CIRCLE WITH A DOT

Something I don't get to talk about often is how much fun I have self-hosting my own content.

What started as mandatory to protect my privacy has become quite a fun side-project.

@argv_minus_one hopefully, yah.

I totally agree and don’t think it’s this big cybersecurity reckoning. When tokens get cheaper, this will be a great tool for good and the bad guys will just be… the same.

@JessTheUnstill I guess we'll see how it goes!

@JessTheUnstill don't try to contribute content.

It's hard... but being active in the live feed is how I go about it. Say important things, contribute to conversations, mastodon is less about the audience for me and more about finding likeminded individuals.

Eventually, they'll pile up.

@argv_minus_one It's also important to note that this is not some magical cybersecurity robot.

Claude is doing something that vulnerability researchers do all the time, just on projects that haven't been hardened or in code that hasn't been looked at.

Vuln research takes time, money, and expertise. While it's cool we've got a bot that can do that now, it's not new, it's just novel.

The projects that #mythos is finding vulnerabilities in (like the BSD kernel issue) are projects that can't afford vulnerability researchers or audit companies.

#ai #cybersecurity

@Epic_Null @RaffKarva

I just don't see any teachers giving out an F for an 18%. I also think that there is this focus put on the 'bad educator' when that is just not the case most of the time.

If a student made the case for their writing and could defend it, I see no issue with a teacher holding that scrutiny in the first place, it's a part of the job at this point.

My school doesn't have access to an AI-checker so if I suspect AI, I'll call the student over for a few minutes and ask them to defend various points in their essay, this isn't because I'm a 'good' educator, it's because I'm an educator.

I would also argue that more and more, teachers are being FORCED to judge the effectiveness of capturing and communicating an idea, because there are so many missing fundamentals in students who were homeschooled during critical academic years.

That being said, the form is not what makes AI standout. The form is also not what makes or breaks good writing. A perfectly formatted and punctuated essay about garbage is still about garbage.

The things that stand out? Stale organizational structure, overly complex word-choice consistently and correctly throughout, short and to-the-point sentences, overly variable word choice (this is a new one and it's different from what it was like 6-months ago when AI didn't have enough variability in word choice).

@RaffKarva This is the thing though... part of reviewing a piece and its credibility IS reviewing the authority of that author.

This is something that I started noticing a gripe with on Medium; publications started accepting any 'well written' piece maybe 5-years ago, the author would then get added to the publication and, via shoddy review processes, pieces that weren't as good as the original submission would get pushed through.

It is up to you as the reader and consumer of news and online content to evaluate who is saying what you're reading and make a decision on if you're going to trust that person. Cross-check sources, review their work, review the publication that they belong to. This due-diligence is not new, it's just more important now than ever.

@RaffKarva lol, clicked on your profile and realized I'm arguing with a linguist about... linguistics.

@RaffKarva @jalefkowit I don't have a nymag sub, but I read the techdirt piece.

This responsibility falls on educators to not rely on this tool. While the "18%" may be scary, it's also going to be ignored in a lot of cases. It's the same when TurnItIn flags an essay as plagiarism when you're citing something from the source.

I'm not saying this isn't an issue, I'm saying that we've been trained our whole lives to detect this. The same thought you get when you see an AI generated image (less and less, I understand that) is the same feeling you get when you read an AI generated piece.

The difference, humans are linguistic creatures first. We are social creatures and we are trained to tell when someone sounds like they're lying or being coy or sarcastic. It may take a bit longer, and some practice, but we can tell when AI wrote something. An algorithm can't.

Here's a good article on some research that was done, albeit with older models.

Just a moment...

(onlinelibrary.wiley.com)

@RaffKarva @jalefkowit It's less the AI-detectors... those are bad.

People have a certain cadence of writing, even academically, that AI does not respect at all.

As a teacher, I see this all the time. Unless the student has rewritten the whole essay in their voice, individual sentences can stand out to me as AI generated.

Trust your gut, read more content from the author, and it's a bit easier to filter out the noise that way.

@jalefkowit I would disagree here

We're really good at picking up on AI generated writing, and if a post sparks that debate, it is almost definitely AI.

Good writing doesn't beg this question.

@Waldorf entertainment advertised as news.

@Sempf 100%, this is part of it, right, Meta wins some, Google wins some, in the end, it's just circulating the money at the top.

It's a PR push, something to get in the headlines, increase users, increase dependency on it, then BAM, if you can't afford the $200/mo subscription, no AI for you.

I think about this a lot as I work with my students, how are they going to fend when they can't sit and ask a chatbot to sort through a problem for them? Right now, they live in a world where that's a non-issue, but that won't always be the case.

@Sempf I mean this is the thing right, Anthropic is bragging about being more profitable, but their usage limits are MUCH more restrictive than OpenAIs.

Claude is down and all I am hearing are people complaining about "not being able to work," what about when they up their sub to $40/month?

@imreFitos as far as I understand it, the UBIs are a bit more robust and made for OSes or server images.

Project hummingbird is, from Red Hat's website, "a minimal, trusted, and transparent zero-CVE foundation for building cloud-native applications"

Sounds like an attempt to break into the SaaS market by providing smaller developers with smaller, more secure images to build with.