HNHacker News
TopNewBestAskShowJobs

jimmaswell

6,461 karma · joined December 16, 2012

https://www.jimm.horse

http://armchair.dashtel.ca/~jimm

submissionscomments
jimmaswell··on As AI eats the web, the internet’s collective memory is disappearing
> Do you feel you have the right to decide whether to exchange your ideas or not?

If I've said something around someone else or posted it online, then I don't feel I have a right to someone's use of that information, no.

> Yes, you’ve decided to share that information freely with that person.

And you decided to make this post on a forum that an AI is certainly scraping as we speak. If you don't want anyone or anything to "extract" your ideas then don't express them. We never would have left the caves if everyone had that mindset, but it's your right.

You keep saying "extraction" as if something is taken from you whenever a machine learns from you, but it's not a zero-sum situation.

Scenario A: Alex tells Sam that PHP 8 has added union types. The world gains 1 point of value because Sam has gained 1 point of value.

Scenario B: Alex tells Sam that PHP 8 has added union types, and a robot overhears it and tells 10 people. The world has gained 11 points of value.

I truly don't understand preferring scenario A.

jimmaswell··on As AI eats the web, the internet’s collective memory is disappearing
What LLM was this? I'm simply not having this experience at any regular frequency these days with GPT Sol as my go-to. If I know I'll find something with a Google query, then I still go straight to Google because there would be no point waiting for the AI. For more vague cases where I don't quite know what I want to search or how, or I want an overview like you mentioned, I haven't had a significant amount of trouble, and I do verify when it matters. It's done well with broad comparisons like comparing options across eBay and Amazon for some odd category of used hardware for example.
jimmaswell··on As AI eats the web, the internet’s collective memory is disappearing
> you're almost always able to find what you're looking for very quickly[...] You'll spend the same amount of time trying to fact-check an LLM

I've been an expert Google user for over a decade and I can only partially agree with the first statement, and not at all with the second. Yes, a search engine alone is fantastic at finding things based on keywords if you know how to invoke it properly. However, there are lots of things one may want to find out which can't be reduced to a keyword search, because you can't have the vocabulary to search for it directly unless you already know the answer. Indirect questions such as "framework options to do x and y in z situation in this language". The best you could hope for pre-LLM was to find forum posts asking the same or a vaguely similar question and comparing a lot of options, finding out you picked a dud after spending an hour on it because it's fundamentally incompatible due to reasons, searching again, etc. It's hard to overstate what a massive improvement LLM's are for this kind of search to find and compare options for exactly what you're asking for given the context of your situation.

jimmaswell··on As AI eats the web, the internet’s collective memory is disappearing
I feel honored if my ideas are processed by an AI and then used to help others. I feel no more entitled to exclusive use or credit for my ideas as used by AI than I would if I had talked to someone at a conference who went on to be influenced by my ideas to do something good after forgetting my name.

The prospect of all human ideas accumulating inside a machine that makes these ideas accessible and useful to everyone on command is beautiful, not discouraging. Humans aren't discouraged from creating or exploring in Star Trek because of the computer, but I could imagine the Ferengi computer being hobbled at the kneecaps by requiring licensing and credit for every single idea inside it, and the user needs to insert a coin every time they want to ask it a question, which gets divided among every living Ferengi and the estates of every long-dead Ferengi whose writings influenced the output. We should not aspire to be like the Ferengi.

jimmaswell··on The AI Apocalypse Is Here
I don't get arrested for saying "Kent State Massacre" (/Tiananmen Square) or insulting the police for a start.
jimmaswell··on The title cards in Blade Runner are amazing
> I’m staring at a fixed-width typeface all day. Apple provides a functional and gorgeous fixed-width variant of their San Francisco typeface called SF Mono, which I’ve been using for months, but, well, I have a short attention span — typographically speaking.

Tangential but I've been enjoying GNU Unifont for this purpose. I like the look of it and it does a good job supporting a huge range of characters. Terminals warn me that it's not technically monospace, but for all intents and purposes in a terminal windows it is. I have it as the default bundled font with a project that renders custom glyphs on the fly for display on real VT320 hardware as DRCS.

jimmaswell··on I’m leaving OpenAI to build telepathy
Could you explain exactly how this project that aims to let people beam thoughts to their computers is "actively making lives more difficult and compromising the social integrity of human civilization" as opposed to "assisting lives and promoting ease"? What could be more empowering and convenient to a user of a computer, especially one with a relevant disability?
jimmaswell··on I’m leaving OpenAI to build telepathy
I feel the complete opposite.

Years ago I was close to disavowing technology and only doing the minimal effort needed to keep up with my job, because I'd gotten bored and disillusioned with it. Anything more than a small hobby project had accumulated too much complexity and busywork for me to feel excited about working on it. I suffered through it for the sake of the few things I cared about most, but it took its toll on me. I didn't have the infinite energy I may have had for my favorite projects in high school anymore.

But then this new wave of AI came along. First there was Copilot, and I was in awe not necessarily at its immediate practicality, but the potential it demonstrated. I felt the embers of passion tentatively peek their heads out from wherever I'd previously buried them.

Every new leap, computing became more and more rewarding again.

First general searches became incredibly easier to find the relevant bits I wanted from. I actually wanted to look things up again now that I didn't have to grok dozens of pages of Google or Stack Overflow. I got my custom home router configured the way I wanted it instead of giving up or settling for copying some known config that wasn't quite what I wanted, after a small amount of trial and error.

Then it became reasonably competent at simple but tedious tasks, at least 90% where the last 10% was easy to fix by hand. Enormous git conflict to slog through? Instead of staring at the screen feeling my heart sink and wandering off to bedrot instead, I felt excited at the prospect that I could instead command the computer to do it for me, and it was actually motivating to get that to work. Instead of feeling too exhausted and annoyed to keep working, I felt rewarded and refreshed.

Then it started writing entire code files, then entire features, then entire applications all on its own, with my guidance and input on architectural decisions and aesthetics. Performing entire FreeBSD upgrades while keeping custom kernel module packages working, fixing Gentoo systems with messed up USE flags and an otherwise intractable dependency hell to sift through, hell, even writing entire kernel patches to fix sleep mode and graphics problems I'd been having on my 2010 Macbook Pro with Gentoo installed. I feel something reminiscent of my memories of opening Christmas presents as a child when the AI is done making something I'm super excited for.

I'm eternally grateful for the current AI zeitgeist in the tech world. We're living in the best time so far in history to be into tech. And things like Conduit? How would you not be excited for something with so much potential to be made? Even if you don't like it, imagine what it could mean for people with communication disabilities, carpal tunnel, etc. I'd love to be able to just type with my thoughts, and sending them straight into an AI sounds awesome too.

If this is satanic then I don't want to go to heaven. I feel more joy, excitement, and liberatory potential in technology than I have at any point in my life. I even feel more liberated to learn more - without the tedium raising my cortisol and stealing my attention all day, I'm actually more inclined to do things like write code in vim on my VT320 or read over code there with no distractions or AI input. I've reaped great rewards from being able to think more clearly and at higher levels of abstraction, free of the bonds of bullshit and tedium, at peace in the knowledge that I will never have to shave a yak against my will ever again.

jimmaswell··on SQLite Critical CVEs or LLM Slop?
It actually is an illusion though. The narrative the brain makes up after the fact for why we did something is not perfectly correlated with the actual reason. It's just very good at its job of spinning a convincing narrative, even to itself. This is demonstrated in split brain patients: they're given instructions to do something that are shown only to one eye, the one connected to the non-speaking hemisphere, and when asked why they did it, the speaking hemisphere confidently hallucinates a wrong answer that the person seems to fully believe.
jimmaswell··on SQLite Critical CVEs or LLM Slop?
I strongly suspect that if you dug deeply enough into how a human brain does reasoning, the answer would boil down to "it tries a bunch of pathways and picks the one that seems right". Pathways weighted by experience, but ultimately still guess-and-check. What would the alternative be?

Maybe our ability to reason is not as mystical and special as our ego might hope it to be, and discomfort over LLM's bringing that to light is the root of some people's urge to continuously downplay and discredit them.

(As far as I can tell as a non-neuroscientist, the literature on how our default mode network and prefrontal cortex interact agrees with my assessment - free association and logical verification respectively)

jimmaswell··on Don't credit the LLM
I find human answers and handwritten code to be much worse of a slot machine. An article (or stack overflow post) has a not-insignificant chance to be riddled with outdated API calls, human-made bugs, unspoken assumptions of knowledge on my part or how the environment it runs in will be set up, etc. Meanwhile a frontier model easily deals with all that for you and gives you something that almost always just works, tailored to your exact requirements, with thorough test cases and validation that it works before handing it off to you. By the time you've hit roadblock number 4 out of 20 trying to adapt that human slop to fizzbar 2.0 because 1.9 was deprecated, the LLM would have been done.
jimmaswell··on GCC steering committee announces AI policy
I'm pretty familiar with the technical details of how an LLM works. They absolutely operate more on a level of abstract concepts than acting like some kind of text-Frankenstein as you're suggesting (as if the model is Frankenstein and its output is his monster).

Here's a very basic example: if you have access to a typical language model's weights, you can subtract the embedding for "man" from the embedding for "king", add the embedding for "woman", and land somewhere very close to the embedding for "queen".

Why is "intelligence", whatever that means, a prerequisite for a machine to process ideas in the abstract?

jimmaswell··on GCC steering committee announces AI policy
It's an invention that can invent things on its own, improve itself, do research, etc. The importance of such a thing should be obvious. The hockey stick that's coming for human progress overall is going to make the industrial revolution look like a flatline.
jimmaswell··on GCC steering committee announces AI policy
How is this encouraging? They're sticking their heads in the sand and dooming themselves to irrelevance. All but the most strongly and wrongly ideologically motivated will contribute to other projects like LLVM when GCC asks them to code with rocks and sticks instead of taking advantage of arguably the most important invention in human history.
jimmaswell··on GCC steering committee announces AI policy
No, you can't copyright an idea, only an expression of an idea, and LLMs operate at the level of ideas. They don't literally stitch together code from training.
jimmaswell··on Kimi-K3 on HuggingFace
I've had a steady 2 gigabits (my maxxed out ISP bandwidth) since I started the download at launch.
jimmaswell··on I think you might be fooling yourself with AI
Somehow I've never had an experience like that on Sol. I find it to never really "cheat" like that. I've had all my best results from Codex CLI though, maybe that's somehow different. I also don't go below xhigh for anything serious and if it's having an issue I escalate the thinking mode.
jimmaswell··on I think you might be fooling yourself with AI
In my case it's usually a video game in question, not a life support system, so I'm not poring over every line of code.

I mainly look at:

- The overall architecture to make sure it's in line with what I want

- The points where it interacts with certain other systems I'm concerned about

And I test the feature myself in addition to the test cases. Sol has proven to be outstanding at good test cases and having a coherent view on overall architecture integration. I basically never think to myself "this entire implementation is slop, I need to rip it out and refactor it" anymore.

I suspect the model's capacity to translate requirements into contextually appropriate test cases, and insistence on thoroughly covering functionality in tests, is a large part of its strength.

jimmaswell··on I Think You Might Be Fooling Yourself with AI
Prior to 5.5 I thought the models were very useful but not trustworthy for certain tiers of tasks and needed more in depth verification. Frequent hand edits and times where it felt more productive to just finish something myself than keep rolling the dice on the model not subtly messing it up. But still great for targeted things.

5.5 on Extra High I found to be a huge leap in what it could tackle and how effectively but only up to a point. Still had common failure modes here and there but a big improvement

When I tried 5.6 I wasn't expecting more than an incremental upgrade, but it blew me away with how "smart" it seemed, how it completely sidestepped failure modes I'd been used to dealing with in 5.5, its success rate on the first try for wide-sweeping architectural tasks, and how I've basically never seen a bad hallucination make it through to the final output. It's the first time I've felt like if I left it with a task on the highest setting and it had any way to verify its output, and told it to keep going, it could solve almost anything. This could all be incremental upgrades in many different areas of the model+harness, but the cumulative effect feels like crossing that "escape velocity" threshold. The point where there are no more rough edges and it "just works".

I can't speak to others, but my feelings on all of these have been consistent since I tried them in the first place. I've also heard model providers draw people in at each new release with the model at full performance and then degrade it over time hoping users won't notice, which has seemed plausible anecdotally but I have no hard proof.

jimmaswell··on I think you might be fooling yourself with AI
> For things you're already highly skilled at LLMs can be handy but the overall gains, after all is accounted for, are not so clear.

Preposterous. I can tell 5.6 Sol to do something like "optimize this entire subsystem to have no ongoing memory allocations, fix these 5 bugs, implement these 2 features" and go work on other code or do something else entirely, and it does it all flawlessly. All things I could have done myself but nowhere near as fast and right the first time with such thorough test cases. We've reached the point where the AI is usually faster and always less effort overall compared to the old days, even if you have to send a few followup prompts.

jimmaswell··on I wrote an bash enumerator because I was sick of xargs
> Find works, but the syntax is arcane.

fd is a lifesaver: https://github.com/sharkdp/fd

jimmaswell··on Delta flight hit by firework while landing at Midway Airport on Fourth of July
Beats the pants off "please sir, may I have some more" as seen in many other places
jimmaswell··on Completing a computer science degree on Coursera
My day job is relatively boring JavaScript components and SPA's, but even there I find things I learned in my Computer Science degree valuable. "Hey, this looks like a finite state machine.." "This could be a simple domain specific language, good thing I had to write compilers in college and I can easily make a simple lexer/parser.." "This other thing is easy to parse if I ingest it into a lexer-resembling state machine.." And I would think the value of understanding algorithmic complexity and so many other fundamental things is obvious, no matter what someone is doing. And you won't waste your time accidentally trying to solve the Halting problem, among other things. Obviously there's nothing a university can teach you that you couldn't theoretically learn somewhere else but I'm seriously not convinced that a Computer Science degree is useless or a poor signal even for someone doing run of the mill React apps.
jimmaswell··on Completing a Computer Science Degree on Coursera
What kind of work do you do now that you feel is entirely unhelped by your time in higher education?
jimmaswell··on Qwen 3.6 27B is the sweet spot for local development
My partner has been trying various models on our server but we haven't gotten anything to run at a usable speed. Q30H engineering sample (Xeon 8570) with two cpus, 56 cores per CPU, 768GB DDR5 RAM running at 5600MHz, two old 3090s in it at the moment with an NVLink and we could put our third in there. We built this server before the prices skyrocketed because we happened across some Tyan boards on Woot that were absurdly cheap for what they are (the motherboards should be $1000+ but we got them for a few hundred).

This thing sounds like it should be a monster but we keep running into issues of the old GPU architecture, lack of support for AMX or AMX not being as big of a help as you'd hope when it does work, etc. Apparently we only got 5 tokens per second trying to set up Qwen 3.6 27B, and a similarly bad result trying to run GLM 5.2 which fits in memory but the custom kernels we had to try to contrive were too slow. I feel like this system should have tons of potential, especially if something was designed to let the AMX and huge system memory shine.

Does anyone have any suggestions? This thing was fun to set up and it's really cool but it's been a bit disappointing not getting any big tangible results so far.

We have a similar system on a single-cpu Tyan board with 256GB RAM that I'm hoping we might be able to use in conjunction with the first one if EXO ever gets good Linux support for GPU/RDMA over InfiniBand.

jimmaswell··on It's death
> why aren't there more people pursuing alternative lifestyles? Similar to the 60s counterculture.

I think the brony and furry community count for some examples. Brony community has been a major counter-culture since its inception and I truly believe it's had a measurable impact on gender norms and some other areas. Many of us see it as something of a lifestyle where it dominates the spaces we primarily engage in etc.

Furry is even bigger and easier to argue as a popular counterculture lifestyle, and growing all the time (record convention attendance every year).

They're also places where some other things like polyamory are more common, and much more LGBT/etc inclusive than average (which is less of a statement now, but they were both far ahead of the curve on that years ago), refreshingly sex-positive (at least the parts of the communities I identify with).

Personally all of those things apply to me and I love being a part of those communities. Brony community has been a hugely important thing in my life since high school and working on game development there was a big jumpstart on experience working in a team/technical experience. Furry I only started exploring more a few years ago but I've made a lot of great friends and met my current partners there too.

Brony's a consistent core of lifetime holdouts like me and a steady trickle of new people at this point, and furry's growing faster all the time - even the little bonfire meetup at a nature preserve I like to go to had a record smashing attendance on opening day this year.

jimmaswell··on AI is slowing down
> Culture is destroyed, truth is destroyed, democracy is destroyed.

Just because those things happened in Dune doesn't mean they'll come true. It should be obvious what massive potential the thinking machines have for humanity as a whole.

jimmaswell··on AI is slowing down
Can't really agree. It's improved my life more than any other single innovation made in my lifetime.
jimmaswell··on AI is slowing down
Why is anti-AI soothing?
jimmaswell··on Ask HN: What was your "oh shit" moment with GenAI?
15A and 20A receptacles on the same circuit sounds fine as long as it's a 20A circuit? And how could it tell which outlet is on which circuit?
← PreviousPage 3 of 34Next →