What answer do you expect?
68 karma · joined June 10, 2023
What answer do you expect?
An AI able to reason, obviously. Ideally able to reason better than human - and be able to explain the universe better than humans did so far. So don't feed me hallucinating LLM stupidity.
> And if a lot of people derive real value out of this new wave of gen AI, but you don't, does that say more about gen AI or more about you?
Good question! I'd say, given the state of things, that it says more about people.
Just reproduce the code with LLM of your choosing. If you manage to actually do so - please feel free to add graphics with any other model.
If you also see six tweets, please reproduce the result following instructions in the tweets and share the recording of you doing so.
Graphics don't matter, let sprites be colored rectangles.
These tweets do not describe the exact and specific steps to reproduce the complete result. They only provide rough and obvious (for anyone ever interacted with GPT) advice, as is par for the course for AI hyping.
This:
> And from there, keep asking for more and more things. And every time something goes wrong, clearly explain the mistake and let it fix it. Patience!
Is pathetic, and proves exactly nothing.
Yep, I did not do that all the times I tried to actually see why GPT4 is so popular, thanks a lot for advice!
Show all prompts step-by-step and let people reproduce.
Someone claims they did X using a couple of AI instruments, without sharing anything to prove the claim, and everyone is excited. I understand that the industry needs a new hype to keep going, but this is just pathetic.
By the way, I have a nice bridge for sale.
Almost 200 comments and Ctrl-F "repro" is zero.
Also let me be straight: I am full-on believer in AI. But not in Midjourney, DALL•E 3 and GPT-4 (or other LLMs) - these are not real AI, just VP hype cycle artifacts.
That's not what I said.
I don't care who does this, I will despise it either way, I just hope that Apple at least has a proper justification to do this.
> Google need it because they want to remove all fingerprinting vectors, and need privacy-preserving replacements for legit use cases
"Legit" defined by whom? Google? I am sorry but at this point Google has burned almost all trust, at least with me.
To be honest, I've been interested in the domain for quite a time and I still want to build a fusor or a polywell at some point just to see it glow. Probably won't happen though.
Though I cannot help but wonder why exactly they did that. Some kind of a corporate requirement?
What's even more interesting is that the fusor - the simplest possible design for a thermonuclear reactor, so simple that anyone skilled in electrical engineering and having access to proper civilan equipment can build one with ease - seems to be invented _after_ both stellarator and tokamak.
That said, I never particularly liked stellarator design. The very _complexity_ of it somehow feels subtly wrong, like doubling down in the wrong direction.
However, this is one of the cases where I would absolutely love to be proven wrong. We are far past due big breakthroughs in the field.
In some workloads short-lived allocations are common. Like memory is allocated for milliseconds and then deallocated - and X% of time will be spent just inside the allocator trying to find the best block to allocate. At this point you want to make sure that you don't spend more time in the allocator than allocations actually live.
I wonder how common business tasks can be coerced to this model.
Also I believe this is not a good excuse - the language should still provide a way to properly manage allocations even if it is not needed most of time.
> The basic rule of manual optimization is that the upper bound for the improvement that can be gained by optimizing any one "thing" is equal to the proportion of the program's runtime that is devoted to that thing.
Yep, and this is why you should always measure things before optimizing. But then again, you can save yourself a lot of time by not doing stupid things like allocating from heap in a hot loop and profiling just to discover such basic mistakes.
> to use different allocators on a per-collection basis
It's far more than that. There are different memory management optimization patterns that are hard to implement in Rust, ie. splitting allocations to different arenas based on allocation lifetime.
> But Rust programs aren't allocating that much in the first place
Are Rust programs allocating mostly from the stack then?
> is going to triple the performance of arbitrary programs
The difference between malloc and any hand-rolled O(1) allocator is enormous, it's just people rarely benchmark this particular aspect.
There is a reason why Rust used to ship jemalloc and why #[global_allocator] is a thing. I would argue there's even more reason to avoid touching heap in the first place.
Edit: formatting.
Rust makes it especially hard to use custom allocators, see bumpalo for example [0]. To be fair, progress is being made in this area [1].
Theoretically one can use a "handle table" as a replacement for pools, you can find relevant discussion at [2].
[0] https://github.com/fitzgen/bumpalo
[1] https://github.com/rust-lang/rust/issues/32838
[2] https://news.ycombinator.com/item?id=36353145
Edit: formatting.
I would expect Rust to be 2x-3x slower in real production code, solely due to approach to memory allocation.
Just looking at the Wikipedia page for this standard makes it really clear why it has never seen widespread adoption (and I, for one, am really happy it didn't).
The project OP is referring to instead adheres to a very small, very clear spec, and not in the sense of implementing it. Instead, the project seeks to provide a set of tools using this limited spec.
I have my own set of criticisms for the OP, but this just doesn't make any sense.