You need to read more of the source blog - he's been pretty pro LLM, but is now acknowledging where it's going too far.

You need to read more of the source blog - he's been pretty pro LLM, but is now acknowledging where it's going too far.

Agreed, this is just a quote on his link blog. Better to post Reddit thread directly. I wouldn’t expect a plain Reddit link to get to the HN front page though.

He's more pro-AI than I am, but his writing has always had nuance IMO

Agree.

Somehow though, this actually might be the best time to for learners, to sit down and engage with topics and don’t be distracted by formal stuff (degrees, grades, points) because the latter is becoming more meaningless with each token being sent down the drain.

> Why not just connect two shitty models together and tell them to hallucinate to each other and skip the whole need to think about anything.

Western countries have better conditions than much of the world for a variety of reasons, but among them is education and culture.

Raising the next generation to outsource all thinking to AI and form a culture around influencing people 45 seconds at a time will destroy those prerequisites to our better lifestyle, and it will be downhill from there.

You might argue that the AI can be a mentor or can guide society appropriately. That's not wholly untrue, but if AI is "a bicycle for the mind", you still have to have the willingness and vision to go someplace with it. If you've never thought for yourself, never learned anything independently, I just don't see how people will avoid using AI to be "stupid faster".

We are normalising everything and everyone into information grey goo.

It's a microcosm of the real economy.

Teachers:

IIRC, it may be better to have the same number of real humans focussing on fewer pupils. Even when they're using VLMs as assistants.

Students:

While humans max out at a higher skill level than VLMs, I suspect that most (not all!) people who would otherwise have finished education at their local mandatory school leaving age, may be better off finishing school as soon as they can use a VLM.

But also #1: There's also a question of apprenticeships over the current schooling system. Robotics' AI are not as advanced as VLMs, so while plumbing will get solved eventually (and a tentacle robot arm with a camera in each "finger" is clearly superior to our human arms in tight spaces), right now it still looks like a sane thing to train in.

But also #2: Telling who is and isn't getting anything out of the education system is really hard; not only in historical systems like the UK's old eleven-plus exams, but today after university graduation when it can sometimes take a bit of effort to detect that someone only got a degree for the prestige and didn't really learn anything.

however, this same action could be useful if it was placed in a different context - for example, if the teacher uses the same AI to produce an artefact, then use it to critique the student as part of teaching (say, to show what might be lacking in a particular piece).