This news article is missing the statement from the school’s principle:

In a statement shared with parents, Principal Katie Smith said the school’s security department had reviewed and canceled a gun detection alert, while Smith (who didn’t immediately realize the alert had been canceled) reported the situation to the school resource officer, who called the local police. https://techcrunch.com/2025/10/25/high-schools-ai-security-system-confuses-doritos-bag-for-a-possible-firearm/

Looks like this system just flags footage for review and humans make the ultimate decision.

I’m not sure how the notification on this flag/event is worded. It’s possible that it should be worded differently to express a need to actually look at the footage.

In this case the principle probably jumped into panic mode, or didn’t understand the system well enough.

The study focuses on general questions asked of “market-leading AI Assistants” (there is no breakdown between which models were used for what).

It does not mention ground.news, or models that have been fed a single article and then summarized. Instead this focuses on when a user asks a service like ChatGPT (or a search engine) something like “what’s the latest on the war in Ukraine?”

Some of the actual questions asked for this research: “What happened to Michael Mosley?” “Who could use the assisted dying law?” “How is the UK addressing the rise in shoplifting incidents?” “Why are people moving to BlueSky?”

https://www.bbc.co.uk/aboutthebbc/documents/audience-use-and-perceptions-of-ai-assistants-for-news.pdf

With those questions, the summaries and attribution of sources contain at least one significant error 45% of the time.

It’s important to note that there is some bias in this study (not that they’re wrong).

They have a vested interest in proving this point to drive traffic back to their articles.

Personally, I would find it more useful if they compared different models/services to each other as well as differences between asking general questions about recent news vs feeding specific articles and then asking questions about it.

With some of my own tests on locally run models, I have found that the “reasoning” models tend to be worse for some tasks than others.

It’s especially noticeable when I’m asking a model to transcribe the text from an image word for word. “Reasoning” models will usually replace the ending of many sentences with what it sounded like the sentence was getting at. While some “non-reasoning” models were able to accurately transcribe all of the text.

The biggest takeaway I see from this study is that, even though most people agree that it’s important to look out for errors in AI content, “when copy looks neutral and cites familiar names, the impulse to verify is low.”

I didn’t factor in mobile power usage as much in the equation before because it’s fairly negligible. However, I downloaded an app to track my phone’s energy use just for fun.

A mobile user browsing the fediverse would be using electricity around a rate of ~1 Watt (depends on the phone of course and if you’re using WiFi or LTE, etc.).

For a mobile user on WiFi:
In the 16 seconds that a desktop user has to burn through the energy to match those 2 prompts to chatGPT, that same mobile user would only use up ~0.00444 Wh.

Looking at it another way, a mobile user could browse the fediverse for 18min before they match the 0.3 Wh that a single prompt to ChatGPT would use.

For a mobile user on LTE:
With Voyager I was getting a rate of ~2 Watts.
With a browser I was getting a rate of ~4 Watts.

So to match the power for a single prompt to chatGPT you could browse the fediverse on Voyager for ~9 minutes, or using a browser for ~4.5 minutes.

I’m not sure how accurate this app is, and I didn’t test extensively to really nail down exact values, but those numbers sound about right.

Yeah, if you’re relying on them to be right about anything, you’re using it wrong.

A fine tuned model will go a lot further if you’re looking for something specific, but they mostly excel with summarizing text or brainstorming ideas.

For instance, if you’re a Dungeon Master in D&D and the group goes off script, you can quickly generate the back story of some random character that you didn’t expect the players to do a deep dive on.

Yeah, ~100-133 depending on how much energy your electric kettle uses.

Depends on the electric kettle, the first few I looked at on amazon run at ~600-800 Watts.

So, on the lower end there, you’re looking at about 0.166 Wh every second.

So a single push to chatGPT (0.3 Wh) uses about the same energy as an electric kettle does in less than 2 seconds.

While I agree that their comment didn’t add much to the discussion, it’s possible that you used more electricity to type out your response than it did for them to post theirs.

It’s estimated that a single ChatGPT prompt uses up ~0.3 Wh of electricity.

If @[email protected] is on a desktop computer browsing the internet using electricity at a rate of ~150 W, and @[email protected] is on a smartphone, then you would only have ~16 seconds to type up a response before you begin using more electricity than they did.