Hacker Newsnew | past | comments | ask | show | jobs | submit | aiiotnoodle's commentslogin

A little while ago when I had a free weekend. I decided I would pre-compute some musings of an LLM trapped in a computer, I had a raspberry pi 1 (or maybe b) lieing around that I had a framebuffer screen for. I think it was a non-standard TFT_ILI9325 from Aliexpress.

The idea is it would display some helplessness comment on the screen like "Help I'm trapped in a computer" etc etc.

Eventually I got the thing booted with it's original OS image and it was running a custom kernel and had the hardware still attached and worked, but it was very outdated so I decided backup and update it, more of a technical challenge than something practical.

I gave it a few tries but eventually the screen would stop working, not helped by the thing being quite slow. Updating in place was probably harder than a brand new OS. Eventually I determined that I'd need to build a custom kernel image using a newer version, something I've never done before and for me it was quite difficult and a little before AI had agents and such to help you along.

I managed to build a custom kernel but the driver for it had been patched out and moved to an optional thing that I had to enable, I was able to build the custom kernel but for some reason the screen wouldn't display anything like it did on the original OS.

I spent a further few weekends on it working really hard to get this thing to work. Eventually I calmly picked up the screen and intentionally pushed down hard with my thumb breaking the screen irreparably.

It was destructive yes but the joy I feel even now being free of this frustration is immense.

I said I would order a newer screen but have yet to do so.


There already is some intelligence fed back to the police somewhere in the 3d printed guns supply chain.

You see it a lot on crime investigation shows. Pretty sure I've seen it on 24 hours in police custody and at least one other show or documentary.

edit: police acting on a source, a lead or some other un-named entity.


I've noticed using antigravity and vscode, Gemini 3 pro often comes back with model too busy or something like that and basically 500s.

Seems like capacity because it works a lot better late at night.

I don't see the same with the claude models in antigravity.


I also noticed that and I also noticed that it starts to struggle when the workspace "tab" you're working in gets longer - it basically gets stuck at "Starting agent ...". I initially thought it must be a very big context that the model is struggling with but since since restarting the "app" and kill -9 fixes it, it suggests that it's a local issue. Strange.


Anecdotally, I notice better performance and output quality across most providers outside of 8a-5p ET.


Yeah that's a separate issue though, it predates the time when the looping issues got really common, for me at least.


How does this compare to LENS? https://lenshq.io/

Always used the free and paid version and never heard of headlamp. Having a look its basically the same but for free.


I never really felt great about Lens. The interface and workflow just never resonated with me. No shade to the crew behind Lens, it’s purely subjective opinion, and I’m sure a lot of people would say I’m wrong.

I like Headlamp. It’s much closer to how I think these kinds of apps should work. Prior to starting work on Luxury Yacht, my favorite app in this category was https://infra.app/ but unfortunately that has seemingly been abandoned.

As far as how they compare… I think they’re all pretty comparable in terms of features. It’s just a question of finding the one that has the UI you like.

I did try to build some stuff into Luxury Yacht that I haven’t seen in other apps, though. There’s an object diff panel, where you can compare two objects (even in different clusters) and it will diff the YAML. I also have a json log parser, that will render json logs as a table so it’s much easier to read. Just small things like that.


Kubernetes Ops person here who opens HeadLamp at start of the day and leaves it open. Lens is much better than HeadLamp IMO but if you are cost sensitive, HeadLamp can probably get you 95% there.


Can you elaborate a little, please? Also, do you use these tools for managed K8s or standard one?



All the cross tenant inconsistency really needs to be ironed out, I'm not sure if it's just my org but half the features of calls are randomly disabled or enabled based on who originated it.

My favorite was when I entered VR during our standup on our otherwise quite locked down and very corporate environment.


While I think this is true now, SEO will adapt. I hope AI companies are keeping their scraped pre AI data sources in their databases as a sort of low-background steel source to use when this happens, although I suppose they are the ones selling shovels.


Practically, verified boot is hard to not have a "this phone has been tampered with" message on boot, the backups generated often have encrypted user data that is usually wiped on boot-loader unlock, you'd also need to unlock the phone or have the user give the pin over and most of the apps that implement root checking SDKs would prevent them from working.

I'm not saying its impossible but it is hard to do at present in a way where if I came and picked up my phone again, I'd not know something happened to it.


Yes and no. In a video I watched on YouTube the people fencing the car had scratched off VINs in the bonnet, door and windscreen, painted and re-etched the exported car's VIN and gone out of their way to find a reasonable fake V5 certificate (UK equivalent of a DMV cert I think) with similar specification as the stolen car (or found the docs first).

The car was sold on, eventually went to Copart with a blown engine and then the YouTuber found out through his videos that the car he owned was stolen and the original had been exported because the interior color was not the same as the decoded VIN. Only when he took the engine out of the car and compared the engine number with the one in BMW's database and the reported VIN in the infotainment was he confident that the car was stolen, same for Copart (who wouldn't entertain the car was stolen).

I think if it wasn't a famous YouTuber who bought the car, it's highly possible that the stolen car would go nu-noticed throughout its lifetime as stolen, even if taken to a main dealer. If I recall correctly the car reports he used (maybe car-vertical) also didn't pick up any discrepancy.

For the criminals its good business, you find a 30k plus car, pay for a clean VIN from cypress or somewhere and then do the damage to the car to re-new it as a different car, even if it costs 10k to do, its a lucrative 20k "profit" and thats on the high end, seems like cars can be stolen overnight, especially ones the criminals specialize in.

edit: https://www.youtube.com/watch?v=RI4S2LT_ntE


Just to add to this, the "clean" exported car's VINs were bought off snapchat for ~4k per car. (for high end cars.)


On your pricing page it is not clear what integrations Plaid has or if my banks are supported.


This is on Plaid’s site. I’ll add a link to it, but here it is:

https://plaid.com/docs/institutions/


This was not my lived experience. I wanted to use the most common banks and most would not let me use it.

Chip contacted me at one point via their live assistant randomly without my doing and told me to stop using the app because they would soon be enforcing that rooted devices would no longer work. I continued to use the app rooted and nothing came of it.

Barclaycard, Nationwide and others don't let you use the app or require some circumvention of their detection to allow access.

Sure there are plenty of other apps, but those apps and banks have a worse product I found.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: