Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Reflections on AI at the End of 2025

antirez.com

Post reply on HN
Print view View original

371–380 of 383 posts

Re: Reflections on AI at the End of 2025

#371

Post by yeasku »

danielbln wrote: post #115
xiconfjs wrote: post #48

Earlier quoted context omitted.

My person experience: if I can find a solution on stackoverflow etc. the LLM will produce working and fundamentally correct code. If I can‘t find a already fullfilled solution on these sites, the LLM is hallucinating like crazy (newer existing functions/modules/plugins, protocol features which aren’t specified and even github-repos which never existed). So, as stated my many people online before: for low-hanging frui…

I don't remember the last time Claude Code hallucinated some library, as it will check the packages, verify with the linter, run a test import and so on. Are you talking about punching something into some LLM web chat that's disconnected from your actual codebase and has tooling like web search disabled? If so, that's not really the state of the art of AI assisted coding, just so you know.

6 months.

Re: Reflections on AI at the End of 2025

#372

Post by yeasku »

danielfalbo wrote: post #21
Aiisnotabubble wrote: post #18

What also happens and it's irrelevant of AGI: global RL Around the world people ask an LLM and get a response. Just grouping and analysing these questions and solving them once centrally and then making the solution available again is huge. Linearly solving the most asked questions and then the next one then the next will make, whatever system is behind it, smarter every day.

Exactly. The singularity is already here. It's just "programmers + AI" as a whole, rather than independent self-improvements of the AI. I wonder how a "programmers + AI" self-improving loop is different from an "AI only" one.

You are all crazy.

Re: Reflections on AI at the End of 2025

#374

Post by redlock »

emp17344 wrote: post #345
redlock wrote: post #341

Earlier quoted context omitted.

“But clearly the difference between LLMs in 2025 and 2023 is not as large as between 2023 and 2021.” This is a ridiculous statement. A simple example of the huge difference is context size. GPT-4 was, what, 8K? Now we’re in the millions with good retention. And this is just context size, let alone reasoning, multimodality, etc.

Gemini’s 2M context window is kind of a gimmick and not useable in practice.

Not true anymore since Gemini 2.5 pro

I have quizzed it with three books (total more than 1500 pages) and it gave great answers.

Initially yes when they released 2 million context with Gemini 1.5 it wasn’t effective.

Try it with Gemini 3 pro/flash now.

Re: Reflections on AI at the End of 2025

#375

Post by rrrrrrrrrrrryan »

yeasku wrote: post #369
dhpe wrote: post #43

I have programmed 30K+ hours. Do LLMs make bad code: yes all the time (at the moment zero clue about good architecture). Are they still useful: yes, extremely so. The secret sauce is that you'd know exactly what to do without them.

I have programed 10 times that. For me LLMs are a waste of time.

300k hours = 8 hrs per day, every day, for 102 years.

Re: Reflections on AI at the End of 2025

#376

Post by yeasku »

rrrrrrrrrrrryan wrote: post #375
yeasku wrote: post #369

Earlier quoted context omitted.

I have programed 10 times that. For me LLMs are a waste of time.

300k hours = 8 hrs per day, every day, for 102 years.

Behind the computer 16 hours a day all their life?

I know a lot of people who does it.

Re: Reflections on AI at the End of 2025

#377

Post by spider-mario »

Anamon wrote: post #342
redlock wrote: post #341

Earlier quoted context omitted.

“But clearly the difference between LLMs in 2025 and 2023 is not as large as between 2023 and 2021.” This is a ridiculous statement. A simple example of the huge difference is context size. GPT-4 was, what, 8K? Now we’re in the millions with good retention. And this is just context size, let alone reasoning, multimodality, etc.

I don't think that refutes the point. I'd readily agree with the parent that in terms of actual usefulness and efficiency gains, we're on a trajectory of diminishing returns.

The point made by the parent seems to be pretty much the opposite of that. They conceded more tooling but questioned the improvements “at the foundational model level”.

Re: Reflections on AI at the End of 2025

#378

Post by dwaltrip »

monkaiju wrote: post #298
dwaltrip wrote: post #103

Earlier quoted context omitted.

Can you provide numbers relative to things many of us already do? - drive to the store or to work - take a shower - eat meat - fly on vacation And so on... thanks!

Whats the point of this question? These are things (some) people do while existing as humans, not sure how thats relevant? The AI is consuming vast resources while note existing as a human, it doesnt get some innate privilege to consume some amount of resources like we do.

AI is a tool I use while existing as a human. I’d like to know the relative downsides.

Re: Reflections on AI at the End of 2025

#379

Post by Hendrikto »

raddan wrote: post #368
Hendrikto wrote: post #304

Earlier quoted context omitted.

https://www.ers.usda.gov/topics/crops/corn-and-other-feed-gr...

The report you link says that 45% is used for ethanol. A lot, but not “most.”

I could have worded that more clearly: The vast majority (90%+) of corn grown in the US is not for human consumption, with most of that 90% being used for bioethanol.

Re: Reflections on AI at the End of 2025

#380

Post by SCdF »

oersted wrote: post #60
feverzsj wrote: post #54

Earlier quoted context omitted.

In most cases, LLMs has the knowledge(data). They just can't generalize them like human do. They can only reflect explicit things that are already there.

I don't think that's true. Consider that the "reasoning" behaviour trained with Reinforcement Learning in the last generation of "thinking" LLMs is trained on quite narrow datasets of olympiad math / programming problems and various science exams, since exact unambiguous answers are needed to have a good reward signal, and you want to exercise it on problems that require non-trivial logical derivation or calculation.…

Late to this, but my interpretation of the parent's point was eg: LLMs still often produce bad code, despite "reading" every book about programming ever written. Simplistically, they aren't taking the knowledge from those books, and applying them to the knowledge of the code they've scraped, they are just using the scraped output. You can then separately ask them about knowledge from those books, but then if you go back and get them to code again, they still won't follow the advice they just gave you.
Post reply on HN