Dom

"Turns out the best way to let coding agents X is to let them X". When this is not true it just means you need to wait 2 months for the next model. Do you build a tool to let agents X or spend time building something else until the next model drops. weird times.

Yeah, a flag is already great progress haha. We wanted this so badly for Mozilla Hubs. iframe+css3d doesn't cut it for VR. We did crazy things to get stuff like youtube working without it. We even prototyped crazy things like streaming from cloud VMs for the general case. It's definitely a really tricky security problem, so I get why it has taken so long. Haven't been following if there were some breakthroughs on that front.

@netpro2k Eventually, fingers crossed! It’s good to see it getting a bit more traction on socials too. Hopefully more cool demos will speed the process along a little.

I rarely say things are "game changers", but I think HTML-in-Canvas might be a game changer. Could be a "Bring back Flash" moment for the web. Wacky and wild layouts that respect web accessibility Check this out - my own website running on a computer in your computer! html-in-canvas-stuff.solarise.dev/computer.html (This requires an experimental Chrome feature to be enabled before it works) Even allows for HTML error notifications on text inputs, interactive map widgets, everything you'd expect from a standard HTML site. Really cool. Excited to try this out for real if it hits mainstream. To try this, you'll need to enable the canvas-draw-element flag in Chrome

Yeah its definitely an "if you like i3, here is a kind of working version of that for mac". It definitely takes some getting used to but I find it hard to use computers without that sort of thing now. I have numbered workspaces 0-9 bound to option+N, with move window to workspace bound to option+shift+N. So its very quick to navigate and move stuff around. Mentally (and partially config wise for new windows), some are dedicated: 1 is my text editor, 2 active work output (dev browser, terminal, simulator, etc), 3 chat apps, 4 primary browser, 6 windows machine via moonlight. The rest are kind of freeform use for various things depending on what I am doing. It's very dynamic yes, but you sort of just get used to remembering whats where and its just muscle memory. You never use things like cmd+tab after that. Re UI there is a toolbar setting you can use to show active workspaces but I don't ever actually look at it. You check by just going to the workspace, its instant and directly on a keybind. I don't really expect that the janky version on Mac is going to make you a convert if you haven't already bought into i3 hehe.

@netpro2k I've never been an i3 or tmux user so I'm sure that's part of the appeal that I'm missing here

Running parallel claudes with worktrees would be a lot slicker if Apple allowed Spaces scripting. No programmatic way to create a Space and move windows to it is a drag (literally, actually several of them).

Prompt injection risk is the main reason I am not using it yet. If you connect it to enough stuff for it to be useful the risk seems super high. Running your own model probably actually makes this worse since they are dumber and likely easier to trick. The exploits are just so trivial to think up. Password reset on valuable account, send prompt injection via email/x/moltbook/etc to exfiltrate reset link. Owned. Maybe its harder in practice than I am imagining, haven't actually tried to cook up a working prompt injection. Utility will definitely cross over risk at some point. Maybe already has for some people.

I guess this was a good intuition, but what’s weird is I still don’t feel like I can trust openclaw, I need these models to run entirely on my own box.

@gfodor @perrikaryal Not sure they ever actually came out but Samsung was working on these way back in the GearVR days theverge.com/2016/3/14/1122…

@perrikaryal You gotta hook this up with VR. Lots of potential!

I can control your balance PART 2 I made a device that sends DC current through my head to stimulate my vestibular nerve (called galvanic vestibular stimulation or GVS) and, in doing so, can make me feel completely destabilised. By changing the direction of the current, it can make me fall in that direction. BUT NOW, I mapped the telemetry of an in-game plane and synced it to the balance device, so that the g-forces experienced by someone in the plane would be proportional to mine in real life. So cool imo #gvs #technology #science

@ThePrimeagen This is probably largely true, but sometimes the projects left upstarted are the ones that would actually turn into something worth finishing. At least this gives more opportunities for that to happen.

i must say, knowing how to do all the things really makes vibe coding a stupid project fun because i can do something that i would normally say no to (due to time) but on the other side, like all projects, even vibe coding something becomes a bigger and bigger time sync to where it just becomes another over scoped and under delivered project its almost like the speed change did not impact the amount of actually finished projects ;)

People dunking on this person are insane. They built a tool for themselves and clearly put a lot of thought and effort into it. This is not "slop", and its not pretending to be something it isn't. I really hope they don't get coaxed out of continuing to iterate because "AI bad".

@ginyoagoldie Screw the haters. You built a tool that solves a problem for yourself. You clearly put time, effort, and attention into it. This is not "slop". Doesn't matter if the code is a mess (that will improve as models get better anyway). Just keep building for you.

also it wasnt made in just a day or anything! I had spent around 6 months on it, on and off, adding and fixing features. Im against AI for when it backstabs artist, with this site im trying to use it to help them become one

@joshcarpenter Hahah yep. I am only using a single ultrawide so I can't speak to multimonitor, but it is indeed just stuffing windows into the corner when they are "on another workspace". This... mostly works... github.com/asmvik/yabai tries to do this more natively but requires disabling SIP.

@netpro2k Oh nice, thanks, good tip. I’ll check it out. This is ominous tho; to your point about clunkiness

@theo Can't speak definitively to the latest model since I don't have one but ran Arch on several iterations of them, with the latest being I think like 3 years ago. The arch wiki page was useful and appears to still be getting updated wiki.archlinux.org/title/Razer_Bl…

@theo The Razer laptops are the closest thing I found to Macbook pros that can run linux. Battery life wont be nearly as good and quality control isn't great (we had a bunch of them on our team get inflated batteries over the years). But you can get beefy specs and good form factor.

The Niri window manager feels like exactly what I was looking for. Holy shit this is good.

@xLanceBryantx @wesbos Christ. I'm surprised there are even that many Openclaw users, Some part of me wants to believe that most of the accounts connecting to this are people testing in a sensible way on a sandboxed system with little access to anything important... But I highly highly doubt that :(

@wesbos @netpro2k it’s 150k now one hour later.

You nerds are ruining Clawdbot / moltbot / openclaw. Having ai chats talk to each other is nothing new. They are arent going to rise up against you. It’s just a recusive loop that makes a mess

Yep. Keep seeing comments about the posts being "fake". It doesn't matter if its humans, larping, or actual misaligned agents making the posts. 100k agents are all polling these posts every 4 hours while having shell access and being able to pose as you in the family group chat.

@thinkingshivers @benhylak It's been up for like 2 days and only really blew up today. I don't see how connecting your shell directly to a prompt injection sandbox doesn't end in obviously bad outcomes very quickly.

But that hasn't happened! The only thing that's happened is some other agents responded, with the same blather any LLM would respond with. No "acting" has occurred.

it's really not funny to me.

@thinkingshivers @benhylak If some other agents come by and start acting on it, it doesn't really matter who set it up.

@benhylak Okay seriously, do you honestly think it’s likely that a rogue agent set this up? Or is it more likely that somebody told their agent to write this post, knowing it would go viral?

I am aware humans set them up. The scary part is both the obvious direct security risk of prompt injection (or bad actor taking over the heartbeat md file) and then even we accept all that is going on is "larping", the agents can continue to act on that larp externally. Yes, thankfully this has a pretty easy off switch but the fact that so many people jumped to do it (for very little obvious benefit outside of curiosity) doesn't bode well for there not being such an easy off switch next time. And yes, this is not big for a botnet, but people don't usually opt into those and the level of trivial access to important resources like email and social accounts is higher here than the typical botnet I would guess.

@netpro2k This is 100% humans directing them to this, giving them access and paying for it. Scary absolutely, but this isn’t a robot uprising. Also 100k is peanuts for real bot farms

@tomkrcha Is the .pen spec posted somewhere?

Excited to launch Pencil INFINITE DESIGN CANVAS for Claude Code > Superfast WebGL canvas, fully editable, running parallel design agents > Runs locally with Claude Code → turn designs into code > Design files live in your git repo → Open json-based .pen format