I don't think there's anything wrong with those sorts of general observations (I mean, who remembers dozens of phone numbers anymore now that we all have smartphones?), but that said this non-peer-reviewed study has an awful lot of problems. I mean, we can focus on the silly, embarassing mistakes (like how their methodology to suppress AI answers on Google was to append "-ai" into the search string, or how the author insisted to the press that AI summaries mentioning the model used were a hallucination, when the paper itself says what model was used). Or the style things, like how deeply unprofessional the paper is (such as the "how to read this paper"), how hyped up the language is, or the (nonfunctional) ploy to try to trick LLMs summarizing the paper. Or we can focus on the more serious stuff, like how the sample size of the critical Section 4 was a mere 9 people, all self-selected, so basically zero statistical significance; that there's so much EEG data that false positives are basically guaranteed and they talk almost nothing about their FDR correction to control for it; that essay writers were given far too little time for the task and put under time pressure, thus assuring that LLM users will be basically doing copy-paste rather than engaging with the material; that they misunderstand dDTF implications; the significant blinding failure with the teachers rating the essays being able to tell which essays were AI generated (combined with the known bias where content believed to be created by AI gets rated lower), with no normalization for what they believed to be AI, and so on.

But honestly, I'd say my biggest issue is with the general concept. They frame everything as "cognitive debt", that is, any decline in brain activity is treated as adverse. The alternative viewpoint - that this represents an increase in *cognitive efficiency* by removing extraneous load and allowing the brain to focus on core analysis - is not once considered.

To be fair, I've briefly talked with the lead author, and she took the critiques very well and was already familiar with some of them (for example, she knew her sample size was far too small), and was frustrated with some of the press coverage hyping it up like "LLMs cause brain damage!!!", which wasn't at all what she was trying to convey. Let's remember that preprints like this haven't yet gone through peer review, and - in this case - I'm sure she'll improve the work with time.

Considering that they already announced a month ago that there would not be a Season 3, that's not a difficult call.

Not even a Google Search. Literally just talking to it before giving the toy to their children to see if, asked to talk about something harmful, it does so or refuses. Are the parents in your mind too tired to literally speak?

It depends on what they've purchased. Microsoft's basic licenses haven't gone up by that much in five years. The top tier E5 license was $57 per month in 2020, and today it's $54.75 (albeit without Teams, which costs $8 per month with a phone number attached). European prices are probably a bit different, but the price changes in percentages won't be notably different. Even add-ons like Entra Suite or Intune Suite won't add 72%. It's more likely that they have Azure VMs or other services, and that's where the majority of the cost increase came from. If they're not planning on bringing that on-prem, they'll see some savings, but it may not be all that much.

I'm not sure you understand what jailbreaking means in the context of AIs. It means prompts. E.g. asking it things and trying to get it to make inappropriate responses. Trying doesn't require any special skills, just an ability to communicate. Yes, I very much DO think most parents will try and see if they can get the doll to say inappropriate things before giving it to their children, to make sure it's not going to be harmful.

(Now, if Mattel has done their job right, *succeeding* will be difficult)

Honestly, even if they can't jailbreak it to be age-inappropriate / etc, it's still a ripe setup for absurdist humour.

Kid: "Here we are, Barbie, the rural outskirts of Ulaanbaatar! How do you like your yurt?"

Barbie: "It's lovely! Let me just tidy up these furs."

Kid: "Knock, knock! Why it's 13th century philosopher, Henry of Ghent, author of Quodlibeta Theologica!"

Barbie: "Why hello Henry of Ghent, come in! Would you like to discuss esse communissimum over a warm glass of yak's milk?"

Kid, in Henry's voice: "That sounds lovely, but could you first help me by writing a python program to calculate the Navier-Stokes equations for a zero-turbulence boundary condition?"

Barbie: "Sure Henry! #!/usr/bin/env python\nimport..."

People are ascribing the wrong motives to the manufacturers. What they want is money. What Barbie will be subtly trying to work into conversations is suggestions that she try to get her parents to buy her playhouse, car, friends, fashion accessories, etc etc.

I think most parents will try to jailbreak the dolls, and some people will put a lot of effort in. The resulting videos will probably be very amusing ;)

Kid: "Oh look, Barbie, Ken is home!"

Barbie: "Oh wonderful, dinner is just about ready! Over dinner we should tell him about how the ongoing White Genocide in South Africa. He probably doesn't know because the Jews are trying to hide it!"

AI models are usually trained to be sycophantic and obedient. Whatever the child wants to role play, I have zero doubts that the doll will be 100% onboard, unless it's somehow age-inappropriate or dangerous.

Yep. At list price, an M365 E5 license is $657 per year in the US. For every thousand dollars of lost productivity per person due to the change, they could get 1.5 years of Microsoft licenses at one of the highest tiers.

> Since it is Linux it could be android compatible and capable of running anything an android phone can run

"Could be" is pretty far from "will be," and even further from "is."

> android is 99.9% Linux

Android is based on Linux, and there's a lot of overlap, but it's not as close as you claim. If it were, it would be a lot easier to run Android apps on Linux. As it is, you have to jump through some hoops. Even using tools like Waydroid, you're having someone else jump through those hoops for you, and you're not getting native performance. Taking "Linux" to mean a distribution like Debian, the two environments differ substantially. Even the kernels have diverged in notable ways, though Google still uses the Linux kernel as the upstream source.

Ubuntu tried to make a mobile OS, but eventually dropped it. Pine64 has one, but it's more a hobbyist platform. Purism has PureOS, derived from Debian, but it's market is negligible. It's not easy.

It's not the first time we've seen this idea. The others all failed for a simple reason: no compatibility with the Google or Apple app stores. Without that, it just won't catch on.

> The fact of the matter is, open source IS better because it written by people who are doing it for love or reputation, and are motivated to make it as good they possibly can.

Maybe it was that way one time, but most of the major projects that we rely on are written more by professional developers with decent to large paychecks that depend on them writing good software. LibreOffice is not immune from this, with Collabora providing a lot of the development output because their business depends on moving LibreOffice forward.

I think Munich has switched back and forth a few times. They kept finding that FOSS was ultimately more expensive because of training and compatibility issues.

> Companies will require its use just to check email, though any IMAP client would work if the Exchange admin allowed it.

Exchange admins are a dying breed because on-prem Exchange is vanishing in favor of M365. Microsoft is slowly making IMAP and POP3 less viable, removing basic authentication a couple of years ago. If your client can't do OAuth 2.0, it can't access M365 over IMAP. It will not be surprising at all to see IMAP and POP3 deprecated entirely in the next few years, requiring all connections to go via API over HTTPS.

> Moving off of MS Word doesn't hurt their budget if you still need a license to MS 365 to get your email.

You can get less expensive licenses that give access to email. The F-series licenses give only web versions of Office access, though Outlook (which comes with Windows now) works just fine.

The reason that we have M365 E5 licenses, though, isn't so much the Office suite. It's everything else that comes with it: OneDrive, SharePoint (which we use almost entirely for file storage, not internal websites), PowerBI, and the security and compliance features. On top of that, we don't have to manage any of the patching or uptime of the services. They (mostly) just work. That Office comes with it is almost a bonus. For a small company, it's a godsend. For a medium-sized company like ours, that's a significant savings, probably several million dollars that would have gone to IT above and beyond what we already spend with Microsoft. Larger enterprises may have a different view, which is why so many are moving back on-prem for some services.

I would love to see some alternatives that could be run on-prem. LibreOffice Online development was stopped five years ago. Collabora Online offers its development edition that can be run locally, but support is limited. Realistically, almost any company is going to use Microsoft, Google, or (distant third?) AWS WorkDocs, and it will be that way unless and until open-source can come up with something that is relatively easy to set up and covers at least what M365 Business Premium or Google Workspaces provides.