richstokes

Same. Only because I broke my M1. Performance wise the M5 barely feels any faster to me in 90% of tasks

Ohhh so that’s why I got those price increases emails. Yay. What are some good alternatives?

Correct

We had the Roamer, British school, very early 90's.

I remember programming a logo/turtle type thing on the BBC Micro, and then I think there was a way to copy the program on to the Roamer directly from the BBC. Or maybe we did it manually. But it was amazing to me, and I always felt sad we didnt use it more!

Stopped using them when I realized I had stopped seeing posts from my actual friends and family.

It’s mostly endless reels and ads now and I have to scroll for way too long before I see a post from someone I’m connected to/friends with. Bored of it all.

Doesn’t load.

Is code being converted to / represented in maths that much more helpful for an LLM that deals in language primarily, by definition? I have no idea, interesting though!

I feel the same sometimes. I don’t see how this doesn’t lead to massive job losses. The thing we spent our lives/careers learning is now worth basically nothing in comparison.

AI is only going to get better and do more with less humans in the loop over time.

That said, I do also relate to the "coding was never the hard part"-type arguments, and much of my day is spent on the stuff in between writing code.. but still.

Do we think that was necessary? What would have happened if OP had just asked it to rewrite and test/validate each piece as it went until everything is verified and complete?

My gut feeling is this is doing way too much, and it would've figured it out.

Having the LLM run the proprietary app through ghidra (which I think even has an MCP now), would likely give it some useful clues too

First thought; this could be turned into a really cool password generator

Had a moment of nostalgia and started building an IRC client for macOS. It’s totally free/open source. And has been fun to customize / add the features I like to see while avoiding the ones I don’t.

https://github.com/richstokes/Netsplit

Idea: Support providing it the URL to your own / arbitrary git repo, so you can see it type out whatever code you like :-)

If the AI could understand the save file format for Minecraft, it could manipulate that

Device-bound would be a nightmare. I don’t want to have to think about different credential for phone vs. laptop, etc.

I got steamos working on my nvidia PC with a bit of messing about. I really like it, it’s a fun experience and after this initial hack, everything has worked great.

https://github.com/richstokes/SteamOS-Nvidia-Drivers

Is it a nightmare? I pasted the first, zoomed out, image of the shirt into ChatGPT, and it decoded it / let me know what the bash script does in one shot

Pretty bad. It seems to have very limited context, and it mostly cant follow up on things if you have additional things to ask. It was ~fine when it first came out, but seems to have made basically zero progress since then, and frankly feels pretty dumb compared to talking with modern LLM-type apps.

Andddd now everyone knows that an arbitrary text string in a device hostname is enough to ground a flight.

What if you deploy to a different CPU architecture than you develop with? Some languages/libraries use pre-compiled binaries as part of their packages.

I wonder if you could repurpose the cars SIM card (eSIM or otherwise) / cellular modem for your own use? Like presumably the manufacturer is paying a data plan to a cell provider. If the connection is not locked down to specific traffic I wonder if it could be reused

It was the Wild West back then. Good times!

Kinda related but I used to practice guitar in my dreams. If I had been learning something I’d often dream about playing it over and over again, and even going beyond that and figuring out “solos” and melodies and stuff over the chord. Can’t be sure if it translated into any real life skill, but it felt like I was actually learning or at least strongly reinforcing what I’d been practicing.

Was thinking the same honestly. GH is very sticky though, especially when you have actions and all kinds of other integrations set up. But it’s just kind of absurd at this point how many outages they have.

Are there benchmarks of this / what’s the best way to compare it against paid models? With all the rate limiting in Claude/Copilot/etc, running locally is more and more appealing.

Do you use it with ollama? Or something else?

My personal Claude sub (Pro), I can burn through my limit in a couple of hours when using Opus. It's borderline unusable unless you're willing to pay for extended usage or artificially slow yourself down.

The problem is half the time you don't know you need the better model until the lesser model has made a massive mess. Then you have to do it again on the good model, wasting money. The "auto" modes don't seem to do a good job at picking a model IME.

Agreed. What network are they forgetting here? Pretty sure I've only ever seen that apply to wifi. Are they trying to say you might need to re-pair a bluetooth speaker on a date? I don't get it.