smokel

45 ms·

541.
by smokel 2y ago
If I understand you correctly, you mean to build up one test suite, and apply it to many diverse applications? I'm afraid that will be pretty hard to accomplish, given that such requirements are not easy to distill from the user interf
542.
by smokel 2y ago
This is a great example of where unit testing does not apply. I see many developers get caught up in rituals, and polishing (and monkey testing for that matter) seem to go against a reproducible approach, and are therefore frowned upon and
543.
by smokel 2y ago
Thank you for taking the time to write this response. Unfortunately, even though I agree that tokenization makes it pretty hard for the LLM to count characters, I'm still not convinced that it is a fundamental problem for doing so. I
544.
by smokel 2y ago
Nope, I'm not missing that particular fact. I'm aware that sentences (and words) are split into tokens, which are vectors. I don't understand how most LLMs can spell out words though, nor do I understand what is causing th
545.
by smokel 2y ago
I'm aware of this. The network could, and apparently does, associate single characters with words. It can associate "red" with "rose", and might associate "r" with "straw", and it might even as
546.
by smokel 2y ago
The "waterfall" way of working still has some merit. Think first, then make a design, and a planning based on that, and then implement the fucking owl. Unfortunately, a lot of unforeseen problems show up during the implementation
547.
by smokel 2y ago
This reasoning is interesting, but what is stopping an LLM from simply knowing the number of r's _inside_ one token? Even if strawberry is decomposed as "straw-berry", the required logic to calculate 1+2 seems perfectly withi
548.
by smokel 2y ago
What makes Holly Herndon and Mat Dryhurst so special? There are tons of artists working with AI, some even before these two particular artists were even born.
549.
by smokel 2y ago
The article addresses your point: > Now, is any of this true art, you might ask, or is it merely entertainment? I’m not sure it matters. Chiang dismisses the value of generative AI for either, defending the craft required for supposedly
550.
by smokel 2y ago
Or is it an obscure reference to the Dutch demogroup "Aardbei", most famous for their 64k intro "please the cookie thing" (2000)? https://m.youtube.com/watch?v=ycmgjZLU0xQ
551.
by smokel 2y ago
How do you define single creator? The person who typed the first keystroke? It seems pretty silly to claim that Algol 60 has thirteen creators. Sure, Wikipedia lists that many names in some table, but that is not really a significant stat
552.
by smokel 2y ago
There are also theories that they will prevent tinnitus because one can play music at a lower volume and still enjoy it.
553.
by smokel 2y ago
I see. But what about adding new knowledge to the models? To be useful in practice, one must train with new data, say every year or so. Who is going to pay for harvesting and cleaning all that data? And as I understand it, some form of R
554.
by smokel 2y ago
> What the fuck is wrong with us? This is simple: we can't just trust each other. When programming started, people were mostly interested in building things, and there was little incentive to spoil other peoples work. Now there
555.
by smokel 2y ago
At least they acknowledge this: "However, the table also shows that for all outcomes (with the exception of the Build Success Rate), the standard deviation exceeds the pre-treatment mean, and sometimes by a lot. This high variability w
556.
by smokel 2y ago
Interestingly, they compare number of pull requests as a statistic for productivity. Not that I know of a better metric, but I wonder if this is an accurate metric. It seems similarly misguided as looking at lines of code. If an AI tool m
557.
by smokel 2y ago
Training and inference are two different things. The incentive to make some model open is unrelated to hardware cost. Perhaps you are missing something in my arguments that I can clear up?
558.
by smokel 2y ago
Before you check yourself in the mirror, note that the article suggests that your pupil size should be larger for improved intelligence, not smaller , which is probably, but not scientifically "proven" to align with cunningness
559.
by smokel 2y ago
Does anyone know why the sizes of these models are typically expressed in number of weights (i.e 1.5B and 9B in this case), without mentioning the weight size in bytes? For practical reasons, I often like to know how much GPU RAM is requi
560.
by smokel 2y ago
What is your argumentation for that? It seems that building LLMs is very costly, and that will probably remain so for quite some time. I think it is equally likely that LLMs will perform worse in the future, because of copyright reasons an
561.
by smokel 2y ago
Along with the switch to Dvorak some decades ago, I remapped CapsLock to be an additional Ctrl, and I started using Emacs key bindings (C-n, C-p, C-f, C-b) instead of the arrow keys. I'm still typing in Dvorak, but I'm almost sure
562.
by smokel 2y ago
If you have doubts, don't switch. It saves you a lot of trouble when you have to work on someone else's system. Dvorak has the advantage of being supported on most operating systems, now, and possibly some decades into the future
563.
by smokel 2y ago
I just asked ChatGPT 4o to generate a Sudoku solver in Rust, and the code compiles and works fine. Edit: tried with Llama 3.1 70B as well, and funnily enough it outputs the exact same Sudoku puzzle as an example to test the code with.
564.
by smokel 2y ago
I'm afraid that is a matter of perspective. This snapshot is from 2007 and the "dot com bubble" burst somewhere around the year 2000. I still see a lot of cool ideas (healthcare, robots, thought organizers) and a lot of idea
565.
by smokel 2y ago
Running LLMs on that kind of hardware will be very slow (expect responses with only a few words per second, which is probably pretty annoying). LM Studio [1] makes it very easy to run models locally and play with them. Llama 3.1 will only
566.
Interactive MNIST Explorer(nathom.dev)
1 points by smokel 2y ago|0 comments
567.
by smokel 2y ago
> We know as a hard fact that LLMs do not understand anything. A bit late to the party, but we most certainly do not even know what "understanding" means.
568.
by smokel 2y ago
No. I just checked the front page [1] for each August 21 from 2014 to 2024, and there is typically no or one complaint in the title. Some lamenting here and there, an uplifting tone every now and then, but most of it is neutral. Perhaps y
569.
by smokel 2y ago
Placing statements on separate lines does come off as somewhat aggressive. It may be subtle, but I got a similar impression.
570.
by smokel 2y ago
To make the point evenly round and extremely sharp, one can use fine sandpaper as a last step. For this kind of work with graphite the sharp point holds for quite a while. Color pencil artists tend to sharpen a lot more, and use automatic

More ›