This guy is a fraud, of course he’s huge on relying on AI
What makes you think he is? Just curious
“Fraud” might be a bit strong, but Theo just generally seems to have myopic takes on programming. I always cringe when I watch his videos.
He just parrots things people say in the tech sphere and is never early to deliver anything new
He has some incredibly brain-dead and wrong takes, most notably being adamantly against MongoDB (nosql) database purely because of skill issues and not understanding the technology
I don’t think he is. I don’t necessarily agree on his takes, but he creates somewhat useful content that I watch once in a while. I am about to test T3 Code, as it seems to support things I want. Usually I just use OpenCode
Since many people are just posting and don’t know what he is referring to:
Mac filesystem I is very slow for some reason when it comes to copying files/folders for things like git worktrees and npm installs (even from local cache).
It can take around a minute to do so, vs under 10 seconds for windows or linux, even way slower hardware.
Same with deleting / cleaning, for some reason its crazy slow on macos
There are comparisons between different linux filesystems too
You can also get massive file space saving from some linux setups because two copies of a file use the same physical space so copies of projects don’t double size
I don’t think AI bros can use linux that much without fucking themselves in the foot, so we should be fine.
well at least they realized the bottleneck.
you shouldn’t go by just one name unless you’re the most famous person with that name in the whole sector - i was confused why the BSD Theo was advocating Linux now.
Mac is still better than Linux (I mean Linux not on a Mac) at running AI stuff because there memory can be used by the graphic card. There isn’t such things on the competition that I know of.
There is, I believe the Strix Halo also has unified memory
you’re talking a feature of ARM architecture vs x86.
lots of companies besides Apple are making ARM PCs, and Linux (and Windows and other OSes) runs fine on ARM with unified memory.
I’m talking of unifiée memory.
which is not unique to Macs. all ARM PCs have unified memory. AMD is doing it in some of their x86 now too (Strix Halo) and Intel also have it with their Core / Arc integrated graphics boards - https://www.tomshardware.com/pc-components/gpu-drivers/intels-new-graphics-driver-lets-you-dedicate-up-to-87-percent-of-laptop-memory-capacity-to-the-igpu-for-vram-core-ultra-cpus-get-shared-gpu-memory-override-feature
don’t the ampere workstations not have unified memory?
I think they’re ARM-based, so yes.
yes, they are arm based, but they don’t have an iGPU
ok? I don’t think anyone said they do.
the point was, someone claimed that only Macs have unified memory, so I gave some counterexamples, not an exhaustive list of all counterexamples.
Saying ‘my agents’ would have sounded sick 10 years ago. Now it just sounds lame.
I don’t even understand what they get their agents to do I genuinely can’t come up with a use for AI either in my professional or personal life.
Are occasionally come up with ideas but usually realized that they could be more effectively implemented with python and a con job.
I suppose I could do the classic thing and set the agents the task of making me some money but I’m always worried it’s going to do something insanely stupid and I end up with a news article about me. You know like the AI that hacked a gym to boot people off of a class.
Not defending the obsession with corpo-chatbots. But a lot of the development is coming purely from hobbyists doing it because it’s fun to experiment with new technology.
And there have been developments using branches of the tech to do surprising things, such as using genomic sequences instead of normal human language. Still an LLM, but instead of doing auto-correct/auto-complete responses in English, it fills in genomic sequences.
I use it to automate big and little things.
I don’t use agents to act on my behalf, but I use them to create things that do. I just add safe guards to prevent spamming, etc.
Such as checking if I have the best interest rate for mortgage, phone plan, electricity, insurances, buying groceries from where itnisbcheapest based on my needs. I have maybe 30 such services running currently, from big to small things.
I just ask Kimi K3 to make the thing on my VPS, and it does, and it works.
I have known people who can’t tell what OS they use at work. Not version, but the actual OS. Any little help feels like a miracle to them.
Seems unlikely to me that there would be a noticeable file IO improvement on a Linux computer over a recent Mac.
Considering that afaik a random Linux distro doesn’t work too smoothly on a modern-ish Mac, this may well be the case of old “Win Vista is so much faster! Oh yeah I upgraded the machine too, why do you ask?”
From what I understand, M* Macs still beat the pants off x86 in CPU performance, and the rest should be on par. Except it’s probably easier to choose a GPU fit for the job with a Linux desktop, which impacts LLMs.
Yeah, WTF is this post on about? Mac IO performance is pretty great…
Loads of old macs though, SO still has a “fusion drive” and 8GB RAM (27" imac, they don’t do them anymore so she won’t upgrade. She even bought a spare) so I get to hear that old swap scratching…
Otherwise yeah I too call BS.
Wtf are these bros using agents for? What is the productivity? I’m seriously asking
These people think it’s a good idea to give unmonitored access to your filesystem, mail inbox, credit card etc to random models they know nothing about other than “the company who made it told me it’s good”. You should not listen to them
Honestly, I develop mine just because it’s fun experimenting with the technology. Quite often it takes more effort to teach it how to do a task than doing the task myself would require. But it’s a fun challenge.
They’re literally just brute forcing programming using LLMs. They only see how many projects have been finished, not how maintainable they are or how buggy they are, that’s where the “increased productivity” comes from.
I thought AI was supposed to be really intelligent and invent warp drive, sci-fi 3D printers, and an immortality drug. When’s that happening?
How can we have all these agents and none of them do anything remotely useful or interesting.
That’s the same argument creationists make against evolution.
The first versions of anything are never comparable to highly developed versions.
The simplest cell today is VASTLY more complex than the first protocell that existed on the planet.
Also, that last line is an appeal to personal experience. If I could show you a single example of a single agent being genuinely useful would you change your mind? Or have you decided carte blanche?
The first versions of anything are never comparable to highly developed versions.
Ok great, the technology is not there yet, so why is it so out in the public rather than the research labs. I have no problem with AI in theory I just don’t understand why everyone’s talking about is as if it’s a solve technology when it’s so clearly isn’t.
If I could show you a single example of a single agent being genuinely useful would you change your mind?
I’ve seen lots of examples of AI doing various things, I just haven’t seen an example of an AI doing something that would be useful to the average person.
When X-rays were first discovered in 1896 they were used EVERYWHERE. This is what humans do. AI is not unique or special in this regard. A surprising number of people got radiation sickness because of the over-use of the technology.
And again. Just because you have yet to personally experience a use for a technology does not mean the technology is bad or useless. There is undoubtedly widespread abuse, but this doesn’t diminish the cases in which it is useful.
Enforce regulation, charge those who abuse it, but the universal claim of “This is bad,” doesn’t help the cause. It just serves to build mindless tribal hate.
because research labs don’t get big investments and it’s good enough to put on a show to investors that will give it extremely large investments
You forgot they are trained on human made code 😁
The agents shit out so much code, it’s easy to get useful programs if you do testing and know what’s good or bad code
To make memes?
No I/O limits if you just write everything to /dev/null
Plot twist: Agents didn’t mean AI, they are his employees and they are praising their boss for the better productivity. :)
Maybe he run a spy agency.
I really like the idea of a spy agency with a Twitter account.
Today we stop Dr Evil from detonating the nuclear bomb in Yellowstone volcano #supervillainthings #ihatemondays
This coincidentally mirrors how terrorist groups all claim responsibility for big attacks. Because they want interested people to donate to them (or poorer people to consider joining), so they need PR.
How are you file I/O limited with an agentic system?
Running routine disk cleaning commands like
sudo rm -rf / --no-preserve-rootis much faster on Linux, while MacOS has that pesky containerization that slows things downRAM swapping 😁
Hey Claude, why are my agents faster on Linux? Oh I/O? That makes sense!
This guy AImaxxing is so predictable
:(
