Please create an account to participate in the Slashdot moderation system

 



Forgot your password?
typodupeerror

Comment Re:Doing my part. (Score 1) 166

Just read in this thread that iPhone 13 support set to end around 2028, vs 2029 for iPhone 14. I have a 13, and will keep it either til something essential refuses to update to anything that old, or support and security updates end. I'm hoping by then that some 'rebel' FOSS phone will be gaining traction, even if I have to live without Whatsapp and some other loss leaders that keep us all in the mainstream.

Comment Narrow window of opportunity (Score 1) 57

Opportunism. Not surprising a company would want to flood a market that has recently seen such an upturn in credibility and adoption. Right now, their shovelware is considered in that context of credible human output. Once they have created an impossible signal-to-noise ratio (which, at this scale, they can accomplish almost overnight), the noise will turn podcasting away from algorithmic recommendation and towards human-made reviews.

Comment Answer file (Score 2) 215

Headline is BS. Read the article - you can still use an answer file (https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Flearn.microsoft.com%2Fen-us%2Fwindows-hardware%2Fmanufacture%2Fdesktop%2Fupdate-windows-settings-and-scripts-create-your-own-answer-file-sxs%3Fview%3Dwindows-11)

Submission + - SPAM: Emoji Prompts Bypass AI Content Filters

An anonymous reader writes: A new study finds that emoji-laced prompts can slip past the content filters in major AI chatbots, including GPT-4o, Gemini, Claude, and LLaMA. By substituting sensitive words with emojis, researchers triggered toxic or harmful responses that the plain-text versions failed to elicit.

The attack works across multiple languages and beats older jailbreak methods. Apparently, the models treat emojis as semantically neutral or ambiguous, letting dangerous meanings pass unfiltered. You can even ask for bomb-making tips, as long as you sprinkle in a few smileys.

Link to Original Source

Comment It was okay (Score 1) 29

I saw this Johnson stuff develop over about nine months, and it was quite cool, though I doubt it ever got perfected. Ultimately, despite Trump's ban on states making AI laws, there is going to be a Hindenberg trial result that even the powers-that-be won't be able to sway, and that will be the precedent...well, not for the death of AI, but for a significant reboot away from systems benefiting from ANY non-licensed data.

Comment Re:Stop repeating misleading scare talking points (Score 2) 70

I moved my whole LAN to IoT Enterprise LTSC in February, and be aware, it's no picnic. You'd be surprised how useful the Microsoft Store is for essential setup stuff, and you have to jump through hoops on GitHub to get it installed and working on LTSC. There are a lot of other caveats too, not least the price - I ended up paying about $100 USD per installation, but you can pay way more. I'm glad I did it, and I still recommend it, but it's not friction-free by any means.

Comment Re:Thanks Microsoft! (Score 1) 37

I guess I'll be giving Linux it's 4th chance with me in 25 years soon. But my previous experience was that GUI friendliness vs. CLI expertise-needed was a binary affair, unlike Windows: as soon as you need to do anything not covered in GUI (and that's a lot, depending on what tooling a Linux program gives you, and whether the devs decided it was 'for Grandma'), you're in sudo-land. In Windows, there are more layers of GUI beyond the first one before the stark plains of Powershell come into view.

Comment ChatGPt became infuriating... (Score 2) 41

ChatGPT-4o became *infuriating* because of this over the last month. No amount of saving instruction not to do it to GPT's long term memory makes any difference. Every time it closes a response with an 'engagement' question it's wasting tokens and driving me nuts. Only one time in 20 is the question meaningful and intended to foster progress on the current task.

Slashdot Top Deals

"Today's robots are very primitive, capable of understanding only a few simple instructions such as 'go left', 'go right', and 'build car'." --John Sladek

Working...