I don't think Slack does, nor Github, which is where I most find screenshotting being a problem.
3,626 karma · joined July 18, 2017
Building modal.com.
Previously data & ML platform @ Canva; also Zendesk and Atlassian.
NYC, USA.
1. I have ‘rubber duck debugged’ my own question.
2. I checked that this question hasn’t been asked before.
3. I have noted in my message what I’ve tried.
4. I have avoided the ‘XY problem’ by clearly detailing the core problem, X.
5. I have provided specifics of my issue, not vague references or descriptions.
6. I have provided URL links to relevant content, and where possible the URL links are immutable.
7. I have not included screenshots of text in my message.
8. I have not used obscure acronyms or abbreviations.
9. I have formatted my message well, particularly paying attention to code formatting and headings.
10. I have not just said “hi” and waited for a reply.
Like other posters, I don't think Apple OCR is sufficient to make up for screenshotting. The biggest problem is search.
1. https://thundergolfer.com/communication/slack/2021/02/24/how...
You can go back and read McLuhan, he's great, but a recent and more approachable book on this is _God, Human, Animal, Machine: Technology, Metaphor, and the Search for Meaning_.
Way back in 1969 the utopian vision of technology put humans at the centre. The Whole Earth Catalogue's slogan was “access to tools.” Just _tools_. That same year, technology put a man on the moon.
Unfortunately, if you realize the extent and the history of the problem, you see we're so far gone, miles away from getting a grip.
The horrible us-east-1 S3 outage of 2017[1] was around 5 hours.
I've used Bazel a lot and contributed to it, but don't feel like it's adoptable by engineering teams <100 in size. mise might be an option though.
> At Greptile, we run our agent process in a locked-down rootless podman container so that we have kernel guarantees that it sees only things it’s supposed to.
This sounds like a runc container because they've not said otherwise. runc has a long history with filesystem exploits based on leaked file descriptors and `openat` without NO_FOLLOW.
The agent ecosystem seems to have already settled on VMs or gVisor[2] being table-stakes. We use the latter.
1. https://github.com/opencontainers/runc/security/advisories/G...
I think he should have remembered your name, but he hadn't forgotten you. Who knows why he forgot your name.
As an engineer this page has a real "trust me bro" feel to it. Maybe fine as a marketing and product positioning thing, but not interesting for HN.
gVisor's achilles heel is it's missing or inaccurate syscalls, but the gVisor team is first class in responding to Github issues so it's really quite manageable in practice if you know how to debug and hack on a userspace kernel.
At some point I was even hearing the claim that digitization (e.g. GenAI) was finally divorcing the tight connection between economic growth and resource extraction. I'd bet it's incorrect, but it's much less fanciful than thinking that growth in oil or beef would help us grow without strip mining the earth.
Bray's first issue —the influence of GenAI on labour's (and also the democratic people's) decreasing power versus capital— is much more important and interesting.
It’s not about being pro-AI or anti, left or right. I just read too much on Bluesky which has me thinking “oh, you really have no idea what you’re talking about.” As Steve says, to verify that they’re wrong is trivial, and yet they’re they are.
Twitter, on the other hand, rarely has this problem if you only look at the “following” tab and curate who you follow.
Trenton and Sholto are very much “talking their book”. They’re doing it well, but it’s highly filtered and partial chat.
source: know a podcast episode which got removed because an OpenAI employee used “black box” to refer to NNs.
The second is a counterfactual, and it is correctly deployed to help show the difference between a valid argument and a sound argument. Graham is saying that a good liar presents pleasing and valid but unsound arguments, or rather sophistry.
I think your confusion here is from reading comprehension problems.
Commenting it out improved iter performance by almost 30%
1. https://github.com/modal-labs/multinode-training-guide/blob/...
Nice to see this because I drafted something about LLM and humans riffing on exactly the same McLuhan argument. Here it is:
A large language model (LLM) is a new medium. Just like its predecessors—hypertext, television, film, radio, newspapers, books, speech—it is of obvious importance to the initiated. Just like its predecessors, the content of this new medium is its predecessors.
> “The content of writing is speech, just as the content of the written word is the content of print.” — McLuhan
The LLMs have swallowed webpages, books, newspapers, and journals—some X exabytes were combined into GPT-4 over a few months of training. The results are startling. Each new medium has a period of embarrassment, like a kid that’s gotten into his mother’s closet and is wearing her finest drawers as a hat. Nascent television borrowed from film and newspapers in an initially clumsy way, struggling to digest its parents and find its own language. It took television about 50 years to hit stride and go beyond film, but it got there. Shows like The Wire, The Sopranos, and Mad Men achieved something not replaceable by the movie or the novel. It’s yet hard to say what exactly the medium of LLMs exactly is, but after five years I think it’s clear that they are not books, they are not print, or speech, but something new, something unto themselves.
We must understand them. McLuhan subtitled his seminal work of media literacy “the extensions of man”, and probably the second most important idea in the book—besides the classic “medium is the message”—is that mediums are not additive to human society, but replacing, antipruritic, atrophying, prosthetic. With my Airpods in my ears I can hear the voices of those thousands of miles away, those asleep, those dead. But I do not hear the birds on my street. Only two years or so into my daily relationship with the medium of LLMs I still don’t understand what I’m dealing with, how I’m being extended, how I’m being alienated, and changed. But we’ve been here before, McLuhan and others have certainly given us the tools to work this out.
I'm quietly betting that agents increase the leverage of deterministic, reproducible devbox tech (eg. Nix, lockfiles, package mirroring), and this will end up being a huge win for us human engineers too.
Economics must have made a term for this part of the venture ecosystem. “Seed capital” misses the significant social and class aspects of the behavior.