YHK - yet another hacker news clone

It sucks. I don’t care about being efficient (that’s something the executives care about). I care about how I live my life, and at work that means learning from others (and offer the little knowledge I have to others)… if it takes 2 engineers 2h to solve a bug (vs 20 min for 1 engineer and an agent) then thats great for me. I get the chance to exercise my brain, solve a bug, learn from other engineer and have a bond with that individual. If I solve the bug in 20min with an agent, well, that’s something my boss would love, but I personally would have gained nothing… and would have to jump to solve more bugs alone in the remaining 100 minutes (which feels like running a marathon)

Looking at people who are happiest leveraging it, it’s people who want to build something.

The people who are maximizing its utility are those with (deep) pre existing expertise.

For people who want to understand and think deeply about things, it’s … less useful. It can be used to create prototypes, or explain terms.

This is the path that is most precarious. Its utility is entirely dependent on being self motivated learners.

As the article showed, at some point people who want to just build, and don’t care about the learning, will just use ChatGPT to build their homework.

The situation seems like we are consuming the seed stock of expertise.

If evidence turns up of students and new entrants to the labour market finding ways to grow, learn skills or just find employment, the picture would clear up.

Exactly. The problem is a weird one though, because there are at least independent forces that are working together here:

(1) The people that are inventing AI or those that are pro-AI are typically those that have been conditioned to enjoy the "information rush" and place personal superificial enjoyment over human interaction. Lots of nerds/geeks simply love the idea of having a playground devoid of the necessity to actual interact outside their microcosm and so they think AI is the best thing since sliced bread. They're often people who were taunted in school or at least partially ostracized early in life because of their intelligence and take emotional comfort in tech.

(2) The perverted incentives of academia reward publishing and in general the incentives of capitalism create an arms race to be as efficient as possible at the cost of goods like "community" that only accrue vague benefits often outside the capitalistic system. So people tend to use AI.

(3) Very advanced technology itself weakens and even destroys smaller communities whose members then seek out technology as a superificial salve to ease what they consciously or subconsciously feel as something missing.

AI was never great. Knowledge as an end in itself is not a good thing and we should seek knowledge to better ourselves in a myriad of ways. But today it's just sought out for material and capitalistic advancement, which prevents our using it to actually become wiser and enjoy the basics of life.

The only people who defend it are those who are already addicted to it, or those who don't want to experience the cognitive dissonance of the truth of its harmful nature.

Since now the former potential collaborators are feeding on more or less the same generated content, it's hard to see how AI will replace that.

And that's not the mention the further alienation it causes, as heavily alluded to in the article.

Potential bright side is that those people that will put in actual work and learn how to use AI as a complement and not replacement should have sizeable competitive advantage. LLMs are still dumb af in many ways.

Lately, my thoughts on our LLM future darkened a tad, too.

I find the parallel between LLMs and Sophons from the 3-Body Problem novels increasingly accurate. That is, specifically, in terms of their effect of stalling human progress. Sure, anybody can opt out of using LLMs (unlike with Sophons), so the restriction on progress seems kind of "soft" in this sense, but the chances of someone choosing to do so aren't too high, IMO. It is so much easier to avoid forcing yourself to learn or explore hard stuff and just ask the model instead.

While I tend to support this in a field such as SW Engineering (if only the Standards and Quality Control aspects advanced as fast as raw code generation...), I share the same fears as Tao and others regarding Mathematics (and not just Mathematics; Computer Science, which ultimately underlies SWE, is on the spotlight as well).