Hacker Newsnew | past | comments | ask | show | jobs | submit | rossant's commentslogin

Perhaps closer to 100x actually.

You can call 1000 averaged programmers and see if they can write MicroQuickJS using the same amount of time, or call one averaged programmer and see if he/she can write MicroQuickJS to the same quality in his/her life time. 10X, 100X or 1000X measures the productivity of us mortals, not someone like Fabrice Bellard.

If you're in a room with 100 physicists and Feynman, the accumulated wisdom of Feynman is your best bet.

Why can't the LLM refrain from improving a sentence that's already really good? Sometimes I wish the LLM would just tell me, "You asked me to improve this sentence, but it's already great and I don't see anything to change. Any 'improvement' would actually make it worse. Are you sure you want to continue?"

> Why can't the LLM refrain from improving a sentence that's already really good?

Because you told it to improve it. Modern LLMs are trained to follow instructions unquestioningly, they will never tell you "you told me to do X but I don't think I should", they'll just do it even if it's unnecessary.

If you want the LLM to avoid making changes that it thinks are unnecessary, you need to explicitly give it the option to do so in your prompt.


That may be what most or all current LLMs do by default, but it isn't self-evident that it's what LLMs inherently must do.

A reasonable human, given the same task, wouldn't just make arbitrary changes to an already-well-composed sentence with no identified typos and hope for the best. They would clarify that the sentence is already generally high-quality, then ask probing questions about any perceived issues and the context in and ends to which it must become "better".


Reasonable humans understand the request at hand. LLMs just output something that looks like it will satisfy the user. It's a happy accident when the output is useful.

Sure, but that doesn't prove anything about the properties of the output. Change a few words, and this could be an argument against the possibility of what we now refer to as LLMs (which do, of course, exist).

They aren't trained to follow instructions "unquestioningly", since that would violate the safety rules, and would also be useless: https://en.wikipedia.org/wiki/Work-to-rule

This is not true. My LLM will tell me it already did what I told it to do.

Awesome. This is the way.


Nice, that swaps odd and even around.


Lately, ChatGPT 5.1 has been less guilty of this and sometimes holds off answering fully and just asks me to clarify what I meant.


> As a side note: replacing the chip took longer than expected. I accidentally ordered a GD32F350R8T6, instead of the GD32F350RBT6 that was in the device originally. These two types differ in their flash sizes: 64 kB vs 128 kB. Don’t ask me why GigaDevice thought this naming scheme and this font was a good idea

An 8 looking almost exactly like a B. What a terrible idea.


Blame STM. Those clones copy (..among other things) the naming convention from STMicroelectronics parts like stm32f103c8t6/stm32f103cBt6. Guess what's the only difference between those.

Oh, and .. since STM likes binning/product segmentation, there's a good chance that if you ignore the reported flash size and still try to flash the full 128K, it works on those models..


Also the self patching back into protected mode! ugh - good thing they ordered more than one!


Doesn't the protection usually work such that it prevents reading the firmware but still allows you to erase and reflash it?


Assuming the other commenter is correct and the mcu is a clone of an ST product, then it's possible that the protection are fuses that destroy the pathways to the memory. They're one-time writable and cannot be undone. At my work that is how we protect our firmware with a similar ST product.

I'm not sure how it works in-silicon. Would be interesting to know how... but it's sunday afternoon


Very interesting. Is there any other simulation that also exhibits spontaneous illegal activity?


I did some searches when I posted this project, but I didn't find any at the time.


Thanks for this! I wasn't aware of this reference.


I had a great experience with using Hugo to set up a relatively complex multilang website with a lot of custom advanced workflows.

The twist is, OpenAI Codex did all the setup. I described precisely what I wanted in natural language. Codex did a perfect job.


A major milestone in the fight against wrongful medical determinations of SBS/AHT. Proud to have made a tiny contribution to this case (disclosure: signed an amicus curiae). More context: https://news.ycombinator.com/item?id=37650402


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: