands gather for Dublin Halloween parade — which didn’t exist

Thousands gathered in the Irish capital expecting to see a ‘fantastic light show’ and ‘giant floats’, as promised by a bogus AI-generated article

www.thetimes.com

Evernote Link

Nazir Ali, the owner of the website which lists Halloween events across the United States, Australia, Canada, Mexico, Ireland and the UK, confirmed the page was written by artificial intelligence but edited by a human being to target search traffic on Google, with the aim of generating ad revenue. The content writer tasked with “optimising” the page for search traffic has been fired after they mistakenly thought the event would happen and published the article.

 

[Free]

If only China would follow policy agreements! </sarcasm>

Open Source Bites Back as China’s Military Makes Full Use of Meta's Llama 2 AI Model

China’s People’s Liberation Army is using Llama 13B for military applications. That’s against the acceptable use policy, but there’s no way to put the AI back in the bottle.

gizmodo.com

 

[Dakota Tebaldi]

Nobody is buying Copilot Pro subscriptions, so Microsoft is going to start adding Copilot credits to Microsoft 365 Personal and Family plans now, and increasing the prices accordingly. "Credits" in this sense are like vending machine tokens, each AI thing you want to do with Copilot costs a credit and subscribers get so many credits each month. For now the change is only in SE Asia/Oceania, for some reason, but expect other regions to follow suit eventually.

Microsoft originally wanted $20 a month - separately from your 365 subscription - for Copilot Pro. Which is silly. Not enough people wanted to pay that, which means Copilot isn't getting a lot of use, which doesn't look good for the AI hype; so MS has decided they can just give AI credits to the semi-captive audience of 365 subscribers and they'll end up using them because they're there. That could be true, but I think success is likely to be limited because those 365 subscribers who actually care about and want to use AI likely have bought ChatGPT subscriptions direct from OpenAI and see Copilot as a "watered down" and "censored" version of ChatGPT - which like, that's not really false? So the only thing MS has going for Copilot is its integration into MS Office apps. And I guess we'll see how many 365 Personal users even can make any real use out of that.

 

[Dakota Tebaldi]

Speaking of Copilot integration into office apps, Microsoft is also adding AI to Notepad and lol I just can't

 

[Argent Stonecutter]

I am so glad I am not using Orifice 365.

Bad enough when they made me buy the whole suite just so I could get Excel, which is the only program in the whole shebang that isn't made of distilled pus.

 

[Dakota Tebaldi]

Love it, by which I mean, hate it

 

[Dakota Tebaldi]

At the end of last month the Open Source Initiative released an official definition of what constitutes Open Source AI, and no presently existing product meets the definition. While most AI companies don't care about this because they've stopped trying to pretend that their model is OS, Meta has not, and they are unhappy with the definition.

Basically the "definition" of open source has not really changed since its inception - the entire code used to build and run the software from soup to nuts must be freely available to the public to use and share. As in, like, someone should be able to (in theory) read the code, duplicate it exactly, and build a completely full-featured identical program with it. OSI's definition specifically regarding AI simply re-asserts that, and makes explicit that any and all training data is included within that scope.

Meta doesn't like that. It says that AI's "complexities" mean the old definition should not count, and that AI developers should be allowed to keep some parts of the software secret and non-public, like weights and training data, while still advertising their products as open source because some of the code is made available. In reality, if Meta were forced to reveal the contents of their model's training data, it would lay open the fact that Meta did not have permission to use much, perhaps even most, of it - certainly not commercially - which could likely cause them legal problems.

 

[Noodles]

Nobody is buying Copilot Pro subscriptions, so Microsoft is going to start adding Copilot credits to Microsoft 365 Personal and Family plans now, and increasing the prices accordingly. "Credits" in this sense are like vending machine tokens, each AI thing you want to do with Copilot costs a credit and subscribers get so many credits each month. For now the change is only in SE Asia/Oceania, for some reason, but expect other regions to follow suit eventually.

Inhate this sonmuch as an M365 subscriber.

I already almost exclusively care about M365 for the One Drive part. I priced things out and the $70/year I pay for 6TB of space for M365 family is "the best deal".

Office was just a "bonus".

($70/year is with a discount through work)

 

[Dakota Tebaldi]

Yeah I have it too and it's 100% for the OneDrive space. I think I only ever made a couple dozen documents over the last several years using the Office apps - one of them is like my personal bills and budget spreadsheet, it's the only document I regularly use, and I was able to convert it just fine to a LibreOffice sheet when I switched to Linux. If I could find a decent alternative to OneDrive that isn't Google or Dropbox I'd be very tempted to give it up.

 

[Veritable Quandry]

Mrs Quandry just lost her position as a trainer for an in-home elder care company. They are moving HR, scheduling, and training to an AI system run out of the California office. Caregivers will no longer contact the local offices, which will be primarily for client recruitment and billing.

 

[Free]

It's been getting better?

What if AI doesn’t just keep getting better forever?

New reports highlight fears of diminishing returns for traditional LLM training.

arstechnica.com

For years now, many AI industry watchers have looked at the quickly growing capabilities of new AI models and mused about exponential performance increases continuing well into the future. Recently, though, some of that AI "scaling law" optimism has been replaced by fears that we may already be hitting a plateau in the capabilities of large language models trained with standard methods.

A weekend report from The Information effectively summarized how these fears are manifesting amid a number of insiders at OpenAI. Unnamed OpenAI researchers told The Information that Orion, the company's codename for its next full-fledged model release, is showing a smaller performance jump than the one seen between GPT-3 and GPT-4 in recent years. On certain tasks, in fact, the upcoming model "isn't reliably better than its predecessor," according to unnamed OpenAI researchers cited in the piece.

 

[Casey Pelous]

It's been getting better?

It is totally winning!111111!!!*

AI Chatbot Joins Mushroom Hunters Group, Immediately Encourages Them to Cook Dangerous Mushroom

We have yet another example of artificial intelligence having no idea what it's talking about.

gizmodo.com

* For certain values of winning, including "killing all the humans."

 

[Veritable Quandry]

It is totally winning!111111!!!*

AI Chatbot Joins Mushroom Hunters Group, Immediately Encourages Them to Cook Dangerous Mushroom

We have yet another example of artificial intelligence having no idea what it's talking about.

gizmodo.com

* For certain values of winning, including "killing all the humans."