This is one of the things that made me ditch Cursor: when they're constantly shifting around keybindings, overriding long-held conventions, and changing UI, etc etc; especially without opt-out (or opt-in for that matter).
Tie that in with a release cycle that force restarts the IDE, its a recipe for flow-breaking frustration that makes a tool untrustable. When its untrustable, it gets removed.
Now I don't let my tools auto-update. It was an experiment to allow the auto-updates to begin with, because new tech paradigms, yet its clear the end user still isn't accounted for properly.
IMO because the product roadmap (such as it might ever be) is usually only influenced by poorly targeted analytics (more is not better), and a small minority of vocal users; leaving the rest of us in the lurch.
To add to this further, I've had great success following The Body Keeps Score; seeing it as a repository of past stress and trauma.
As part of this, I've been able to locate and work through stress and trauma activations in my body, where normally they'd cluster around my head and never actually get resolved.
Every time I go to work out, I pay attention to what areas of my body arent responding, are activating oddly; and I'll work to strengthen the foot-to-neck paths. It started with a back injury and has resulted in me finding I needed wide foot shoes and changing my entire stance, posture, complex movements, etc.
Some times I find it odd that I don't have that daemon running around yelling, because hes now activated in my body, and all I have to do is stretch.
For me, it was when a movie wasn't the way I remembered it. Then I found a pirated copy.
Turned out the 'official' release was heavily edited, with tone, characters, and even some plot had been completely reshaped. I've found this to be increasingly prevalent, and not just in a "made for TV" or "adapted for Flying" type modifications.
I did something like this, in a much more limited form, when putting together my personal site a few weeks ago. https://1ps0.info/
I had a friend try to run 'sudo shutdown -r now'. It inspired a much more thorough approach to the terminal functionality, but I didnt want to rabbit hole too far.
As it stands, initially it was a cyberpunk theme but i wanted a vscode-like professional theme as default, so you can toggle between them through cli. Lots of fun to be had with eastereggs.
My experience informs my opinion, that structure is more important than specific tone.
IMO If LLMs are made from our language, then terminology semantics plays strongly into the output, and degree of control.
Some people rage when the machine doesn't work as expected, but we know that, "computers are schizophrenic little children, and don't beat them when they're bad."[1] ... right? Similar applies to please.
I've had far better results by role playing group dynamics with stronger structure, like say, the military. Just naming the LLM up front as Lieutenant, or referencing in-brief a Full Metal Jacket-style dress-down with clear direction, have gotten me past many increasingly common hurdles with do-it-for-you models. Raging never works. You can't fire the machine. Being polite has been akin to giving a kid a cookie for breaking the cookie jar.
It is funny though, to see the Thinking phase say stuff like "The human is angry (in roleplay)..."
Following in the tradition of the Tao of Programming[1] and the Unix Koans[2], this initial set of Koans for LLMs seeks to capture the line between understandable tool and misunderstood hype.
This is great, going to see how it integrates with my own agentic work. Code looks well laid out and concise.
I really like how you demonstrate the clear workflow and obvious outputs in the video.
FYI feedback on the branding end: I'm starting to notice a negative reaction to anything termed 'vibe' (though i love the wordplay in the project name).
Also I was a bit startled by the splash image.
That said, this is definitely a cut above the average project with a clear targeted tooling win. Congrats!
I ended up taking my own path towards small-scale tooling.
I've been using the binaural audio trick through youtube for some years now, particularly the rendered 40hz trick to clear my head.
The insight was that a lot of the noise pulling me off center was coming from built up clenching of muscles around my head and back, and a destabilization of core posture while sitting, leading to shaky leg and such.
The project I built was simply a mixable tone dashboard for different frequencies. Ive been dogfooding it hard, and its my goto to push around my headspace to clear out the noise and make a protected space for work. The easiest litmus for tone choice has just been, 'does this help or hurt'. On my better days its like im giving my brain a massage and I have the most productive times.
A note: i load the file directly from repo when I use it on Firefox, theres an audiocontext policy i havent been able to figure out. Should work great on mobile and elsewhere.
Barkley opened the door for me to the term 'executive dysfunction'. This better describes my symptoms than the blanket ADHD term that is now oversaturated IMO. Its given me much better anchors to reason with the issues that mess up my day(s), and ive made notable progress since the discovery of his work, at a time when I had been plateauing.
Simply being always-online and sending telemetry, doesnt preclude the ability to remotely intercede in the car's operations.
I think there's precedent with the OnStar system in other cars, and there has anecdotally been evidence of various systems being operable by some kind of admin terminal.
I think "summon mode" is actually a great demonstration of how you can remotely access and do some form of piloting the car without a driver. There's also the disabling of features remotely on resale, and probably other examples.
That's always been the catch for me; its an always-on remotely controlled camera-mobile, of which you happen to sit in the drivers seat and pretend you own while it drives you around dangerously.
A small note: 1.96.2 is the VSCode version, the Cursor version latest i think is 0.46.x.
I'll also say that "Restore checkpoint" often causes crashes or inconsistency in the indexed files. I've found using git and explicit full reindexing has solved more problems than the AI itself.
After working with this tooling for about a month, I've found the biggest struggle is adapting my mental model as the paradigms change. Because make no mistake, this is a research-stage tool that has been thrust into the spotlight while in its infancy.
There's definitely shift between versions of cursor. One time I had an unpleasant crash that also upgraded the version, and suddenly .cursor/ files I thought were working (but hadnt been) showed up mid prompt, and derailed the entire composer run.
I've tried ideas ranging from TODO documents, append-only work logs, blueprint documents, faux-team interactions through role definitions, and much much more.
The most success I've had is by adding manual lines at the end of every prompt, akin to phrases like "be respectful of the existing code and functionality, make deliberate changes with caution and consideration". Beyond the normal "plan and take it step by step" type phrasing.
It would also be nice if .cursor/ rule files actually were actively used, instead of sometimes being hooked in if you wrote the description right.
Sometimes I have great success at utilizing specialized roles, with named team members and explicit direction, to elect the "appropriate" role for the items being worked on, and discuss pros/cons and facilitate brainstorming.
At the end of the day, I think I end up doing an annealing pattern for changes:
- Build the features and breaking is okay so long as we get the features.
- Distill the built features into a doc with examples and roll back the code by git.
- Build a v2 based on the distillation.
- Rinse repeat, usually until v4 or v5, when i tear it all down and write it 99% by hand from the distilled documents.
I havent personally tested it, but I also havent allowed it on my phone.
I figure either its time to drop the smartphone concept, as I barely use it for more than calls, or try to wait it out for them to realize I/we didnt want or ask for it.
I'm curious, for such a straight-forwarded added functionality, why you would:
1. add a very heavy editor framework (monaco) for a configuration page?
2. use actual javascript to check partial urls one at a time?
For 2. this would be better handled by a json or similar that just maps Firefox's approved url match patterns[1]? Using javascript adds potential execution vulnerabilities and possibly parsing complexity that isnt needed.
Also a word on partial urls. Given that many sites will use subdomains and redirects as a norm, expecting a site to be like "www.amazon.com" seems like it would generate a lot of excess code in order to maintain url targeting. This comes from working on a macOS plugin that did similar javascript injection based on url targeting (BeardedSpice[2]) and would have regular targeting errors, especially with more complex SPAs.
Either way, keep up the good work! Love to see projects like this attempt to improve the QoL around the browsing experience.
A question I've had on my mind for a while now, watching this progression of fake or poisoned jobs, and miles of automatic systems to navigate:
At what point do people consider the well poisoned? Where they just check out and stop applying, to specific companies or in general, because its very very obvious that there isnt actually a valid hiring market at all.
I ask this question, because I've already passed this threshold, and have instead devoted the maximum of my time to personal ventures.
> Macroplop