love2read

This answer makes sense if you know that LLMs have layers, if you don't this answer is not super informative.

If I were to describe this to a nontechnical person, I would say:

LLMs are big stacks of layers of "understanders" that each teach the next guy something.

Imagine you are making a large language model that has 4 layers. Each layer will talk to it's immediate neighbor.

The first layer will get the bare minimum, in the LLM's of today, that's groups of letters that are common to come up together, called "tokens". This layer will try to derive a bit of meaning to tell the next layer, such as grouping of letters into words.

The next layer may be a little bit more semantic, for example interpreting that the word "hot" immediately followed by the word "dog" maps to a phrase "hot dog".

The layer after that, becoming a bit more intelligent given it's predecessors have already had some chances at smaller interpretations may now try to group words into bigger blobs, such as "i want a hot dog" as one combined phrase rather than a set of separated concepts.

The final layer may do something even more intelligent afterward, like realize that this is a quote in a book.

The point is that each layer tries to add a little meaning for the next layer.

I want to stress this: the layers do not actually correspond to specific concepts the way I just expressed, the point is that each layer adds a bit more "semantic meaning" for the next layer.

Is it becoming a thing to misspell and add grammatical mistakes on purpose to show that an LLM didn't write the blog post? I noticed several spelling mistakes in Karpathy's blog post that this article is based on and in this article.

How is posting on this website with your full name not career suicide?

> How can I reasonably trust that the tools I'm using are built in a sound manner, when the barrier to producing good-looking bad code is so low

Honest answer: You never could.

So funny when people point at electron as if it singlehandedly makes every program unusable.

Also, I would assume there are not many significant pages on $B/Trillion companies that take 5 seconds to load text that are used frequently.

> I know I'm tired of reading them, but don't people get bored of writing them?

People never get tired of reading or commenting on commentary on their hobbies.

I think what you are finding is that people's definition of "Good Code" differs the same way two people's definition of "good food" differs.

> IMO, you need to have the capacity to write Good Code to know what Good Enough Code is. I completely agree, and its one of the biggest problem of trying to talk about "how you use agents". A lot of the people that may use the same agents with the same workflow may see wildly different results depending on their ability to evaluate the end result.

> The problem with this argument is many do not believe this sort of leverage is possible outside of a select few domains, so we're sort of condemned to stay at a low level of abstraction.

I think theres a similar tangential problem to consider here: people don't think that they are the person to create the serious abstraction that saves every future developer X amount of time because its so easy to write the glue code every time. A world where every library or API was as well thought out as the virtual memory subsystem would be an overspecified but at the same time enable creations far beyond the ones seen today (imo).

> Even the SICP authors admitted programming had become more about experimentation and gluing together ready-made parts than building beautifully layered abstractions which enable programs to just fall out of easily.

> Either sleeping away carefully on the craft of your code manually writing it, or shitting out first pass vide-coded stuff without really taking care to specify the problem or iterate on the code afterwards.

I think the thing you are missing is that people are

> shitting out first pass vide-coded stuff without really taking care to specify the problem or iterate on the code afterwards

to assume that people will take a path other than the path of least resistance now when they never did before, such as copy-pasting directly from stackoverflow without understanding the implications of the code.

none of this even kind of addresses why the article implies that people stopped writing good code. why are we going to spend "a lot more reading code than before"? is this an ai generated comment?

the rise of "good enough" was the death of "good code"

Every single blogger is running to make a post talking about how ai can visit websites now in different words.

I went to a university class in my last year that had the same “choice”. Any time the professor starts asking about “ai” tends to be worrisome in my experience.

Could this be used to make something like same.energy?

I love this, thank you

Would that just be the ability to chat without making new memories while using existing memories?

Cool, who's the audience?

is this ai generated? i literally don't know most of the words in this post...

while you are making up connections in your mind I made revisions for designs for a business using nano banana 10x in the same hour the logo designer i contacted told me he would only be available in december. i don't get what you are fighting for? a slower, less resourceful people?

Crazy that so many seem to be so against remodeling the whitehouse.

It doesn't matter. When I need to insert a screw I don't stick to a screwdriver because it gives me more exercise, I use a drill, because I can and I have access to one, and that's all there is to it. Any time you find yourself justifying people use a less powerful tool for a job you are probably in the wrong...

> Are engineers this terrified of originality that celebrating generic imagery would become a battle cry of mediocrity?

I think it's OP who is more terrified of the originality of letting anyone create with AI.

> This is unusual, a drive to embrace uniformity, and perhaps this is what internet pluralization is about. Those lacking visual imagination would come together celebrate mediocrity of imagery as a battle-cry, rally around software that roboticizes creativity.

It's almost comical how opposite my read of this was. Forcing people to "learn how to draw" is embracing uniformity rather than letting them create with whatever tool is available to them.

> The tone of these replies seems to reveal a rage that's about defending the right to eviscerate creativity.

Is creating tools to make it easier to create really the evisceration of creativity or the creation of new venues to create? In my mind it's the latter.

> It’s as artificial and soulless as corporate style.

Even if this is true (it's not, corporate style has much less soul because it has to comply with the clean corporate image), so what? Who cares that it's soulless? Corporate image is used all the time, why can't AI be too?

> You want a drawing of something? Learn how to draw. Sure, it will consume some time, money, and goodwill, but I promise you, it will be a billion times more valuable than something fabricated.

I don't really think you understand: a) how lazy people are b) how worthless the time to do a task is if there is a machine that can do good-enough work for a fraction of the time and cost

> Don’t get me started on searching for an image of something via a search engine. It’s filled to the brim with AI images if you don’t turn the filter on, and even if you do, it’s imperfect and still shows some.

Seems like a different issue?

> GPT-4 consumed tens of millions of water.

You're gonna make me explicitly write that water can't be consumed? You should have put this article through chatgippity.

> the amount of water used will soar persistently

You did get one thing right here, the water is used, not consumed...

> In the fourth generation, to generate a 100-word email, it consumed about 500ml of water (17 oz). So writing a 400-word email or re-doing a 100-word one will consume as much water as an average human needs per day ≈ 2000ml (½ gallon)

Even if this is true, the common link is that both "usages" of water will eventually be returned. There is no permanent consumption here of water.

> Even if there was some crazy optimisation reducing the use down to 50% it’s still approximated to be 300 people's daily water amount needed for some dumb AI video.

One man's trash is another man's treasure.

This article is very much man yelling at clouds. It really is true that even an atheist has a religion, and this person's religion seems to be a hatred against AI. Which is fine, but recognize that you don't have any actual reasons, what you have is just an unhappiness with the general climate of the world.

1) I would argue that the year that you created your bank account is not a good proxy for age. 2) I would question where you think the uuid representing your age from your bak would leak to considering it’s still a bank account id 3) I would question whether you consider that the vast majority of uuids aren’t used for high stakes ids such as online banking ids

My first thought too!

Looks great!

The design looks great!

Of course not, the alternate is writing these projects in a more maintainable language

they could have provided a seperate api over top of this and error’d on use? It’s a clearly error-prone api.

does this actually mean anything more than using a type system?

*though it could have been fixed by typescript had they cared.