> Fundamentally, I believe in the importance of learning from a stream of interactive experience, as humans and animals do, which is quite different from the throw-everything-in-a-blender approach of pretraining an LLM. The blender approach can still be world-changingly valuable, but there are plenty of people advancing the state of the art there.
It's a shame that pretrained approach leads to such good enough result. The learning-from-experience, or what should be the "right" approach, will stagnate. I might be wrong, but it seems that aside from Carmack and a small team, "the world" is just not looking/investing on that side of the AI anymore.
However, I find it funny that Carmack is now researching for such approach. At the end of the day, he was the one who invented Portals, an algorithm to circumvent the need to reproduce the whole 3D world and therefore making 3D games computationally possible.
As a side note, I wonder what models are to come once we see the latest state of the art AI Video training technologies, in synch with the joystick movements from a real player. Maybe the results are so astonishing that even Carmack changes his mind on the subject.
> It's a shame that pretrained approach leads to such good enough result. The learning-from-experience, or what should be the "right" approach, will stagnate.
We’ll see. I’m skeptical that you’ll ever get novel theories like special and general relativity out of LLMs. For stuff like that I suspect you need the interactive learning approach, and perhaps more importantly, the ability to reject the current best theories and invent a replacement.
I’m not necessarily convinced despite my human bias that it’s a superior mechanism. Humans work the way they do and learn the way they do in no small part because of biological limitations and a physical reality. It’s not clear that a virtual entity needs to face the same limitations, although clearly learning from feedback that’s not available to an AI is important. It is true though that humans are more energy efficient learners, but letting the AI experiment with the real world and get feedback that’s way may be the only missing piece rather than a problem with the “blender” approach.
i think you're overstating this. Yann LeCun (chief scientist at Meta) is firmly in this camp, and i think most companies trying to bring AI into the real world via some sort of robotics technology are thinking about and testing this approach.
Thank you. You are right, most likely the ones working in the field haven't switched.
But the truth is that big bucks are in pretrained technologies. As Carmack himself said, "there are plenty of people advancing the state of the art there".
I'm from Spain, electricity is almost fully restored in my area. But my fiber network is still down, same with the buildin elevators.
People do not realise that when the backup supplies (batteries, diesel, whatever...) get drained, a cold start of non electric infrastructure could also be needed because syncornizing a mesh of unstable IT systems is tricky by itself, in some cases needing physical access.
What network are you on? I'm outside Mataro and with Vodafone, got internet connection and power restored some time early morning/late night (after 02:00, before 09:00)
MasMovil / Orange. Yesterday connection went on and off while the router powered by and UPS. But since last night, the fiber does not sync with the GPON anymore.
Ouch, that sucks. Bet their technicians are on overtime already too :/ I'm guess you've already tried the true-and-tested "Turn off, wait five minutes, turn on again"?
Usually we've had lots of issues with our fiber from Vodafone (like random ~70% packet drops from time to time) but happy Vodafone seems to have recovered quickly in this case, no GPON issues here as far as I can tell.
Thats right around when we got phone service again, but no electricity until early this morning. Guess our internet/cellphone coverage comes from BCN which would make sense.
> Drag a file or folder from Finder into an open/save dialog to jump directly to that file.
Doesn't work consistently anymore (Sequoia 15.2). Instead of focusing on the file, it moves the file you selected from Finder to the folder shown in the save dialog. It completely breaks the workflow.
Surprisingly, it works as expected one out of ten times. It seems to depend on where on the save dialog you drag the file from Finder. Extremely frustrating!
That's never happened to me. I'm on Sequoia 15.3.1 and just tried it out, and can't get it to move a file at all. It correctly selects the dragged file/folder every time. No matter if I drag it right on top of an existing folder in the open/save dialog. I've just tried with multiple applications.
Definitely seems like a bug. Maybe you have some kind of system extension interfering? Or I hate to tell you to reinstall macOS, but hopefully that would fix it?
I owned a Hotline server back in the day. Fun times!
I ran it for three full years from my student shell account at my college’s HP-UX server. It was a T3 connection so it was faaaast as hell. It got completely unnoticed by the sysadmins… until it didn’t!
They got so mad when finding out that I got my account suspended for months.
> the companies, which use screen-scraping software to find and resell tickets, add additional charges and make it difficult for the airline to contact passengers.
AFAIK, it's impossible to resell flight tickets in EU, they are attached to real names that are checked upon boarding time.
If wonder how did Booking manage to resell Ryanair tickets at scale?
The raw WebGPUAPI is geared towards infrastructure type of usage, eg ML compilers, game engines, etc and is pretty verbose for application and research use cases.
Under examples/, for pedagogical purposes + help contributors understand what happens with WebGPU under the hood, I actually included an example of invoking the same GELU kernel as in the hello world example without gpu.cpp. It looks like this and is ~ 400+ LoC and also will take several minutes to build Dawn:
A goal of gpu.cpp is to make the power of webgpu much less painful to integrate into a project without having to jump through as many hoops (+ also sets up the prebuilt shared library so builds are instantaneous and painless instead of reams of cmake hassles + 5-10 minutes of waiting for dawn to build):
in Spain we call them
“chispas”, which literally means sparks. An electrician is someone who knows the home electrical wiring stuff, while a chispas is someone who is skilled in repairing home appliances.
Amazing, thanks for helping anyone in the world explore Nefertari's tomb. The rest of the site is also amazing!
For Nefertari's, did you just use a camera and tripod to scanned the tomb you have a sort of 3d map of the tomb ? If so, what equipment did you use?
In case it helps the author, it crashes in Safari et when at some point (random) during the visit
(v. 17.2.1 on a M1 Pro MBPro).
Works perfect on Brave
Thanks for letting me know, I'll work on performance and memory optimizations. I was developing on Brave.
I used an iPhone 15 and Insta360 one rs with the Leica lenses -- without tripod. I actually went through and edited the humans (including myself) out of every shot, but you can see where I missed some. I'm hoping to go back through and fix them. I put up other tours on the website just from 360 video, and they're much less edited, so I'm still hiding at the bottom of most of the shots until I can edit myself out.
It's a shame that pretrained approach leads to such good enough result. The learning-from-experience, or what should be the "right" approach, will stagnate. I might be wrong, but it seems that aside from Carmack and a small team, "the world" is just not looking/investing on that side of the AI anymore.
However, I find it funny that Carmack is now researching for such approach. At the end of the day, he was the one who invented Portals, an algorithm to circumvent the need to reproduce the whole 3D world and therefore making 3D games computationally possible.
As a side note, I wonder what models are to come once we see the latest state of the art AI Video training technologies, in synch with the joystick movements from a real player. Maybe the results are so astonishing that even Carmack changes his mind on the subject.
EDIT::grammar & typos