What’s one big weakness of AI that doesn’t get talked about enough? \ stacker news ~AI

pull down to refresh

What’s one big weakness of AI that doesn’t get talked about enough?

150 sats \ 11 comments \ @Kayzone 5 Sep AI

view all related items

61 sats \ 2 replies \ @nichro 5 Sep

Basically. It kisses your ass too often depending on prompt wording. One of the recent South Park pikes fun at it.

0 sats \ 0 replies \ @SimpleStacker 5 Sep

Sycophancy, indeed

0 sats \ 0 replies \ @Kayzone OP 5 Sep

Funny but true...Lol

52 sats \ 5 replies \ @teremok 5 Sep

That people are becoming dumber and lazier by the minute.

Both qualities are universally known to make your life worse.

You see, when I learned most of my coding, my mentors specifically asked NOT to use autofill or intellisense.

Why?

By typing it, you learn it. You see your mistakes after compiling doesn't work.

I use intellisense now of course, but only after I honed my skill.

And sometimes I still skip it and program raw, just for the sake of keeping my brain working. Just like I still do pullups or go for long walks even when a machine called a car can take me further.

You don't learn car mechanics or woodwork by watching. I watched my dad do it for decades and I can't fix a damn broken light switch. At most I can change a tired because he forced me to learn when I wanted to drive his car.

You learn by doing and failing, and people are in for a surprise when they realize doing is hard.

This was my prediction when I worked my fiat contracts and kept programming by hand while others vibed. I was certain that as projects grew, these people would have no clue.

Fast forward to today, and people who vibe now realize the complexity is completely out of their league.

I've seen one too many peers become dumb. Takes them one hour of back and forth LLMing to solve something it takes me 15-30 minutes to code or design.

This is how I keep my edge and stay as a top 1% talent.

0 sats \ 4 replies \ @gmd 5 Sep

Sounds smart... but do you use it at all?

63 sats \ 3 replies \ @teremok 5 Sep

Not at all. I research its limitations very often.

I tried it for coding but I wasn't convinced. It's good when I am not an expert in the programming language but far from better than me in my expertise (frontend).

Simple example, I asked it how many calories in a 5kg whole watermelon.

It will say a 100g portion has 30 calories. So it's a simple 50x30=1500 calories, it says.

Well what about the rind? I asked.

Then it realizes its mistake and does the correct math, but any nutritionist with a PhD level education wouldn't make that mistake. In fact most people who can still think for themselves wouldn't either. I love how people say it's because I am not prompting it right.

Sam Altman couldn't be more wrong about GPT5 being a PhD level expert on ANY given field.

Same mistakes happen in my field.

10 sats \ 2 replies \ @gmd 5 Sep

well I'm in medicine ... LLM's are pretty damn good! lol

I also used to do software decades ago... I was pretty impressed how quickly I could vibe code a crappy game

21 sats \ 1 reply \ @teremok 7 Sep

The devil is in the details. Crappy != good.

I shipped a videogame a while ago. Didn't sell that much but was praised by some big magazines

(Imagine, it was a time when video game mags were a thing)

After coding 100,000 lines of code I can safely say that good video games are by far some of the most complex software projects in the world.