>>110325372
Is AI not the advancement? What the fuck is he talking about?

Post more. Like a paragraph n sheeeiiit you fuck.

>>110325372
>If AI solved coding why is my software still the same
This unironically reads like one of those "If god isn't real why did daddy touch me at night?" Charlie Kirk shitposts.
>Jonathan Blow
Oh yeah right.

>>110325384
try some thinking on your own some time
maybe if you try really hard you can figure it out

>>110325372
>Still the same
Its all worse now. The only thing I've seen from AI is shitty barely functioning clones of existing programs and
>WOWEE ZOWEE DOOD WHAT IF MARIO WAS IN ELDENRING WITH A SKATEBOARD :O
bullshit

>>110325372
wtf I got the same video recommended

>>110325372
nobody is ever going to have an opinion against what he says in this video because only his fanboys would watch him talk for 1 hour and 40 minutes.

>>110325442
It is almost like there is an algorithm behind it all and some (actors) behind it can push certain content

>we are at the beginning
>"why hasn't everything changed already? why aren't we on the moon with flying cars?"
amazing question

>>110325479
>AI NPC:
>BASICALLY ALREADY AGI
>WORKERS ARE OBSELETE
>EVERYTHING 10X NOW
>AI NPC when pressed why no one ever made anything good with AI:
>It's still early, you can't expect such rapid progress

It's not the same though, it's clearly worse

>>110325372
>game coder
not a real coder, his opinion is worthless.
a megabrain such as himself wouldnt waste his intellect on silly puzzles for kids

>>110325547
>>AI NPC:
>>BASICALLY ALREADY AGI
nice headcanon

>>110325372
>If cars solved transportation why is my horse still the same
Grandpa asks

>>110325844
>sam altman every month for past 3 years:
>AGI in 2 weeks our next model is AGI trust me bro

yada yada if you had a job you'd be in the know. these things have only gotten to the point of "close your eyes and let it do it" around june or so and there's still plenty of way to go in terms of model improvements and harnesses, and everyone is only just learning on how to use it effectively. obviously it's still early

>>110325610
>silly puzzles for kids
You probably though the same about gaming GPUs, yet here we are.

>>110325402
One would at the very least expect software to have less bugs if AI is doing as good a job as we are told.

>>110325935
>muh Scam Saltman
Yawn. What's next? Seething about Musk being a fraud? Yeah no shit Sherlock. Get new material.

>large organizations can rapidly transform and change course
jon blow sure seems to knows how the world works

>>110325872
It's be more like asking *if cars are so great why haven't they improved logistics? "
Of course motorvehicles very much improved logistics compared to before their invention, but AIs haven't made any noticeable improvement in any major software and if anything it has only hastened enshittification.

>>110326004
Right now we're in a stupid early era where Sukdeep Dikshit just tells the agent to
>Saar make bug go away
And they just merge whatever the fuck it does in response.
But these things are so immensely useful I got turned into a believer over the past few months. I can do 3d modeling so much faster, I can do animation and so much more I couldn't do just a while ago and it costs me $50 per month

>>110325959
>obviously it's still early
>>110325479
>we are at the beginning

Is this the best you've got?

>>110326050
blow's retarded question doesn't need any other answer

>>110326044
Yes, it can work fast to create something mediocre. For code that quickly adds up to incredible amounts of slop that it can't handle after a while and for 3D modeling that means quickly making passable 3D models.

>>110326044
i've seen a colleague do this
he hastily types something akin "make it easier to understand" without any further input. prompting is the new programming and there's already retardation going on

>>110326071
Why?
Steel clad ships instantly made all wooden ships obsolete, and it was obvious why.
AI coding agents are supposed to be even more transformative and yet bugs and outages only seems to increase in frequency.

>>110326037
>enshittification
As per Cory Doctorow, who coined that word, it means when a middleman gains a dominant market share, and then squeezes both sides to extract maximum value.
That is not what's currently happening with AI companies at all, as they are still in the growth phase.
You simply meant to say that the use of AI has made software worse, but decided to use a buzzword, whose actual meaning is unknown to you.

>>110326004
Where do you think the flood of bug discoveries are coming from? Do you really believe people just suddenly decided to do open source bug hunting in droves? You're complaining about the very thing you're currently living through.
For another example see Valves recent output in patches which has been revealed to also have been AI assisted. Companies know better than to announce they're using AI, they're still using it evidently.

>>110326112
>AI coding agents are supposed to be even more transformative and yet bugs and outages only seems to increase in frequency.
because the people using it have not yet learned how to use it effectively (that includes management who pushes it this way). very simple

>>110326087
larger problems can be divided and conquered as usual. it's all skill issue

>>110326112
i'm sure early ironclad ships were shitty at first.
the same analogy for cars works against you:
early cars could only go 20mph, were completely unsafe and shitty, whereas today today 100 years later, they can go 300mph

>>110326115
I was talking about the effects of AI agents replacing programmers, this has hastened enshitification.
If you didn't rely on AI to summarize my post but instead read it yourself that would've been clear.

>>110326037
>but AIs haven't made any noticeable improvement in any major software
discord was rewritten so it isn't electron anymore :^)

>>110326115
Yes, AI has found and can find many bugs and this is a good use case but they also introduce a lot of logical bugs and unintended consequences when projects get vastly enlarged codebase due to all the slop they generate.
This will only get worse if code bases continue to get larger.

>>110326037
>but AIs haven't made any noticeable improvement in any major software
AWS just released an AWS emulator that you can develop against locally

>>110326163
>early ironclad ships were shitty at first.
They weren't, the first ships beat every wood ship they faced without breaking a sweat and AI is sold as the same thing for coding, we hear coding is solve every day.
Bill Gates said it a few days ago.

>>110326215
>they also introduce a lot of logical bugs and unintended consequences
when they are misprompted. most often people don't write detailed prompts and leave out design decisions that should be written explicitly because they are lazy and they think the LLM will figure out what they want. They are essentially misprogramming the LLM, which is what introduces bugs, not the LLM itself

>>110325403
No, asshole. Stop making vague low effort threads about other people's assertions, if you want to accuse anyone of lazy thinking. Whole fucking topic's creation was off loaded to another platform and person. Retard.

>>110326213
Rewrites are nice and might be a good use case, but the consequences remains to be seen.
Now no one is really familiar with the code base and who knows how difficult future development will be.
>>110326235
And this is completely AI developed and they did not have any plan for this before they stated using AI coding agents.

>>110326259
>when they are misprompted. most often people don't write detailed prompts and leave out design decisions that should be written explicitly because they are lazy
Do you expect this inherent flaw of humans to suddenly disappear?

>>110325372
why can't you post the video?

>>110326280
good engineers will take it to the moon and bad ones won't.

>>110326273
>but the consequences remains to be seen
fair, but the bun rewrite to rust made a bunch of people throw a hissyfit and it's been totally fine and i think sumner's landing another million line PR soon as well. that said, he does work at anthropic so he has infinite tokens. but tb h a company the size of discord would also have effectively infinite toks

>>110326318
>made a bunch of people throw a hissyfit
Why? I don't use Discord but I know the client was anything but popular with people, are they now making up reasons why actually electron was good for you?

>>110326308
Good engineers won't exist if managers don't see the difference in output between proper engineers and semi-retards.
It'll lead to a lower code quality and more slop, "proper engineers" will only decrease and decrease in frequency.

>>110325372
He got that Andrew Tate chin going on lmaeo

File: Little_DebHee-Hee-68504710.jpg (100 KB, 750x750)

100 KB JPG

>>110325372
It just made it easier for programmers who know what they're doing to make shit. And possbile for some people who don't know what they're doing to make simple or low quality shit.

A tool is a tool. Nothing more and only as good as the hand that holds it.

>>110326341
you're just making shit that makes you feel good. obviously good engineers will continue to exist because humans are being born every day and many of them are smart and they will find effective uses for the tools they have access to. prompting is the new programming and this is a brand new field, and you're asking why people are unskilled at it yet

>>110326340
i meant hissyfit about bun rewrite, not the discord one. almost everyone except vehemently anti-ai people are happy about the discord one.

>>110326374
Oh my god I'm fucking retarded I misread sorry. Still don't get that one either.

>>110325384
I've listened to half of it, he mourns the ever growing complexity of software combined with the low experience of common devs, being thrown into deep water of it, so to speak.

>>110326371
No, I speak from work experience.
If you work at a company as a programmer you most likely have a manager.
That manager most likely won't understand anything about programming and can't understand why problem A is simple and problem B is complex and can't understand why solution A might be good and solution B might be bad.

Sometimes bad programmers say soluton B is good to the manager and since the manager can't understand the difference between a good and a bad programmer or a good and bad soluton you you will have to implement solution B.

Now imagine if the managers think almost anyone can code because you just say your problem to the LLM. this is where we pretty much are at my company and I am pretty fucking sure we are not alone

the rate of advancement will not be uniform but proportional to the degree to which the individual and organization can take advantage of the technology
there are people who are, right now, 30 years ahead of their software goals compared to where they otherwise would be
there will be some set of people at the top of that Pareto distribution who start outperforming proportionately huge swaths of the bottom

>>110326430
good engineers will simply go find work elsewhere. managers will arise who do know better. the Jeff Deans of the world will continue to pop up. the laws of money aren't magically going away

>>110325372
Because Jeets stagnated software by 30 years.

johnathan blow had the sense and fortune to ride the "indie" video game wave while it was still rising. his programming acumen and language are the middest shit ever and i wouldnt trust him to see progress if he were being physically buried by an LLM-driven backhoe

>>110326456
>there are people who are, right now, 30 years ahead of their software goals compared to where they otherwise would be
Yeah that's me on my personal projects. Thing is I see less than zero reasons to publish any of them. First up they're niche tools adapted to my specific needs adding to already niche programs or standalone tools no one would use anyways and I frankly don't feel like dealing with retards too stupid to read documentation spamming me asking how to compile or use any of it. I suspect a LOOOOOT of other people using LLMs are in the same boat. There really aren't any big niches no one has filled yet in terms of software so I don't know why some people expect a massive explosion of indie projects. We'll probably see an acceleration to indie game dev but programming? I don't see it honestly.

>>110326482
>managers will arise who do know better.
No, not at scale. Their ignorance is critical to the bureaucracy.
Maybe you should read the Managerial Revolution.
We neither live in a libertarian paradise nor can human organizations expand much without quickly losing efficiency so your laws of money don't mean much in the real world.

Anyway once we have to start arguing politics to convince each other I can pretty safely say your arguments are now based on hope and dreams.

>>110326509
the logical result of the price of software approaching zero is that everyone will just write custom whatever for anything they can't pull off the shelf for free or for less than the token cost of creating it

File: David_Hume_2-2484355614.jpg (297 KB, 1024x1200)

297 KB JPG

>>110325547
>>BASICALLY ALREADY AGI
This is 100% true, and not "basically" either. We have AGI. These clankers are it.
>>WORKERS ARE OBSELETE
>>EVERYTHING 10X NOW
Both of these are false, and you can observe that. It's cringeworthy delusional fantasy and marketing. You should revile that shit.
It's just that AGI was overhyped. Not because of some god of the gaps argument that these things lack some vague, undefined semantic notion, and they're not "true" AGI. Don't be fucking retarded. AGI is here and it's kind of shit. Get over it. Your fantasy was never real. If you weren't a half-wit or at least spent time around high percentile intelligent people, or even just stopped to think about the existence of idiot savants for 5 seconds, you'd have already been comfortable with the fact that "intelligence" is not the end-all-be-all.

Maybe with that realization, you won't have to sit there like an asshole and blithely lie that something reliably constructing novel, high level mathematical proof is not AGI. It sucks at creative writing. It sucks at design. It has no taste. It gets things very, very wrong. It is a machine that artificially constructs a general degree of intelligence. Nothing more. Nothing less. There will be no machine god. The apocalypse is not coming.

>>110326565
It can't even learn, LLMs are fundamentally unable to learn anything new without retraining.

>>110326558
I think people are actually overestimating that, historically companies were always hesitant on foundational software without SLAs or support contracts, maybe that will change now. But we 100% will see much more tailored software, inhouse modifications as well as from the supplier directly and you're probably right that it will drive down prices overall.
In my opinion massive SaaS still has a moat for now, smaller tools are definitely in danger.

The new wave of software is software meant to used by agents, not humans. That’s why your software is still the same.

>>110325372
Solid interview.
https://www.youtube.com/watch?v=IRK8h67ol7Q

It is not solved
https://status.cloud.microsoft/

>>110326550
i don't need any hope. i will continue to work on personal projects where i use ai to save myself a lot of time to get the results that i want. and no i'm not producing software. i am only interested in its result. it is fun however to iterate and add features or improvements as i think of them, experimenting, etc.

>>110326596
What use would support in that world?
No one one from the company that produced the software would understand the programs they sold so support wouldntwbe much more than reading a list of features.
If developed in house you'd have far more control over what you have and you could even let AI agents look through the code if things turn to shit.

>>110325384
Okay. So the main thing promised with AI is that things will advance beyond prior comprehension in terms of scope. Instead, everything is just shittier and everyone is just lazier. There has been no transformation of society or zeitgeist or methodology save for pushing laziness because the CEO is invested in the technology that enables laziness. Because it isn't true AI. It's a mechanical Turk.

>>110326644
That's great and I'm happy for you but we were talking about software development in a professional setting.

>>110326588
that's not entirely true
they can self improve a harness which becomes effective learning if the context is organized and called appropriately
it may not be learning per say, but its precomputing, storing and retrieving the solution at the right time
close enough
>>110326596
you are correct. I should clarify that the price of software will only approach zero for people capable of fully taking advantage of the new tooling.
I think, for everyone else, it will become something like 'custom software as a service' where there is still someone you can talk to when something goes wrong.

>>110326613
The world must truly be magical for those with 95 IQ, like you.

>>110325414
Can it ever do anything beyond this? It was trained on humans so how would it ever do better than humans? It just scales existing things up or mashes them together. But maybe this is all it takes.

>>110326662
>it may not be learning per say
No, it's not learning and compaction just makes it all the more obvious.

>>110326657
>It's a mechanical Turk.
Or maybe a mechanical Indian.

>>110326645
>What use would support in that world?
It's less about actual support and more about the legal security and security theatre checklist bullshit that comes with it.
If you develop inhouse and it fucks up it's your fault. If you outsource you can always shift blame to the supplier.
>Why wasn't this feature implemented a year ago? Well the supplier...
>>110326662
It will definitely increase price competition but it will be incredibly unevenly distributed. I think even big SaaS isn't necessarily safe long term but the institutional inertia and asscovering is so loadbearing even if we continue to see these levels of model improvements for another few years instead of slowly tapering off middle management will still want their assess covered as much as possible so yeah bespoke software will become huge instead of the classical one tool SaaS with maybe a few optional modules. The competition between these suppliers will be brutal though.

>>110325384
>What the fuck is he talking about?
if it's truly the advancement, software should have already jumped ahead, but it absolutely hasn't.
All it can do is generate trash code and automate tiny demo porn examples.
Where's the Von Neumann Architecture replacement?
Why do we still need compilers?
Why is software performance not largely solved with automatically fast software now?
Sloppers promise the future, and then keep pushing it back another year, another model

>>110325384
If AI is so great at making software better why hasn't my OS had an update that makes it noticeably faster? Why haven't my games had updates that make them faster?

>>110326659
i was done arguing. you are right, we were talking about that

>>110326712
>why hasn't my OS had an update that makes it noticeably faster?
Your OS maker is retarded.
>Why haven't my games had updates that make them faster?
Rimworld got an update that just tripled my framerate by moving the engine and my 500+ mods load order worked without issue. Maybe buy games from developers that give a shit?

>>110326703
I think the two majors reasons we are where we are now is that LLMs are very good at stroking the egos of executives and managers and that they are also good as generating proof of concept demos, which said executives think is 90% of the way to creating the real thing

>entire field of math has been solved

it will stop there guys though, it's not good

>>110326588
It does not matter. Not only are you confusing machine learning jargon with a generalized notion of "learning", not only are finite metalanguages capable of describing infinite object languages, it still fundamentally does not matter. Why? Because learning and intelligence are orthogonal concepts. If you had even the smallest amount of exposure to real world intelligence, you wouldn't even think of the conflation. It's something you literally just invented, nobody else gave to you certainly not any respectable literature, to prop up your delusional fantasy of what intelligence is. In this manner, you are no different from the morons driving the hype-train boldly off a cliff.

>>110326727
Was that update made with only AI?
Devlog source?

>>110326762
>only AI
Shifting goalposts so soon?

>>110326758
You need to be able to learn to be able to apply your knowledge in new ways.
A worker that starts at a company is often expected to be worthless and a net drain when they start ni matter their raw intelligence.
You are very disengeneous.

>>110326712
The advancement, clearly stated and free of marketing lies, is the ability to automate cranking out shitty web applications, and do a whole host of other things in a mediocre way. It is not to make better code. Welcome to the future.

>>110326778
I don't see why you would even bring it up otherwise.

>>110326789
We will look back to the time when Indians started to dominate software development as the good old days when programs mainly dis what they were expected to do.

>>110326778
>AI hasn't made my software faster
>Uhhh but Rimworld recently got an update that made it faster
>Was that update made with AI?
>WHO GIVES A FUCK???

Holy shit lmao.

>>110326565
i think you underestimate it. we don't know what kind of behaviors might emerge. something like layering. these things might only just be building blocks for other things we can't imagine yet

>>110326657
>>110325372
If you think about it, the ability for the people involved to actually understand the technology is actually a huge bottleneck. Both the developers maintaing the software as well as the normalfag users. Because how do you make extremely sophiticated technology that the average fucking moron can use effectively? You cant, not really. Even most consumer grade technology has to be severely hampered and knecapped to be even remotely usable by the general public.

>>110325372
I think the wave of AI-assisted recompilations is quite impressive.

>>110326793
>>110326828
What is it then? It only counts if it's fully AI written instead of AI assisted? Can you pick a lane so I know what cope you'll pull next running down your usual scripts?

>>110326969
he is suggesting the possibility that the improvement may have been found by a human and implemented by a human by hand

>>110326981
But surely other software besides
>*checks notes*
Rimworld has received performance boosts right? AI should be able to take any piece of software and find improvements right?

>>110326781
>You need to be able to learn to be able to apply your knowledge in new ways.
"Learning" in the sense as it is used with these models and what they cannot do, does not mean taking in new information. It's a very specific notion of metaprogramming, in which a program (in this case, the weights) is automatically derived from a glut of data. AGI is a program, already derived. Learning, as you mean it and define it here, is more broadly the intake and retaining of information, which yes the models are actually capable of. They can construct libraries from external phenomenon, recording them to disk, and react to them according to a coherent notion of interpretation.

The ML notion of learning doesn't define the boundary of information. Obviously not, otherwise there would be no place to put the text in which they predict the next token of. But there's a hell of a lot more flies in the ointment than that. To say they can't "learn", only means that contexts form their own universes, and even that isn't true as contexts can modify weights on disk or inference environments, or what have you. What matters is do they have the first principles and the structure to use them to produce a generalized reconstruction of intelligence. And the answer to that anybody with eyes is a yes. Their shortcomings and mediocrity was also to be fully expected, and it's not a bug.

File: 1787463369576317.jpg (81 KB, 1080x1124)

81 KB JPG

>>110326680
>>110325372
AI should, in theory, be able to replicate and implement pretty much anything.

The issue here is that theory doesn't take into account that companies and people in general are constrained by time, not competency. Given infinite time, a single developer should be able to replicate and implement pretty much anything. AI allows developers to write much more code in the same amount of time, but that doesn't necessarily mean better quality code, just that now said dev has more time to implement more features. Projects become encumbered by feature creep and complexity as a result.

The tricky part is that the more complex a program grows, the worse AI fares for two reasons: 1- AI by its very nature is unable to understand what it's writing, so it will always try to implement the "correct" way of code, even if it's too complex or unnecessarily verbose for the task at hand. 2- AI has a limited context window and increasing the context window is exponentially more complex than increasing computing and while there's tricks to ameliorate this, it means that given a complex enough program LLMs will be unable to write code correctly without breaking things.

Another issue with AI-assisted coding is that programming may be tedious and time-consuming, but it is the easy part. The hard part is the architecture and the logic behind a given program. AI automates decisions of architecture and logic. So programmers become increasingly incapable of understanding what the code does.

This also happens with education and is the reason why AI-assisted education is a terrible idea and why students that use AI perform so poorly: when learning something, there is a friction and a struggle. That friction is the learning. AI removes the friction and merely provides answer, so the user doesn't actually learn anything, just memorizes statistically-generated text.

This is why AI-assisted programming has not resulted in better programs, just more complex ones.

>>110325372
>I am a real genius listen to me and my takes!
>I make games!
Only one of these can be true at a time brother.

>>110327288
Video games are among the most complex software.

>>110325372
If there's a basketweaving app that works, what do you expect, remake it in AI? An entire new basketweaving app? Thats stupid.

>>110327288
But can you actually address his point?

>>110326565
>It sucks at creative writing. It sucks at design. It has no taste. It gets things very, very wrong.
Skill issue.

>>110327571
Have these "fixes" actually been tested? Or is this just Ramaman Sukdik padding his resume?