thundergolfer

Experiencing life not as living but as an anticipated memory.

I agree with the core premise that the big AI companies are fundamentally driven towards advertising revenue and other antagonistic but profit-generating functionality.

Also agree with paxys that the social implications here are deep and troubling. Having ambient AI in a home, even if it's caged to the home, has tricky privacy problems.

I really like the explorations of this space done in Black Mirror's The Entire History of You[1] and Ted Chiang's The Truth of Fact short story[2].

My bet is that the home and other private spaces almost completely yield to computer surveillance, despite the obvious problems. We've already seen this happen with social media and home surveillance cameras.

Just as in Chiang's story spaces were 'invaded' by writing, AI will fill the world and those opting out will occupy the same marginal positions as those occupied by dumb phone users and people without home cameras or televisions.

Interesting times ahead.

1. https://en.wikipedia.org/wiki/The_Entire_History_of_You 2. https://en.wikipedia.org/wiki/The_Truth_of_Fact,_the_Truth_o...

> There are no great FaaS options for running GPU workloads

modal.com exists now

This is an absolutely beautiful personal website. Go click around it, there's so many 'wow' moments. Go to the homepage (https://maximeheckel.com/) and then click 'about', the transition is :chefskiss:

The blog's design style reminds me a lot of Alex Harri's[1], who also does excellent work. I wonder if one influenced the other.

Edit: Oh the homepage has its inspirations actually listed. Harri isn't there.

1. https://alexharri.com/blog/planes

https://github.com/butter-dot-dev/bvisor is pushing in that direction

yeh of course I worked with him for a few years! Agree, smartest person I've ever worked with, and there's a smart crowd at Modal.

Modal engineer here. This isn’t correct. You can DIY this but certainly not by wrapping EC2 which is using the Nitro hypervisor and is not optimized for startup time.

Nearly all players in this space use Gvisor or Firecracker.

Fair, though my guess is that the growth rate of Waymo's market cap will far exceed Google's as Waymo scales. I wish I could invest in Waymo, so I'll take that 2.7% exposure.

I presume if you invest in Google you are indirectly (but significantly) invested in Waymo, like it is with Anthropic?

Waymo is the best service I've used in many, many years. The jump from Uber->Waymo is similar to the quality jump from Taxi->Uber 12 years ago, but I don't see an obvious way for Waymo to get enshittified.

The G stands for "Google", does it not?

Totally agree, and this is covered in an (identically named?) Google Research blog [1].

Just last week I was writing a demo-focused Python file called `safetykit.py`, which has its first demo as this:

    def praise_dryrun(dryrun: bool = True) -> None:
        ...

The snippet which demonstrates the plan-then-execute pattern I have is this:

    def gather(paths):
        files = []
        for pattern in paths:
            files.extend(glob.glob(pattern))
        return files

    def execute(files):
        for f in files:
            os.remove(f)

    files = gather([os.path.join(tmp_dir, "*.txt")])
    if dryrun:
        print(f"Would remove: {files}")
    else:
        execute(files)

I introduced dry-run at my company and I've been happy to see it spread throughout the codebase, because it's a coding practice that more than pays for itself.

[1] https://www.gresearch.com/news/in-praise-of-dry-run/

Whether Nystrom realizes it or not, and I think he does, this piece is shot through with Marxist thinking on use value, exchange value, socially necessary labour time, and the general structure of capital social relations.

Capitalism has been a fantastically productive system that has also produced a great deal of labour alienation. Nystrom has a deep need to labour for those he cares about, he needs to make the scarf, slowly and badly, for his grandmother.

But the socially necessary amount of labour to make a scarf is now extremely small, and so Nystrom labours in software to earn a higher wage.

The wage doesn't fulfil him so much, because it's for labour power directed for the purpose of value valorization (aka. profit), not to help those he cares about.

He's skilled and lucky, so he has plenty of excess after labouring to poorly make a scarf. But if he does not already have plenty of capital, he has to work, and his capital has to be put to work too, on things other than badly making scarves, lest it too whither away.

It does pretty obviously read as descriptive. I think people just uncharitably read it.

You're not misunderstanding, the PCIe does indeed outperform on the memory bandwidth tests. But it gets dominated on FLOP/s and real-world application bencharks.

Author here. That 1:50-100 ratio looks roughly right based on my research, but my numbers have GPUs faring even worse.

  Component                      Type       MTBF (yrs)  AFR
  ─────────────────────────────────────────────────────────

  SSD                            Hardware   ~100        ~1%
  RAM uncorrectable error        Hardware   ~75         ~1-4%
  NVIDIA A100 critical error†    Hardware   0.18 (65d)  -
  NVIDIA H100 critical error†    Hardware   0.15 (50d)  -

† “Critical error” refers to a NVIDIA Xid or sXid error which is not recoverable, requiring application and GPU reset.

Only a minority of GPU 'failures' appear to be permanent hardware problems, such as row remapping errors. A lot seem to be, like another comment says, a consequence of operating too close to the operational limit, tipping over it, and then requiring a power cycle.

Was the LLM using vision capabilities to verify the correctness of it's work? If so, how was that verification method guided by you?

The first 500+ words of this article are focused almost exclusively on the dynamics of corporate capitalism, not AI.

Read the article.

I definitely see what you're saying, but:

> has come at everyone, equally, all at once

is not true. It's obvious that certain people and certain fields are technological laggards or technological early adopters.

Other computing and IT technologies also provided a good training ground for this stuff. LLMs have really interesting new properties, but all have familiar properties and decade+ old methods of distribution.

This stuff is difficult, sure. But we have long set a low bar for education management and the results—declining literacy and math in countries which have become stupidly wealthy—speak for themselves.

Read my comment again. I deliberately did not use the word "teachers".

> education academics and professionals

I've listened to a handful of podcasts with education academics and professionals talking about AI. They invariably come across as totally lost, like a hen inviting a fox in to help watch the eggs.

It's perhaps to be expected, as these education people are usually non-technical. But it's definitely concerning that (once again) a lack of technical and media literacy among these education types will lead to them letting (overall) unhelpful tech swarm the system.

Thought this would be a thread about the crazy offers being handed out to join AI teams at Amazon, Meta, Google, OpenAI, Anthropic, Apple, or any of the massively capitalized 'neolabs'.

Hey, author here :)

I wrote this as a first step in exposing our internal GPU reliability management, inspired in large part by SemiAnalysis' focus on industry best practices in the ClusterMax report and Lepton's `gpud`.

NVIDIA GPUs have 172 "Xid" error codes and increase the active population with each new major driver release. Coming out of 2025, we have a good handle on which Xids (and SXids) are critical and can be automated away.

The interesting next frontier are the Xids that sit ambiguously as maybe hardware issues or maybe application issues. Xid 31, GPU memory page fault, is the most dreaded code, because ~95% of the time it's an application exception but it's pretty tricky for users to debug and confirm.

Automating Xid 31 handling is my new GPU reliability holy grail.

"selling premade software assets" is a business, and it's the business both Tailwind and Envato were in. Both businesses got hit hard by AI. Check out Envato's homepage now. It's unrecognizable from what it was in 2017, and completely genAI oriented.

It is a business. Envato was a billion dollar business in 2017. I agree that AI makes these kinds of businesses vulnerable, but it's overstepping to say that these things aren't businesses.

haha, of course it was posted by eatonphil.

I doubt 47 points was enough to get anywhere near the top of the front page? To me this is way too impressive for 47 points and 6 comments.

This has been posted to HN about a half dozen times now and never taken off, so apologies but this needs an editorialized title.

"Dude comes out of nowhere, posts one of the most impressive tech demos/environments ever seen, refuses to elaborate and leaves"

"This is beyond insane. A lot of work must have went into making the underlying system, but the developer experience looks absolutely stellar. Taking deterministic execution to a whole new level."

They’re completely different characters and their qualities are completely different. You can easily admire tragic figures, as well.

As whole essay could be written about how this comparison misunderstands The Great Gatsby.

Unironically dril's tweets[1].

Part of the problem of our time is that shared culture has significantly receded. There's little capacity to maintain "classics" as we understand them today. Take any massive artistic output (film, book, TV show) and it's nowadays either not seen/read/heard by more than 20% of the population or it's a flash in the pan hit which will be forgotten in another year or so (e.g. Barbenheimer).

1. https://www.reddit.com/r/dril/comments/cqde0e/pound_for_poun...

This article ends up seeming like an ad for some dubious derivations of the original novel.

- "Take, for instance, Michael Farris Smith's new novel, Nick. The title refers, of course, to Nick Carraway, the narrator of Gatsby, who here gets his own fully formed backstory."

- "Jane Crowther's newly published novel, Gatsby, updates the plot to the 21st Century, and flips the genders to feature a female Jay Gatsby and a male Danny Buchanan."

- "And Claire Anderson-Wheeler's The Gatsby Gambit is a murder mystery which invents a younger sister for Fitzgerald's eponymous anti-hero: Greta Gatsby – get it?"

Writing that is aimed at the truth should be written to best convey it. But truth is not the only aim of writing, and it's most often not the primary aim.

Truth is also not the aim of these:

> Suits for interviews. Combed hair for dates. Makeup to feel prettier.

The latter is for the pursuit of beauty. The first is because the almighty British empire started wearing suits and we all followed.

Wearing a grubby t-shirt to a job is not because you value truth so much. It's failing to value the other things.

The lads on the Manhattan project almost surely dressed better than a truth-loving rationalist.