Comments by "Gilad Barlev" (@GSBarlev) on "Fireship" channel.

  1. You've just made Nix users extremely happy. That's right, all eight of them.
    6600
  2. I'm legitimately impressed by 3:10. Either the model is actually that good or Jeff put a ton of effort into that scripted response. Either way, very impressive.
    2300
  3. To clarify, this attack didn't actually affect any production systems—every stable distro was at least two minor versions old, and no rolling release was built in a way such that to be vulnerable to the attack vector. The one system where this was out in the wild was... macOS systems with Homebrew. But don't let Tim Apple find out, or he'll try to block macs from "sideloading" software.
    1100
  4. I've actually taken professional docker courses. This is better than 90% of them. I will absolutely add "Docker certified by Jeff Fireship" to my resumé the next time I'm on the hunt.
    953
  5. Yeah, some state agency is extremely pissed right now that their op was busted after two years of work, and before their backdoor could actually make it into the wild.
    819
  6. So apparent Crowdstrike Falcon broke a Debian image about three months ago, but because Linux doesn't actually force software updates, it fucked the VMs of a few dozen nerds who reported the issue and rolled back to the previous image before the entire global ecosystem went down. Seems like there's a few lessons to be learned here.
    812
  7. Database programmers are a different breed, my dude.
    809
  8. My favorite algorithm is probably Perlin Noise. You touched on procedural generation in your wavefunction collapse portion, but the fact that the result is deterministic (based on the initial seed) and independent of the order of the observations is just mind-blowning.
    776
  9. It's important that a nonzero percentage of the TOR network be used for innocuous purposes (such as leaving 💩post replies to YT comments) or else legal defenses against evidence of TOR usage boil down to, "But you can't prove which crime I committed."
    402
  10.  @LosFarmosCTL  Pretty sure you're right about Homebrew from what I've found—the general consensus appears to be that the FOSS world dodged a bullet by Freund discovering this in March and not May, by which point this version would have actually been deployed in the intended target—Ubuntu 24.04 LTS.
    183
  11. I suspect two things are going on here: 1. He has such absurd amounts of wealth that he can afford all the image consultants, fitness coaches, plastic surgery and makeup / prosthetics needed to make him look however he wants. 2. SBF completely ruined the whole game of "you can always trust the pasty nerd with no rizz"
    178
  12. That's something I would really hate to get called on. "Oh, you're fluent in erlang? Great! You're now responsible for keeping this extremely important but ancient codebase from exploding! Here's your company pager—hope you don't believe in nights or weekends!"
    177
  13. 💯 Back when I was a physics undergrad (15ish years ago) our closest equivalent to an Arduino was a multi-hundred-dollar GPIO EVM hooked up to a computer running LabVIEW. (We also had to walk one mile to the physics building, uphill, in the snow)
    96
  14. It's not even the first time. Or was that "Searches cures for depression > Gets directions to the Golden Gate Bridge" screenshot debunked?
    92
  15. ​ @rohithreddy75 Install an OS that you actually control? Like Windows 10 (EOL be damned) or that thing with the 🐧
    84
  16. same. Fixed in 131.0.2, my system was already on 131.0.3
    83
  17. All the Python devs I've talked with have been extremely skeptical of Mojo (especially when we've had pypy and numba for years). Now that it's general availability, I'm looking forward to some unbiased cross-comparisons (and, hey, may do some myself).
    79
  18. If you do a deeper dive into open microcontroller computing, I hope you cover how the original Arduino developers got completely shafted. I also would love to see a "100 seconds" on microPython (/ CircuitPython) and cover how these days you can don't even have to write static programs--you can just start up a REPL (or, better, WebREPL) and just run python commands directly on the device. The level to which this has cut down on iteration time and increased prototyping speed has made it really hard to justify going back to Arduino.
    77
  19. Just to clarify, the top spot still goes to Phippy & Friends and their Illustrated Children's Guides to Kubernetes (which, yes, do cover Docker as well).
    76
  20. Arch Nix 🤝 SteamOS since 3.5
    76
  21. Gotta love how, now that the Supreme Court has ended Chevron Deference, these complicated issues will be decided exclusively by political appointees with no technical experience who think a RESTful API is something that happens in the bathroom after you've had a good night's sleep.
    76
  22. This is why microSD cards are OP. So much more flushable.
    75
  23.  @DJ-dh3oe  You're missing the point, though: he's been rocking the robot look for decades. Why suddenly go surfer boy chic?
    73
  24. This is a huge leap forward in Sam Altman's ability to separate AI bros from their trust funds and crypto hodlings.
    70
  25. Guess what? Everything he discussed and showcased in this video still applies to FOSS frameworks like pytorch-rocm. Get that bag, Jeff.
    67
  26. This 🐐 develops postgres for a living. With all the db exploits he's probably seen over the years, this was probably a giant snooze-fest for him.
    63
  27. The thing that bothered me about the sponsor bit was the "1 month free trial" part—like, 75% of the JetBrains suite has "Community Editions" that are perpetually free.
    60
  28. Learning to code is now and has always been a horrible idea. Learning to program is useful no matter if you never touch another computer in your life. It's all about thinking logically, which, as Jeff says, will always be needed, and a computer science degree (as opposed to a four-week for-profit boot camp) is a great way to build those skills.
    59
  29. There's a few performers who I'd hope would get into politics—then again, they might not be willing to degrade themselves to that level.
    47
  30. Mark Twain: "I apologize for the long pull request—I didn't have time to write a short one."
    44
  31. "AI" is a meaningless buzzword. Pre-trained transformers and foundation models have legit useful profitable applications, but it's more because it's allowed capabilities that used to be reserved for supercomputing clusters to be commoditized to running in your garden-variety cloud, not because the models of today are "smarter" than they were five years ago.
    34
  32. People around me have been pushing "natural language code gen" for a while now in the data analysis space, to which I say—anyone who can execute a clear and unambiguous data ask using natural language more efficiently than they can construct the ideal SQL query or DataFrame op is a savant, of one form or another.
    33
  33. I know enough C to know that I should leave C to the people who know how how to write C. For example, a large part of my PhD work regarded matrix multiplication. I wrote a multicore version in C, an extremely multicore version in CUDA... and then ended up mostly using numpy on a single thread to take advantage of the hundreds of collective lifetimes the numpy authors had spent on optimization tricks and algorithmic shortcuts.
    30
  34. ​ @Ztaticify The Gentoo people were supposed to be delivering the roasts, but they got stuck compiling the latest version of Ladybird. They'll be here eventually.
    29
  35. QBASIC was my second programming language. 😎
    24
  36. Static HTML builders like mkdocs and Hugo FTW.
    23
  37. With how sea levels are rising, that's 💯 the more relevant and marketable skill.
    21
  38. Linux has been my primary operating system since 2006. I still end up Googling the syntax and command-line flags for the tar command once a week. But in the last year I did finally learn how to exit vim!
    18
  39. Remember Popehat's Rule of Goats: If you [kiss] a goat, but tell everyone "it was satire!," you're still a goat[kiss]er.
    18
  40. ​ @yourlinuxguy Yep. To really turn the screw, in their exit interview they'll tell you they're letting you go because you're not "delivering your own results" and point to your mentees as paragons of productivity.
    17
  41. It's a thing with these billionaires. Bezos famously loved The Expanse; every week we hear about a new start-up inspired by a Black Mirror episode; heck, Zuck named it the flipping Metaverse after Snow Crash.
    17
  42. Given how I get most of my breaking news from Jeff, it's pretty awesome, not only that Alberta scooped him, but that she's got enough of a following that a huge portion of the top comments are talking about how she already covered this!
    17
  43. Eh. Jensen wants you to believe that you need to pay him a fortune for a 4090 and sell your kidney to afford the power bill, but the truth is, not really. I've run Stable Diffusion XL and Mistral 7B on my local computer, which, sure, has 128GB of RAM, but only a 12GB last-gen Radeon GPU. Hooked up a kill-a-watt, crunched some numbers, I can answer maybe a dozen proompts for about $0.14 in electricity. Sure, GPT-4 Turbo, Sora and Gemini are an order or magnitude or two more expensive, but they're also running on dedicated (and thus, higher efficiency) server hardware. There are plenty of reasons to be anti-AI, but environmental impact isn't one of them, especially since, once people realize that these capabilities are massively over-hyped, the demand is going to be nowhere near the numbers that Altman and Huang are forecasting.
    16
  44. Counterpoint: I'm surprised the NSA didn't notice this themselves: "Hey, there's this weird 500ms slowdown in our botnet playbooks. Someone needs to dig into that."
    16
  45. Not only would a geothermal plant at Yellowstone produce enough power to run the entire country indefinitely, it would also prevent another supervolcano mega-eruption.
    15
  46. Is it not common knowledge that Guido is on Microsol's payroll?
    15
  47. Arch users to each other: "Is that how annoying we've sounded all this time?"
    14
  48. Seeing all this drama just made me so dang grateful that I learned Hugo from you in 100 seconds.
    14
  49. "Why is a video from 10 years ago showing up at the top of my subscription feed?" "Wait, was Fireship around 10 years ago?"
    14
  50. Props for explaining the visual encoder. It seemed like such a waste to think they were just diffusing every image of every frame, and this approach makes a lot more sense. If I were designing a transformer model to generate videos, I'd take a full step back and basically train it to interface with Blender. That is, have the model outputs decode to models with articulation; lighting rigs; textures; paths and physics, then use conventional rendering to turn that into video. Way better use of computing than requiring these models to synethesize unconstrained.
    13