IKantRead

> AI is (still) eating the world.

Citation needed.

I work in this space so am fairly optimistic, but it's worth pointing out we're still mostly talking about AI hype. Nothing has really been eaten at all yet from what I've seen.

The article claims that there are tons of startups in this space, which is true, but that doesn't mean any of these startups have actually solved any major problems yet.

As a reminder, we still haven't "solved" autonomous driving quite yet and we were much further along that road 5 years ago.

Working on shipping LLM driven products everyday, I'm becoming increasingly concerned that there's no way the products proposed can possibly match the hype for them. Which is a bit of a bummer since these models have a lot of potential in the sub-hype space, but I fear backlash in the future will lead us to squander that potential.

I think what you're describing is basically something I've pointed out in different contexts: for many areas concerned with AI, content is already effectively written by "AI", it's just humans following the algorithms rather than fully automated.

- content marketting is already AI writing, it's just humans churning through the SEO optimized algorithms.

- Mass digital stock photograph is AI generated "art". People that do that for a living have an exact formula for reproducing images that will sell en masse.

- As you point out, a good chunk of generic pop music is churned out by musicians following algorithms behind the scenes and then manufactured to look like the work of a public performer.

The parts of our content work threatened by AI are already cold, sterile, and machine generated.

As someone who has been doing a lot of work in SD, I just wanted to strongly second this.

Stable Diffusion and the like are much closer to when photoshop first appeared than they are to than the completely automatic art generation machine people make them out to be.

I remember well both digital cameras and photoshop receiving initial criticism from "real" artists... until they were both widely adopted by these communities.

The critique of digital photography was that it's not really art because you can just take a bajillion pictures and then filter out the ones that don't look good. Photoshop users (especially web comic creators) got similar flack that it didn't take real artistic skill to make nice images in photoshop.

For anyone young enough to always have photoshop and digital photography will likely be surprised that there was such resistance and snobbery around using these tools. There was also the same fear that these tools would ruin art by making "real" artists obsolete.

I was a sophomore in college, it was strange because people didn't have smart phones and many people didn't have cell phones. I remember class starting and a student said, very calmly after receiving a text message about, "oh, that's weird it says an airplane in the world trade center". We all assumed it was just a small private Cessna plane that must of accidentally bumped into the one of the towers, and then class began as usual.

My roommates and I spend the next week completely glued to the television. Which is why this interface is particularly great for capturing that feeling, but it is tough to rewatch.

> CEO says firm is smaller than ideal but positive for margins

Sounds like "quiet cutting"[0] to me, only I'm guessing the CEO's target was slightly less people quitting.

It's remarkable how quickly startups have become aggressively anti-worker.

0. https://www.wsj.com/articles/first-it-was-quiet-quitting-now...

Just yesterday I was thinking to myself "I wonder how long it will be until we start generating prompts using code and then we're just writing code again"

I recently came across an in-progress book from Manning, "Peer-to-Peer Web Applications" [0] that, despite a fair bit of crypto/web3 references, seems to be targeting an audience of people looking to build more robust/complex version of this.

I'd recently been thinking about how it's unfortunate that we have seen so little growth in the peer-to-peer part of the web, which to me is the most resilient strategy to resist a few large tech companies taking over everything and turning the web into a gigantic advertisement.

Curious if anyone has read the MEAP for that book so far and has any opinions on it?

0. https://www.manning.com/books/peer-to-peer-web-applications

Thank you for pointing this out! I have quite a few Chinese friends and they never cease to be entertained by the long list of "Chinese" proverbs that American's claim to know.

It turns out that since for most of Western history China was synonymous with the "mysterious far east", it become common practice to just tack on "Chinese" to the origin of any saying to let it exotic credibility.

Unsurprisingly "May you live in interesting times" is not a Chinese curse.

> The focus on the top 10 in vector search is a product of wanting to prove value over keyword search

I'm surprised to see less mention of the use of vector search in RAG. The reason top 10 (or less) is so important in RAG is that you only have so much context to work with (and even in large context models, you still would prefer to minimize the total amount of context you're sending the model).

Everyone I know working seriously on vector search is using it to plug into a prompt for an LLM. For this use case top 10 is a fairly important metric.

> You're saying, let's pull that lever.

First off, I'm not saying "starting tomorrow now fossil fuels!", I'm saying you must keep fossil fuels in the ground to avoid worst case. And the start of that means you have to identify some fossil fuels that will never be extracted.

> since we have ample time to do so.

I seriously have no idea where you got that impression, but with climate change once you start feeling the effects you're already in the danger zone. We've already signed up for famine and potentially billions dead even if we had zero emissions today.

I do suggest you read up a bit more on the topic. No serious research around climate change would claim we have "ample time".

The bigger question is: how close are we to climate "tipping points". We do know that geologically it seems there are points where positive feedbacks start accelerating climate change rapidly. We just don't know where the line is for those tipping points.

If we seriously want to avoid extinction of the species we should already be starting to strategize a schedule of what fossil fuels reserves we promise to keep in the ground, and what we're going to do about future discoveries.

To be honest, I don't really think we will do this, but if we wanted to survive at all we should start talking about it very seriously.

> I'm curious how you think that the damage of climate change can be worse than this.

Oh, much worse.

First off we, fertilizer or no, we're already facing crop failures in the US [0] and will continue to face even more extreme crop failures [1]. Even if we had unlimited fertilizer we've already signed up for massive famine.

But if you really want to talk about the unmitigated climate change path, which is what happens if we choose not to keep hydrocarbons in the ground, I really recommend reading Peter Ward's Under a Green Sky [2].

He's a respected geologist that makes a compelling case that the vast majority of mass extinction events were caused by rapid rises in CO2.

One of the realistic scenarios we're facing, he argues, is the break down of the AMOC ultimately leading to the oceans becoming anoxic and releasing hydrogen sulfide rather than oxygen. The entire marine ecosystem is essentially wiped out except for some cyanobacteria. The oceans are the foundation of all of our food systems. It would make the planet uninhabitable my most complex life of today. This has happened before in Earth's climate history.

All of the "bad-awful" but not-extinction event scenarios assume that we do not burn all of the fossil fuel reserves currently leased. We are already looking at a grim future, but given that we're rapidly pumping millions of years of stored CO2 into the atmosphere, upper bounds for the damage we can do are tremendous.

0. https://www.nytimes.com/2023/08/09/us/kansas-wheat-harvest-d... 1. https://www.aljazeera.com/news/2023/7/4/scientists-warn-of-c... 2. https://www.amazon.com/Under-Green-Sky-Warming-Extinctions/d...

Wait until you see what climate change does with continued carbon emissions!

But this has always been the problem with stopping climate change.

We are choosing between lesser, but certain immediate pain vs larger but less certain future pain.

In the 1880s we could have easily stopped climate changed, but been forced to slow growth. The pain would have been minimal for those living, but the probability of extreme catastrophe seemed remote so nothing was done.

As we move forward the probability of extreme catastrophe becomes more certain, but so too does the immediately pain.

I do believe that included in "only action" is: keep fossil fuels in the ground.

The idea that we can only mitigate is insane. Yes we've already signed up for a lot of pain as a civilization, but as long we keep burning fossil fuels we are looking at an increasingly worse future.

It's not like a smoker who has gotten lung cancer, where the best we can do is try to ease the pain, and more smoking won't really matter.

The biggest first step in preventing the worse case is to be able to point to an established oil field and say "that will never be extracted, and will remain in the earth for at least a 1000 years."

Every ton of carbon we can keep in the ground that way, the better of our future, even if that future is already slated to be bad.

I still chuckle a bit when people get anti-scrum, because I've been around long enough when agile/scrum was the disruptive thing.

Of course in practice agile/scrum/whatever has turned into the same cargo-cult nonsense that the original promoters of the ideals were fighting against.

One thing I thought was interesting was:

> First, the most common jobs among the people who told me I was wrong were "Agile Coach" and "Scrum Master." They feel very strongly in favor of Scrum, but I'm not sure why.

It seems funny to critique people's job titles when the poster's Twitter profile reads:

> I teach Machine Learning and run http://ml.school

I mean loudly critiquing a mainstream practice in a flippant way is pretty much the textbook approach to bring attention to yourself and establish as an expert in... teaching people about these topics. It's content marketing 101.

I don't think you should criticize others just because you've adapted to what is clearly a dark pattern.

I clearly see the "continue reading" now, but I definitely recall the first time I saw this popup not seeing it and closing out of the window because I thought substack was demanding signup.

I have no doubt they've a/b tested the hell out of this UX and patted themselves on the back when first time readers started subscribing at record rates... never bother to ask what this meant for the actual UX.

In the US, which is the region I'm citing for covid deaths, HIV causes ~20,000 deaths a year [0].

Last I checked people are quite considered with global HIV death, especially since they can obviously be reduced dramatically.

0. https://www.hiv.gov/hiv-basics/overview/data-and-trends/stat...

Or the million+ deaths in the US from Covid, which I have seen brushed off on HN as not a big deal.

It's remarkably how quickly people incorporate tragedy as normal when it doesn't fit their worldview. If you travel to 2000 and told people "in 2020 a pandemic will kill over 1 million Americans, and a good portion will brush it off as a 'nothing burger'" people would think you were being ridiculous and would dismiss it as unbelievable fiction.

> wet-bulb event

It's probably already too late, but I keep seeing people confusing the concept of "web bulb temperature" (all temperatures have a wet bulb temperature, even ones very pleasant for people) with "the human survivable maximum wet bulb". I frequently see people saying things like "when we reach wet bulb temps..", which feels a bit silly.

This always feels like talking about a "Kelvin event" or "Kelvin temps" referring a temperature reaching absolute zero.

> Crypto was a total scam.

As someone who have been very anti-crypto for a long time, it wasn't always a complete scam.

The first wave of the crypto boom, before anyone that wasn't a programmer had even heard of it, there was a lot of real work being done that very much mirrors current AI work. Lot's of very sharp developers learning about block chain, figuring out how to implement things, experimenting with ideas. Back then everyone owned their own wallet and you would meet at coffee shops to exchange cash for BTC.

Most of the serious engineers that were really into crypto during the first crypto boom of 2012 left in disgust when the second boom came around.

Having worked in AI/ML for a long time, I myself can start to see how they felt. We do have some really cool technology in front of us, I think it has a lot of potential, but so many of the loudest voices in this space are entirely out of touch with what's possible, and far more interested in hype and making money than the underlying technology.

I work in this space and think that it's far more rational to accept the axiom that LLM progress will not be significant than to bank product work on assuming it will increase drastically.

I do think we still have yet to squeeze the most value out of current LLMs, but most people's radical AI dreams are completely out-of-touch with reality for anyone working closely on these problems.

My biggest fear in this space is that disappointment in the inability of these tools to live up to the hype will cause people to irrationally abandon exploring the spaces where they do work.

I think they're HF averse because HF is fairly anti-NSFW content and it's pretty clear that civit.ai has a fairly large NFSW focused audience.

On top of that HF is just hard to use for your average user. Civit.ai is just "click to download" while HF is "look here's a broken model card... you can figure it out from here".

Despite the cute logo, I think most people find that HF comes across as fairly anti-user. Despite having years doing ML related work, I still find HF a bit byzantine to navigate.

Sampling is widely accepeted to have originated from the 1970-80s Hiphop scene, where traditional DJs (playing other people's music, like the do today) started to get clever with how they were combining different tracks. Rappers would then start reciting lyrics over these remixed version of well known funk and soul tracks. Although the wikipedia does a good job tracing it back even further to the 1940s[0]

Copyright issues only came up later once Hiphop had begun to see more mainstream acceptance (i.e. make more money) and influence other musicians.

They're may be cases where sampling is used to avoid other legal issues, but it's origins are most definitely in reusing existing music to create something new.

0. https://en.wikipedia.org/wiki/Sampling_(music)

I do find it interesting how little in the larger AI art conversation the idea of borrowing already exists in various art forms and each art forms unique culture of accepting this comes up.

In folk music for example, performing someone else's work is not disparaged at all historically. It was common for most musicians to travel around and teach each other songs.

In pop music we see sampling has increasingly become a major part of the art, and only really controversial when someone tries to avoid paying the original artist their fair share (which is itself a topic of debate).

In the visual arts it's much different. There is of course the famous Picasso quote: 'Good artists copy, great artists steal.' But even in the tone of that quote you can tell there's a different cultural acceptance of reusing work.

Writing is tricky because referencing or quoting another author without attribution is a common feature of "high brow" works. T.S Eliot for example borrows heavily form past authors, often without attribution, but assuming an audience which knows enough to not require attribution. However at the other extreme you can get in trouble for plagiarizing your own work.

Personally I think Diffusion models and LLMs are just bringing up an issue in the visual and literary arts that was already forced to the attention of musical artists during the rise of early hiphop djs who showed that you can genuinely create something new nearly entirely by cut and pasting existing works.

One thing I would point out: It seems to me that copying/borrowing/reusing are equally present in all creative fields, they just are willing to admit it to different degrees.

This is a great comment. I was self-taught and then went back to school, so I have pretty good experience with both paths.

I hear a lot of time from developers "I never use those fancy algorithms!" or "I've never needed any math beyond HS algebra!" but I find that very often it's precisely because programmers aren't familiar with those solutions that they don't see the areas they can be applied to.

One of the biggest insights from my after-self-taught CS degree was that one class in particular stands out as a the dividing line between feeling knowledgable and not: Compilers.

To anyone wanting to improve their CS background: If you can't get back to school full time, try to audit a course on compilers (this is one area where having a structured course really helps). Pretty much all aspects of Computer Science are touched on in that one area: algorithms, data structures, graph theory, theory of computation etc. Plus it's just a lot of challenging programming to be done.

But once you see a high level program that you wrote compiled to assembly by a compiler that your wrote every piece of, you really feel like you understand both CS and software.

AI will only replace the work of creatives who have already turned themselves into AIs long ago.

There are plenty of writers and illustrators out there who have trained themselves to churn out reproducible garbage over the years in order to fulfill the demands of content marketing. These jobs will be replaced by AI soon, but by creating content from a formula they've already be using a crude form of AI.

I really love Stable Diffusion, but, as a means of creating art (including the most common forms of popular art), it can only supplement existing work not replace it. I pay for plenty of real art in my home and the best works on my walls could never be replicated by an AI because what makes them beautiful is precisely the human touches that I have yet to see AI generate (and suspect it can't). Latent Diffusion Models also have a pretty poor imagination.

At the same time Stable Diffusion has gotten me thinking about creative projects I could undertake that would have been impossible years ago. But it's obvious that all of these projects will take plenty of work to create, and SD will ultimately just be another tool in the creative process.

I vividly remember when photoshop started to gain major acceptance and there was a similar anti-photoshop sentiment among "real" designers and artists. What's funny is how many webcomic authors I see critiquing AI art, when I remember quite well pen and ink comic artist similarly scoffing at web artists that used digital tools to create their work.

Hopefully we'll see SD and similar tools accepted as more tools to create cool art, rather than a misplaced focus of peoples career anxiety.

> It fails to deliver a consistent representatiuon of reality.

I'm curious by what precisely you mean by this. Last I checked, reality itself has problems consistently representing itself.

I really want to agree with this, but the trouble is that an unstated prerequisite is that you need a team of developers that have a strong sense of software design. A sensibility I've come to believe is drastically in decline in more junior software engineers (though don't mistake this for a statement that the quality of these devs appears any less).

Just take Ruby on Rails, which is probably a great example of a framework that runs into the problems you're describing. My experience with more contemporary, generally more junior, dev teams is that the "Rails" part of the framework is essential for these developers to maintain any loose semblance of adherence to any kind of architecture or design principles.

Earlier in my career I remember, even junior devs, spending a fair amount of time thinking about about the abstractions they were using, making design decisions, and often times being more guilty than not of over thinking abstractions.

Today I've seen so many very bright and talented engineers whose sole view of good software is minimizing the time from request to PR (and code reviews that strong favor minimizing the time from PR -> prod).

If you have a bunch of senior engineers (not SV "senior", but truly experienced), and they're the type that still pick up SICP form time to time, or spend a Saturday reading a section of TAOCP, then sure I think the framework-free approach is best. But you also don't need to tell that to a team of engineers like that.

Otherwise, frameworks and even languages that do the most handholding when it comes to design decisions are going to remain essential.

For a first pass I highly recommend reading the Ethics backwards, and only focusing on the collaries and the notes at the end of each section. Ignore the axioms and proofs for the first few passes through the book.

The content and notes at the end of sections are much easier to read and quite beautiful.

When I first read Ethics in the order presented I was deeply confused as to why Spinoza was considered the "prince of philosophers". After revising my approach he's rapidly becoming one of my favorite.

Despite it's presentation, the Ethics is much better approached in a very non-linear fashion.

edit: I also recommend Deleuze's Spinoza: Practical Philosophy. If you've read Deleuze's other work you may or may not be put off by this recommendation, but his work on Spinoza covers a lot of ground an is very readable.

The best advice I ever got regarding Spinoza's Ethics was to read it backwards, and focus on the corollaries (and notes at the end of sections).

If you start the Ethics at the end you find a great introduction to reading Spinoza:

> But all things excellent are as difficult as they are rare.

It's from this I also derive my belief that if you find Spinoza "approachable" then you probably haven't spent enough time with Spinoza.

Existential dread.

Nietzsche was right to be concerned with the "death of God". People have a very hard time dealing with the inevitability of our own death and the fundamental meaninglessness of our existence. Even people who think they are comfortable with these concepts tend to still have many tactics of avoiding this reality.

For most of human history we solved this problem by pointing to the past (e.g. "when I die I become one with my ancestors") or another world (e.g. "when I die I will leave this vale of tears, and find myself in the world I belong in").

For most of modern society we have replaced this with finding salvation in the future (e.g. "My life is meaningful because my work is contributing to a better future for humanity"). I think most of us, striving to work on hard technical problems, imagine ourselves as ultimately being a small screw in the generation ship that will send humanity into the stars. Or, at the very least, most hope to live on through their children.

Catastrophic climate change challenges that fundamental system of meaning many people have established to resist this terror of existence. If society as we know it has no future, this is roughly the same as telling a devote Christian that their is no heaven. No matter how much evidence you gather for this, at a deep level they will be unable to accept this. And if they are able to accept this, it will only because there has been an alternate method of meaning proposed.

This is why Nietzsche was so sincerely concerned with the death of God. We can't simply go back to being a world that derives meaning from ancestor worship, or from religion. Sure some subset of the population might "find God", but realistically we can't trivially go back. Nietzsche saw how fragile our systems of meaning had become and worried what society will look like when they start to break down en masse.

Nietzsche was write, these climate deniers and many other similar radicals are near perfect representation of Nietzsche's "Last Man" https://en.wikipedia.org/wiki/Last_man