Post

I'm hearing getting good results out of smaller models will become more standard/expected within the industry. That does require some knowledge and skill, but can result in less expense. You could always "pay to win" with daddy's money, investors, etc. imo, not doing so just looks different

Yep true - if we look at this in terms of building something successful - but I think more out of the pov of a young person starting to get into this profession. I started when I was 12 with my home computer that maybe cost 150€ maximum back then.

Besides the computer and notepad I wouldn't need anything else to start to learn code. Nowadays the shift with coding without ai is stupid would mean you'd need some kind of subscription to pick this up - as a job or a hobby. As a young adult maybe it's possible but not as a teen without income.

AI coding is the future, the cat is out of the bag and we can’t put it back in. And, yes, it is going to be “pay to win” but, it also sort of always had been?

Like we lived in a strange time where a guy with a laptop could out perform a hundred people. But that never really existed any other time in history. This is sort of going to be like the whole, “well in the 1950s a man with a high school degree earned enough to support a wife and 2.5 kids”

I unfortunately think that we are in this weird time where, for boomers, every year got better. And for us, every year gets worse. It’s really just things reverting back to the unfortunate mean that it always was.

Kind of, before you would just need some device to code on and pay for that once, means even as a teenager living in a not-so-well-off family if you had some computer at home you could theoretically pick up the skill and start to code.

Obviously you can still do this today and probably in 2030 - but how realistic is learning that with the job later? I think until local compute is possible on most devices we'll see people being pushed into pricey compute subscriptions.

I think the barrier to entry is fairly low. $20/mo for a hobby is very approachable, and cost of inference seems likely to trend downward or stay the same. I think the real pay to win risk is big tech companies having better models for employees only. Even that seems low risk right now.

i'm not sure what you're doing with $20 a month that will compete with someone spending $200+

It may be easy to pay off as an adult but I think a lot of us started coding when we were teens (at least I assume) and $20 per month was far away for me back then. In contrast you would have been able to just open notepad or whatever editor your 200$ parents computer came with

There was a big barrier to entry even before AI. A smartphone is the only device for a vast majority of people. Those who can't afford a PC are severely limited in what they can do. Ai will make this worse but we might also get cheap AI chips that are capable to run models like qwen 3.8 27b

Oh man, so coding has become golf / horse riding (both are ~expensive / pay to play)?