Who wants some good news?

Last week I spoke to a middle school class about cybersecurity as a career, and AI came up. I asked if they had tried to use LLMs to do their schoolwork—they all had. But every student in the class said it didn't work. Why?

"ChatGPT is too stupid."

Kids are alright.

@mttaggart Which begs the question. If middle schoolers can figure that out, why can't adults? 😉🤦‍♂️
@jbhall56 @mttaggart because being stupid and making a metric ton of fat cash are not mutually exclusive. Greed is a solid motivator and mankind’s single greatest worst invention.

@PetterOfCats @jbhall56 @mttaggart fear is the other worst invention.

Those two are responsible for a huge fraction of the world's misery

@jbhall56 @mttaggart ...or high schoolers...yet.

@mttaggart

It is like trying to use AI to do science, all they know is dogma, they are literally trained on current text books and take everything in them as the way things are with no variations allowed. Anything that doesn't match dogma is forbidden. Way to think outside, but pressed against, the box.

@mttaggart

Wait...so ChatGPT can supposedly pass a bar exam, but it can't do 6th grade homework?

What is on bar exams anyway?

Did OpenAI’s GPT-4 really pass the bar exam?

The large language model’s claims of a top 10% score may have been relative to test-takers who had already failed the exam at least once, according to an MIT researcher.

Fast Company
@MylesRyden it depends on the goal of the questions you ask. If you ask the students to just repeat, chatGPT is good at it. If you ask questions that check if they understand, chatGPT fails (and some students fail, too) @mttaggart
@mttaggart yeah this checks out. There were some stories about high schools in the Netherlands that banned mobile phones from the classroom. The students came up with the idea. They felt more relaxed and talked more between classes. Young people usually understand the good and the bad of new things better (said as an aging millennial)
I think what this really says, not so much that LLMs are "stupid", but that sixth graders are smart enough to lie when asked if they cheat (and this is a quality lie: "I tried it but didn't like it").
@mttaggart
@scottcain Sorry, but I don't agree, and that's a pretty cynical take. 1) They had no reason to lie to me, 2) I have a career behind me of being able to detect middle school lies, and 3) the unanimity and immediacy of the response doesn't track with that theory.
I'm really sorry--I forgot the smiley at the end; I was just joking.
@mttaggart
@mttaggart lol "Kids smarter than the average tech CEO"
@Lazarou Based on what I've read, Twitter would have done better with X Æ A-12 as the CEO than Elon.
@mttaggart Sadly, the same is not true for writing corporate docs/emails/presentations, which speaks a lot about the general quality of those.
@mttaggart That is very good news. Made my day. Tx
@mttaggart ChatGPT and other are great tools to help your brainsmog to clear a bit up (iterating your thoughts and stuff)

or getting rather simple answers.

but those tools are getting dumber nowadays, as they learn shit from LLM outputs. Meaning they learn mostly nothing at all. LLM, similar to Web3, are now at the top of the hill, right before the turning point.

Some stuff will survive as its helpful, but the golden chalice turned out to be mostly made out of papier mache painted gold.
@mttaggart I think a lot of us adults need to try and remember doing schoolwork when we were kids. We /all/ tried to use shortcuts, tools we found online, etc. The kids trying to use ChatGPT are just doing the same thing we did when we used Spark Notes to write that book essay that's totally due in half an hour and we didn't read the book.
@mttaggart Were any of these students asked if they understood where the underlying LLM obtained its source material, and if they knew how much CO2 was generated by each query on AI?

@mttaggart @rysiek

Seems like this behavior changes once they start working. My colleagues think it's great. They use it all the time and don't notice how the replies don't add any value because they don't like & don't want to think about the task/question/problem. They don't want to learn how to do it by themselves. They believe AI can do it better - and that it's fine that way.

I can't even put in words how I feel.

@mttaggart

A friend of mine and college teacher asked me to do a presentation on cyber security over zoom to his class.

That was a lot of fun. Kids all had great questions. This was a few years before chatGPT became common.

@mttaggart NPR is still babbling on about the usefulness of AI or something. They went hard for a solid year on this shit with practically no examination of its big glaring obvious flaws, such as it's over hyped bullshit and the massive energy consumption is just an unfortunate but necessary side effect. Is it because they receive funding from Microsoft
@BLTpizza I would lean more toward NPR's boomery "hmm interesting" tendencies about tech and their lack of critical thinking in reporting whatever businesses say as fact.
@mttaggart I'd agree mostly with that but everyday for nearly a year looks more like an agenda. Maybe it was to gin up interest in buying stock in an AI company?
@mttaggart I think that's exactly it's problem, it's too bad at doing stuff correctly, but they are very very good at quickly doing a lot of stuff that looks correct at first glance. Stuff that becomes false news stories, political ammunition and a sea of misinformation used yo drown out real sources :(
@mttaggart We need those kids to do some TED talks for our CEO's to watch!
@mttaggart also interesting is that the kids were smart enough to recognize how stupid it was. Many adults are not.
@mttaggart drunk people and kids speak the truth (old Dutch saying)
@mttaggart too bad not most students are like this. One can only hope things will change sooner rather than never.
@CrochetMom_78 Yes, they are. But I don't have time for people who don't have faith in kids.

@mttaggart
I think it's more a case of adults know Chatgpt isn't perfect, but they care more finishing the task rather than doing it perfectly.

Like a student who is satisfied with a B may use chatgpt and accept chatgpt getting them a B given that the student didn't have to any work.

Businesses would also rather have 100 support tickets dealt with like 15 mistakes rather than just 10 perfect support tickets.

Of course not everything is like that, but this is what I heard from people I know.

@mttaggart the three ish years where gpt wasn't a phenomenon but I still could use it to write my essays was so goated

One time teacher read my essay full of hallucinations and I just bullshitted my way out of it cuz it wasn't plagarized

Def a net negative but man English class was so much easier

@imadnyc Wow that behavior sucks

@mttaggart No wonder — Elon Musk had been a co-founder of OpenAI. 😵😎

Artificial intelligence appears to be attractive, when natural intelligence never reached the beta stage.