>There isn’t a single day where I don’t have to deal with software that’s broken but no one cares to fix
Since when does this have anything to do with AI? Commercial/enterprise software has always been this way. If it's not going to cost the company in some measurable way issues can get ignored for years. This kind of stuff was occurring before the internet exists. It boomed with the massive growth of personal computers. It continues to today.
I think the point the author is trying to make is that there are many problems in plain sight we could be spending our efforts on, and instead we are chasing illusory profits by putting so many resources into developing AI features. AI is not the source of the issues, but rather a distraction of great magnitude.
> Commercial/enterprise software has always been this way
All software is this way. The only way something gets fixed is if someone decides it's a priority to fix it over all the other things they could be doing. Plenty of open source project have tons of issues. In both commercial and open source software they don't get fixed because the stack of things to do is larger than the amount of time there is to do them.
Thought exercise: has any of the money Apple has spent integrating AI features produced as much customer good-will as fixing iOS text entry would? One reason for paying attention to quality is that if you don't, over time it tarnishes your brand and makes it easier for competitors to start cutting into your core business.
Then you missed the point of my post. That money never did. It went back into the hands of the investors, the investors that are now putting money into genAI.
> What could have been if instead of spending so much energy and resources on developing “AI features” we focused on making our existing technology better?
This is a bit like the question "what if we spent our time developing technology to help people rather than developing weapons for war?"
The answer is that, the only reason you were able to get so many people working on the same thing at once, was because of the pressing need at hand (that "need" could be real or merely perceived). Without that, everyone would have their own various ideas about what projects are the best use of their time, and would be progressing in much smaller steps in a bunch of different directions.
To put it another way - instead of building the Great Pyramids, those thousands of workers (likely slaves) could have all individually spent that time building homes for their families. But, those homes wouldn't still be around and remembered millenia later.
> instead of building the Great Pyramids, those thousands of workers (likely slaves) could have all individually spent that time building homes for their families. But, those homes wouldn't still be around and remembered millenia later.
They would have been better off. Those pyramids are epitomes of white elephants.
I wonder about the world where, instead of investing in AI, everyone invested in API.
Like, surfacing APIs, fostering interoperability... I don't want an AI agent, but I might be interested in an agent operating with fixed rules, and with a limited set of capabilities.
Instead we're trying to train systems to move a mouse in a browser and praying it doesn't accidentally send 60 pairs of shoes to a random address in Topeka.
LLMs offer the single biggest advance in interoperability I've ever seen.
We don't need to figure out the one true perfect design for standardized APIs for a given domain any more.
Instead, we need to build APIs with just enough documentation (and/or one or two illustrative examples) that an LLM can help spit out the glue code needed to hook them together.
The problem with LLMs as interoperability is they only work sub 100% of the time. Yes they help but the point of the article is what if we spent 100billion on APIs? We absolutely could build something way more interoperable and that’s 100% accurate.
I think about code generation in this space a lot because I’ve been writing Gleam. The LSP code actions are incredible. There’s no “oh sorry I meant to do it the other way” you get with LLMs because everything is strongly typed. What if we spent 100billion on a programming language?
We’ve now spent many hundreds of billions on tools which are powerful but we’ve also chosen to ignore many other ways to spend that money.
Today I’ve compiled a few thousand classes of Javadocs in .978 second. I was so impressed, with a build over 2 minutes, each byte of code we write takes a second to execute, computing is actually lightening fast, just now when it’s awfully written.
Time of executing bytecode << REST APIs << launching a full JVM for each file you want to compile << launching an LLM to call an API (each << is above x10).
> LLMs offer the single biggest advance in interoperability I've ever seen.
> ... we need to build APIs with just enough documentation (and/or one or two illustrative examples) that an LLM can help spit out the glue code needed to hook them together.
If a developer relies on client code generated by an LLM to use an API, how would they know if what was generated is a proper use of said API? Also, what about when lesser used API functionality should be used instead of more often used ones for a given use-case?
If the answer is "unit/integration tests certify the production code", then how would those be made if the developer is reliant upon LLM for code generation? By having an LLM generate the test suite?
And if the answer is "developers need to write tests themselves to verify the LLM generated code", then that implies the developer understands what correct and incorrect API usage is beforehand.
Which begs the question; why bother using an LLM to "spit out the glue code" other than as a way to save some keystrokes which have to be understood anyway?
As if the challenges in writing software are how to hook APIs together.
I get that in the webdev space, that is true to a much larger degree than has been true in the past. But it's still not really the central problem there, and is almost peripheral when it comes to desktop/native/embedded.
Agree. I often prefer to screen scrape even when an API is available because the API might contain limited data or other restrictions (e.g. authentication) that web pages do not. If you don't depend on an API, you'll never be reliant on an API.
Basically the opposite has happened. Not only has every API either been removed or restricted. Every company is investing a lot of resources in making their platforms impossible to automate even with browser automation tools.
Mix of open platforms facing immense abuse from bad actors, and companies realising their platform has more value closed. Reddit for example doesn't want you scraping their site to train AIs when they could sell you that data. And they certainly don't want bots spamming up the platform when they could sell you ad space.
We work with American health insurance companies and their portals are the only API you’re going to get. They have negative incentive to build a true API.
LLMs are 10x better than the existing state of the art (scraping with hardcodes selectors). LLMs making voice calls are at least that compared to the existing state of the art (humans sitting on hold.)
The beauty of LLMs is that they can (can! not perfectly!) turn something without an API into one.
I’m 100% with you that an API would be better. But they’re not going to make one.
I feel like it’s not technically difficult to achieve this outcome… but the incentives just aren’t there to make this interoperable dream a reality.
Like, we already had a perfectly reasonable decentralized protocol with the internet itself. But ultimately businesses with a profit motive made it such that the internet became a handful of giant silos, none of which play nice with each other.
While I'm somewhat sympathetic to this view, there's another angle here too. The largesse of investment on a vague idea means that lots of other ideas get funding, incidentally.
Every VC pitch is about some ground-breaking tech or unassailable moat that will be built around a massive SAM; in reality early traction is all about solving that annoying and stupid problem your customers hate doing but that you can do for them. The disconnect between the extraordinary pitch and the mundane shipped solution is the core of so much business.
That same disconnect also means that a lot of real and good problems will be solved with money that was meant for AGI but ends up developing other, good technology.
My biggest fear is that we are not investing in the basic, atoms-based tech that we need in the US to not be left behind in the cheap energy future: batteries, solar, and wind is being gutted right now due to chaotic government behavior, the actions of madmen that are incapable of understanding the economy today, much less where tech will take it in 5-10 years. We are also underinvesting in basics like housing, or construction tech. Hopefully some of the AI money goes to fixing those gaping holes in the country's capital allocation.
If a million families each has a $1,000 to invest in new business, how would you envision the money to be invested collectively? what would be the process?
it’s peculiar because i love to use chat gpt to fill my knowledge gaps as i work through solutions to building and energy problems that i want to solve. i wonder how many people are doing something similar and, although i haven’t* read through all the comments, i doubt much is being said let alone giving credence to that simple but potentially profound idea. learning amplified.
The reply defining terms from busterarm was flagged, so I'm repeating them here:
> TAM or Total Available Market is the total market demand for a product or service. SAM or Serviceable Available Market is the segment of the TAM targeted by your products and services which is within your geographical reach. SOM or Serviceable Obtainable Market is the portion of SAM that you can capture.
More less you could say similar things about most of the crypto space too. I think maybe it's because we're at the point where a lot of things that tech can do, it's more than capable of doing, but they're just not easy to do out of a dorm room and without a lot of domain knowledge.
There is still so much one can build and do in a dorm room. The hardest part is still the hardest part in every business, which is getting sufficient money to get sufficient runway for things to be self sufficient.
>What could have been if instead of spending so much energy and resources on developing “AI features” we focused on making our existing technology better?
I've been watching this my whole life. UML, SOA, Mongo, cloud, blockchain, now LLMs, probably 10 others in between. When tools are new there's a collective mania between VCs, execs, and engineers that this tool unlike literally every other one doesn't have trade offs that make it only an appropriate choice in some situations. Sometimes the trade offs aren't discoverable in the nascent stage, a lot of it is monkey-see-monkey-do which is the case even today with React and cloud as default IMHO. LLMs are great but they're just a tool.
The big difference is LLMs are as big as Social Media and Google in the pop culture, but with a promise of automation and job replacement. My 70 year parents use it every day for work and general stuff (with generally understanding the limitations), and they’re not even that tech savvy.
We haven’t mapped the hard limitations of LLMs yet but they’re energy bound like everything else. Their context capacity is a fraction of a human’s. What they’ll replace isn’t known yet. Probabilistic answers are unacceptable in many domains. They’re going to remain amazingly helpful for a certain class of tasks but marketing is way ahead of the engineering, again.
IoT wasn't exactly a waste of money. If anything, the problem was that companies didn't spend enough doing it properly or securely. People genuinely do want their security cameras online with an app they can view away from home. It just needs to be done securely and privately.
I have 4 cameras, a home security system, a remotely monitored smoke detector, a smart plug, 4 leak sensors, smart bulbs, a car whose location and state of charge I can track remotely, a smart garage door opener, a smart doorbell, and 7 smart speakers.
The author doesn't seem to appreciate that investors aren't incompetent, but malicious.
Investing 100 years of open-source Blender does not give them any fraction of monopoly.
Even if scientists present 100's of proposals for computation (optical, semiconductor,...) they will specifically invest in technologies that are hard to decentralize: growing monocrystalline ingots, reliant on dangerous chemicals, ... if there is no money in easily decentralizable processor manufacture, then it could easily be duplicated then proposals to pursue it would basically be equivalent to begging investors to become philantropists. Quite a naive position.
It's in the interest of the group to have quality software, manufacturing technologies, ... so the onus is on representatives of the group of taxpayers to invest in areas investors would prefer to see no investment in (even if someone else invests it). Perhaps those "representatives" are inept or malicious or both.
There is real value being created by creating interactive summaries of the human corpus. While it is taking time to unlock the value, it will definitely come.
Readit News
Since when does this have anything to do with AI? Commercial/enterprise software has always been this way. If it's not going to cost the company in some measurable way issues can get ignored for years. This kind of stuff was occurring before the internet exists. It boomed with the massive growth of personal computers. It continues to today.
GenAI has almost nothing to do with it.
All software is this way. The only way something gets fixed is if someone decides it's a priority to fix it over all the other things they could be doing. Plenty of open source project have tons of issues. In both commercial and open source software they don't get fixed because the stack of things to do is larger than the amount of time there is to do them.
Things that are easy, fun, or "cool" are done before other things no matter what kind of software it is.
FaceID has proven pretty popular tools.
This is a bit like the question "what if we spent our time developing technology to help people rather than developing weapons for war?"
The answer is that, the only reason you were able to get so many people working on the same thing at once, was because of the pressing need at hand (that "need" could be real or merely perceived). Without that, everyone would have their own various ideas about what projects are the best use of their time, and would be progressing in much smaller steps in a bunch of different directions.
To put it another way - instead of building the Great Pyramids, those thousands of workers (likely slaves) could have all individually spent that time building homes for their families. But, those homes wouldn't still be around and remembered millenia later.
They would have been better off. Those pyramids are epitomes of white elephants.
This was not the original intent of the construction though.
Yes, but they'd have homes. Who's to say if a massive monument is better than ten thousand happy families?
It's not. The pyramids have never been of any use to anyone (except as a tourist attraction).
I'm referring merely to the magnitude of the project, not to whether it was good for mankind.
Like, surfacing APIs, fostering interoperability... I don't want an AI agent, but I might be interested in an agent operating with fixed rules, and with a limited set of capabilities.
Instead we're trying to train systems to move a mouse in a browser and praying it doesn't accidentally send 60 pairs of shoes to a random address in Topeka.
We don't need to figure out the one true perfect design for standardized APIs for a given domain any more.
Instead, we need to build APIs with just enough documentation (and/or one or two illustrative examples) that an LLM can help spit out the glue code needed to hook them together.
I think about code generation in this space a lot because I’ve been writing Gleam. The LSP code actions are incredible. There’s no “oh sorry I meant to do it the other way” you get with LLMs because everything is strongly typed. What if we spent 100billion on a programming language?
We’ve now spent many hundreds of billions on tools which are powerful but we’ve also chosen to ignore many other ways to spend that money.
Time of executing bytecode << REST APIs << launching a full JVM for each file you want to compile << launching an LLM to call an API (each << is above x10).
> ... we need to build APIs with just enough documentation (and/or one or two illustrative examples) that an LLM can help spit out the glue code needed to hook them together.
If a developer relies on client code generated by an LLM to use an API, how would they know if what was generated is a proper use of said API? Also, what about when lesser used API functionality should be used instead of more often used ones for a given use-case?
If the answer is "unit/integration tests certify the production code", then how would those be made if the developer is reliant upon LLM for code generation? By having an LLM generate the test suite?
And if the answer is "developers need to write tests themselves to verify the LLM generated code", then that implies the developer understands what correct and incorrect API usage is beforehand.
Which begs the question; why bother using an LLM to "spit out the glue code" other than as a way to save some keystrokes which have to be understood anyway?
I get that in the webdev space, that is true to a much larger degree than has been true in the past. But it's still not really the central problem there, and is almost peripheral when it comes to desktop/native/embedded.
Mix of open platforms facing immense abuse from bad actors, and companies realising their platform has more value closed. Reddit for example doesn't want you scraping their site to train AIs when they could sell you that data. And they certainly don't want bots spamming up the platform when they could sell you ad space.
LLMs are 10x better than the existing state of the art (scraping with hardcodes selectors). LLMs making voice calls are at least that compared to the existing state of the art (humans sitting on hold.)
The beauty of LLMs is that they can (can! not perfectly!) turn something without an API into one.
I’m 100% with you that an API would be better. But they’re not going to make one.
Like, we already had a perfectly reasonable decentralized protocol with the internet itself. But ultimately businesses with a profit motive made it such that the internet became a handful of giant silos, none of which play nice with each other.
Every VC pitch is about some ground-breaking tech or unassailable moat that will be built around a massive SAM; in reality early traction is all about solving that annoying and stupid problem your customers hate doing but that you can do for them. The disconnect between the extraordinary pitch and the mundane shipped solution is the core of so much business.
That same disconnect also means that a lot of real and good problems will be solved with money that was meant for AGI but ends up developing other, good technology.
My biggest fear is that we are not investing in the basic, atoms-based tech that we need in the US to not be left behind in the cheap energy future: batteries, solar, and wind is being gutted right now due to chaotic government behavior, the actions of madmen that are incapable of understanding the economy today, much less where tech will take it in 5-10 years. We are also underinvesting in basics like housing, or construction tech. Hopefully some of the AI money goes to fixing those gaping holes in the country's capital allocation.
The elephant in the room is that capital would likely be better directed if it was less concentrated.
A surface-to-air missile?
As funny as that would be, maybe you should define your terms before you try to use them.
> TAM or Total Available Market is the total market demand for a product or service. SAM or Serviceable Available Market is the segment of the TAM targeted by your products and services which is within your geographical reach. SOM or Serviceable Obtainable Market is the portion of SAM that you can capture.
Dead Comment
I think we'd still be talking about Web 3.0 DeFi.
I think IoT was more than just hype.
Investing 100 years of open-source Blender does not give them any fraction of monopoly.
Even if scientists present 100's of proposals for computation (optical, semiconductor,...) they will specifically invest in technologies that are hard to decentralize: growing monocrystalline ingots, reliant on dangerous chemicals, ... if there is no money in easily decentralizable processor manufacture, then it could easily be duplicated then proposals to pursue it would basically be equivalent to begging investors to become philantropists. Quite a naive position.
It's in the interest of the group to have quality software, manufacturing technologies, ... so the onus is on representatives of the group of taxpayers to invest in areas investors would prefer to see no investment in (even if someone else invests it). Perhaps those "representatives" are inept or malicious or both.
There is real value being created by creating interactive summaries of the human corpus. While it is taking time to unlock the value, it will definitely come.