kblaney.bsky.social
194 posts
100 followers
136 following
Regular Contributor
Active Commenter
comment in response to
post
"Oh, what a lovely ladder I used to get here. Better pull it up behind me."
comment in response to
post
Also a link to the original article on arxiv:
arxiv.org/pdf/2506.088...
Which, admittedly and hypocritically, I did not read the full text of and instead am relying on interpretation from other people.
comment in response to
post
As a complete aside, a huge issue is the commodification of education. If the grade and the degree are more important than the information, of course students are going to optimize their time toward the grade and degree goals.
Educators have been warning you about this for literal decades.
comment in response to
post
For ChatGPT, that info is going to be in a different place every single time. No physical anchor probably makes it harder to recall later.
comment in response to
post
The third thing that might be a factor is the memory trick of remembering where something is on the page. You look something up in a book or a website and you remember roughly where that important info is located first. Makes it easier to reference later.
comment in response to
post
... and as we saw, people got really good at that process. For other tasks that require additional processes, I wonder if we'd see a different impact. For example, asking ChatGPT how to make scrambled eggs and then going and actually making them, people would likely wean off of ChatGPT.
comment in response to
post
For 1, the cost is lower to reference because you don't need to remember how you looked it up. Just a vague description is usually good enough.
For 2, that's just the nature of GenAI: create a certain amount of novelty. For writing essays, the process will always be the same (ask, copy, paste)...
comment in response to
post
In the instance of ChatGPT as reference material, I imagine there are two (maybe three) things that work against this learning:
1. The cost of looking something up is much lower
2. ChatGPT is far less likely to be used to reference the same info (by nature of what it is used for)
comment in response to
post
Notably for educational purposes, repetition over a long period of time will still help people learn because, at a certain point "just remembering" is easier than needing to look the same thing up again and again.
comment in response to
post
Survival tip: if someone invites you, do not... under any circumstances... "come out to play".
comment in response to
post
teeth gang, once and forever
comment in response to
post
Kind of amazing that before exposing himself as a complete dope on twitter he spent shovel fulls of money on building a "billionaire genius philanthropist" image that almost everyone fully bought into.
comment in response to
post
What answers could you possibly need? It happened on video to someone you know. You already have the answers.
Publicly support impeachment. Loudly and often. Do it now.
comment in response to
post
Don't forget the crushing realization that you are not the main character of the universe
comment in response to
post
Putting the NYT Pitch Bot out of work over here.
comment in response to
post
"If we trained an AI on mortgage data, even if we explicitly exclude racial data, the AI would learn to be racist" was my go to example when teaching about Tensorflow. Good to know my old lesson plans have held up.
comment in response to
post
"Written"
comment in response to
post
Revealed preferences would show us that they do like racist policies but don't like being called out for liking racist policies.
comment in response to
post
I wonder if individual Republicans themselves can even keep up with the corruption (other than the parts they are directly involved in). They just know they need to continue to kiss the ring.
comment in response to
post
The singular of sheep should be "shoop".
comment in response to
post
No, it rhymes with "knife"
comment in response to
post
comment in response to
post
The execs only protection is that they tend to be major shareholders. Otherwise they are just replaceable employees.
comment in response to
post
Hey look, its Death Bed, The Bed that Eats People.
comment in response to
post
I literally skip over it without reading like the banner ads in the internet of yore.
comment in response to
post
comment in response to
post
Awesome. Thanks. On the list to be notified at launch.
comment in response to
post
What's the game and where can i buy it? If you can share that is.
comment in response to
post
Hey now... I *also* have a ton of video games in my backlog to play and a few 4 hour long youtube video essays to watch.
comment in response to
post
Looking for the "and that's how I accidentally admitted to my professor that I was getting high during his class" post.
comment in response to
post
Trump just found out that stealth planes don't have cloaking devices.
comment in response to
post
And that matching algorithm, regardless if it was AI or not, is there to influence your behavior. It is not automatically less so if your data passed through a bunch of linear alegbra to get there.
comment in response to
post
For example: take a personality quiz for a dating site. You answer a bunch of multiple choice questions and then it assigns a compatibility score with other people on the site and sorts highest to lowest.
Do you think that matching algorithm was AI? Solid maybe these days.
comment in response to
post
A lot of digital ink has been spilled over the question of determining AI agent vs human, but precious little has been dedicated to AI vs not AI.
That is, if I let you interact with a process, would you be able to tell if the process used AI processes or not?
comment in response to
post
I default to "He Who Walks in Blood" not out of reverence but out of how-do-you-say-that so often.
comment in response to
post
That's awesome. Are all of the characters in the party similarly "would be the monster in a horror movie" styled, or just her?