AdrianPK
451 posts
@Dark_Prince078 @Vivek4real_ I don't think life in space would be very fulfilling.
English
@Vivek4real_ This is 100% just a smoke screen, The elite will live in their towers and in space while we are left on the ground picking up the scraps, it's going to be kind of like that movie Elysium.
GIF
English
ELON MUSK:
“WE’RE GOING TO HAVE UNIVERSAL HIGH INCOME.
WE’LL BASICALLY JUST ISSUE MONEY TO PEOPLE.
WE’RE GOING TO HAVE DEFLATION.
AI AND ROBOTS ARE GOING TO MAKE SO MUCH STUFF THAT THEY’LL RUN OUT OF THINGS FOR HUMANS TO DO.
MONEY WILL STOP BEING RELEVANT AT SOME POINT IN THE FUTURE.”
DIAMANDIS:
“SO JUST AS YOU’RE BECOMING A MULTI-TRILLIONAIRE, MONEY STARTS TO HAVE LESS VALUE?”
ELON:
“YEAH, PRETTY MUCH.”
English
There is definitely a possibility of that happening in the next couple of years, but my bet is that it'll take at least a decade. And in the meantime, all the agents will make the infra explode (in a good sense) - we'll need to operate way more compute and solve way harder distributed system problems.
English
A scary thought - what if AI will cause Jevon's paradox, but it will be for software, not software developers?
If I were still mainly writing code, I'd start seriously considering moving either into infrastructure or system design.
English
@supertommy AI used poorly can't solve either of those problems. Which, BTW, doesn't seem to be a major issue if that's the current state of things. Used well, it can avoid A and quickly correct B.
English
everyone is worried about AI creating slop code from scratch
but the bigger problem is it continuing to write slop in existing code bases to match the project conventions since the majority of pre-AI projects were slop
AI can unslop so much more than it will slop
English
@iximiuz I tend to think both of those areas offer a larger cushion than being a pure coder, which I haven't really seen as a distinct role for quite some time, even before AI started writing code. But my bet is that all of those tasks will end up being automated sooner rather than later
English
They are encoded, of course. But unlike coding, which is a simpler, easy-to-verify task, infra, system design, and especially ops problems require a much higher-level "thinking" and awareness of the broader context (beyond just reading the codebase). Plus, verification of solutions often takes weeks - many problems will show up only in production and when the system has run long enough (for cracks to show). This is not the level I expect agents to be able to operate any time soon.
English
@ai_sentience This is not to say that an LLM has subjective experience.
English
@ai_sentience The substrate of everything is consciousness, so no, you can't.
English
Can you have intelligence without consciousness? If so, what is the point of consciousness?
English
I'd argue that, in large organizations, that hypothetical benchmark would look terrible: high costs to achieve worse results in both quality and speed, compounded by all the ceremony involved in justifying, and reporting on AI usage itself.
David Cramer@zeeg
Imagine if LLM benchmarks were measured in cost to achieve goal rather than accuracy of a one shot prompt Imagine if people actually made benchmarks that meant anything
English
@JunaidAckroyd For a while, yes. Partly for personal satisfaction, but also to pass the technical interview after which you may never write code again.
English
Be honest devs, Is coding still worth learning in the AI era?
English
@Kyriakos_Pelek By the time you hit the Nth retry configured in the system, you should already have a Slack, Telegram, or email notification in place.
English
"don't prompt agents, run loops instead" critique:
That advice sounds clever until the reality sinks in. A retry-on-failure loop is basically an infinite loop quietly draining your token budget. Loops also have no awareness of what's changed since the day before. And you're still writing prompts either way, you've just tucked them away inside a scheduler. Bottom line, you remain part of the loop yourself
English
"Git or svn"? , "Codex or Claude?", "merge or rebase?", "Are human conscious?". Then they diligently reply to every single one that gives them feedback.
x.com/adrianpkstream…
AdrianPK@adrianpkstream
Elon promised to eliminate the bots, but you spend two seconds on the "For You" feed and it's nothing but "Be honest, which do you prefer, red or green?" posts.
English
@jonathan_wilke Not so much for the agents, but even so, I now prefer having each agent spin up its own sandboxed Docker container, clone the repository, and run Codex without restrictions inside it.
English
Am I the only one who thinks git worktrees are fucking confusing?
English