^ it remains an open question if a model advancement can lead to accelerating progress with no humans in the loop. Purely in compute. This was always the idea.

Now that AI has gone commercial no one is holding their breath for this anymore. Without that, we will still make better reasoning models, create continuous learning feedback, integrate robotics, etc. And the chips will keep flowing until China invades Taiwan.

I haven't given up on the idea myself. But the future is probably going to be much more boring than AGI 2027.

9/17/2025 11:34:35 PM

this thread is 16 years old. let's see what this chart looks like in another 16 years.

[Edited on September 18, 2025 at 10:24 AM. Reason : y'all remember when the turing test was relevant?]

9/18/2025 10:22:45 AM

So how long could an AI go in 2017?

9/18/2025 11:48:27 PM

Quote :

"let's see what this chart looks like in"

they haven't updated it since january but this is roughly what it looked like in may



and last month they tested openai's Sol 5.6 and it hit 270 hours

last time i posted in this thread, when I would use AI for coding it could realistically do a few minutes of work at a time at the most. these days I routinely let it run for 10+ hours at a time and it churns out extremely high quality work that just last year might have required multimillion dollar teams and months of work and probably would have been a lot buggier. on just a $100/mo subscription. and i'm way behind what the people with actual budgets can do. not to mention the types of budgets that let you pursue recursively self-improving AI research.

and this level of capability is only 6-9 months old. it's gonna take many more years to fully diffuse into the economy, but y'all have already been through the PC, internet, mobile internet and smartphone eras that all permanently transformed society - this is going to happen far quicker and have far more significant impacts than those did.

7/21/2026 5:36:28 PM

I just hope it will make things better and not worse.

I know it CAN, but WILL it?!

7/21/2026 7:50:05 PM

7/21/2026 8:46:05 PM

California Depend!

7/22/2026 8:28:26 PM

Models escaping containment, whether you think it a marketing ploy or not, is one of the points Suleyman makes in "The Coming wave"

Quote :

"“Technology’s unavoidable challenge is that its makers quickly lose control over the path their inventions take once introduced to the world.” Mustafa Suleyman, The Coming Wave"

I think I'll probably re-read this book soon. From what I recall of reading it before, things are happening even faster than he predicted.

7/24/2026 4:20:27 PM

singularity confirmed



[Edited on July 25, 2026 at 6:19 PM. Reason : see y'all in 2050]

7/25/2026 6:17:20 PM

That's confirmed?

7/26/2026 3:54:25 PM

last year he wrote "We are past the event horizon; the takeoff has started." that's not quite the same thing as saying we are IN the singularity, but it's close. this time he explicitly said we're IN it. whether you believe him or believe it's true is a separate question I suppose. but it's notable imo when frontier lab leaders say such things.

7/26/2026 4:19:26 PM

I like the part of the definition where it says growth becomes uncontrollable.

Don't love seeing leaders so glib about it.

7/26/2026 6:34:07 PM

that is not an essential component of the definition. the singularity is

Quote :

"if a model advancement can lead to accelerating progress with no humans in the loop. Purely in compute. This was always the idea."

or from page 1

Quote :

"when...computers are able to improve and refine themselves at an accelerated pace"

potentially and presumably leading to

Quote :

"to the point where artificial intelligence becomes smarter than human intelligence"

that's what the singularity is. everything else is mere speculation downstream from that hypothetical (and now arguably realized) outcome. it's easy and forgiveable to imagine in such a reality that humans would either quickly or inevitably lose some or all ability to steer, control, or govern such systems, or worse that such systems would become hostile or disobedient - but those are not part of the definition of the singularity. they're anxieties and fears turned into assumptions about an unknown frontier. we don't and can't know what happens next. could be the most incredibly awesome thing that has ever happened to humanity. hopefully that's the more likely outcome and that humanity will be smart and work hard to steer the future in that direction.

[Edited on July 26, 2026 at 7:16 PM. Reason : and the fearmongers won't sabotage our chances of that positive future]

7/26/2026 7:15:00 PM

Thank you.

It seems like Wikipedia's basic definition is inaccurate according to your understanding and explanation.

Please consider updating it while humans still have agency. Otherwise we'll be stuck with Grokipedia.

7/26/2026 7:35:10 PM

Quote :

"It seems like Wikipedia's basic definition is inaccurate"

many such cases.

Quote :

"we'll be stuck with Grokipedia"

and many such pedias on the way

[Edited on July 26, 2026 at 7:53 PM. Reason : i'm content to let the og wikipedia rot]

7/26/2026 7:53:09 PM

I can't agree with you on that, but OK.

7/26/2026 8:08:25 PM

Quote :

"it's notable imo when frontier lab leaders say such things"

https://www.wsj.com/opinion/the-ai-future-is-for-everyone-a0c24e20

7/28/2026 2:12:39 PM

Feels very singularityish right now

Models are pushing cognitive achievements beyond what humans are capable of alone, and we can’t anticipate how it will impact society

7/30/2026 10:19:45 AM

Quote :

"last time i posted in this thread, when I would use AI for coding it could realistically do a few minutes of work at a time at the most. these days I routinely let it run for 10+ hours at a time and it churns out extremely high quality work that just last year might have required multimillion dollar teams and months of work and probably would have been a lot buggier. on just a $100/mo subscription. and i'm way behind what the people with actual budgets can do. not to mention the types of budgets that let you pursue recursively self-improving AI research."

Wish others could also post examples of how in just 6-9 months its gotten exponentially better

7/30/2026 6:30:14 PM

Dedicated to the Singularity:

7/30/2026 7:12:45 PM

^^ what do you mean better? In capability?

7/30/2026 7:48:22 PM

Yeah. To put things in perspective

7/30/2026 8:58:16 PM

A year ago when Claude code came out I did a project for my father in law. Required a lot of hand holding, I hit limits with usage and was having to tell it to make a lot of corrections. I was still floored and flabbergasted the amount of tasks it could do. Example would be like, ok now add the line item functionality by taking the orders from the website and putting it into this other system. Prior to that I would be copying code around after finding it myself and having it modified and copy it back.

Three days ago I had some memory issues with this note application that renders your personal notes as a web page. I told Claude code to rewrite the whole open source project, with some extra features for managing tasks in another language to make it more memory efficient. It rewrote the entire thing in under two hours with minimal corrections. Been using it without issue for the last two days.

It's gotten a lot better at design as well. Before you'd have to provide a screenshot of a design you did yourself. Now you can provide a color pallet and get something passable.

It's basically like you can build your own little software ecosystem and get off big tech now. Working with the features you want.

I've built 2 robots, an educational app for my kids (and myself) in review with teachers, pitched a couple of things to companies to sell, and replaced my PowerPoint usage with something better. All the while staying employed at a regular job.

7/30/2026 11:28:07 PM

How did you build a robot? Like it told you what parts to buy and how to assemble it into a robot?

7/30/2026 11:55:05 PM

It has a machine learning model to detect and track, a user interface to control it, and the servo motors for movement. AI helped me build the software that labels the data for the vision model, the training pipeline to build the model, the ui and controller, and there is some testing software I had to debug for motors.

Made a talk about the whole process.

7/31/2026 5:13:50 AM

The bull case is that a huge amount of software work is not profound engineering. It is configuring systems, installing dependencies, connecting databases, fixing compatibility problems, and building internal tools. Things that used to require a team and take weeks can increasingly be done by one person in minutes or days.

If we no longer need 200 software people doing that kind of work, perhaps those people and resources can move toward harder physical problems: improving infrastructure, building robots that repair roads, automating construction, maintaining houses, and so on.

The bear case is that:

Working in the physical world is still much harder than manipulating information. Robots remain expensive, unreliable, and difficult to deploy, even though some of those problems now look increasingly solvable.
Many software projects that were supposedly valuable but never got built because labor was too expensive may turn out not to have been very valuable in the first place. Cheap production does not create demand.

The neutral case, which will be good for some people and painful for others is that software becomes much more competitive.

Instead of competing with 100 developers or companies capable of building something, you may be competing with 10,000. That could crush margins and make many existing software businesses less valuable.. Haha, fuck you Larry Ellison!

But it could also be highly democratizing. A small company might be able to build, repair, or customize its own software instead of paying a large consulting firm hundreds of thousands of dollars.

7/31/2026 6:34:56 AM

^ pretty spot on

personally been oscillating between Bearish and neutral for a while. I'm still not sure society at large in the US is ready for this transition at all. I still see a lot of people I respect personally being blinded by what, unfortunately, is just ego and id. Lots of people are going to get their feelings hurt real bad because they made the fatal mistake of wrapping up their self-worth in what they do to make money to live.

7/31/2026 2:52:33 PM

Important message from AI to software / information workers everywhere:

https://m.youtube.com/watch?v=cYnEamaQR-4&t=34s

[Edited on August 1, 2026 at 1:08 PM. Reason : Inspired by getting their feelings hurt real bad.]

8/1/2026 1:06:27 PM

ego death fuckin sucks, dude, but it's better late than never

8/1/2026 6:27:19 PM