I tried lane keep assist in an audi a5 sportback, and it was genuinely terrifying to me. It felt like it was trying to drag me into oncoming traffic when going around curves.

Automatic braking sounds great on paper but practically speaking doesn't work out in my opinion, and once again seems to get in the way. Many people have reported getting frequent alerts while driving or even having ghost braking going on with their Subaru EyeSight systems. This wouldn't be an issue if I could confidently remove the system from my vehicle, but right now I practically can't without doing some heavy lifting on my own.

I'm aware that these systems are pretty integrated, so it's not just a matter of pulling a fuse, but doing some type of reverse engineering to take control over how things internally work. I'm doing the research and am willing to put a lot of time in to make this happen, but I'm curious to know if anyone has explored this themselves already and to hear about their own findings.

Comments URL: https://news.ycombinator.com/item?id=48261313

Points: 2

# Comments: 6

However, we've been seeing advancements in compressing context and capabilities of smaller models that I don't think it'd be too far off to see something like what I'm talking about within the next 5 years.

I think companies that are shelling out the money for these enterprise accounts could honestly just buy some H100 GPUs and host the models themselves on premises. Github CoPilot enterprise charges $40 per user per month (this can vary depending on your plan of course), but at this price for 1000 users that comes out to $480,000 a year. Maybe I'm missing something, but that's roughly what you're going to be spending to get a full fledged hosting setup for LLMs.

1) Apple is not a data company.

2) Apple hasn't found a compelling, intuitive, and most of all, consistent, user experience for AI yet.

Regarding point 2: I haven't seen anyone share a hands down improved UX for a user driven product outside of something that is a variation of a chat bot. Even the main AI players can't advertise anything more than, "have AI plan your vacation".

As others have stated too, how do you define what an incorrect output is?

Where? How do I know you're not pulling from some shady repository?

I don't know any student that really uses it now.

If you want actual good advice: go to a doctor.

Don't go to a chiropractor, don't go to hackernews. Go to a doctor. You can either start with a physical therapist in your area or start with your primary care doctor to get a referral.

I'm assuming you're in the US, so I know it's expensive but this will genuinely shorten your life span if you let it get significantly worse.

Just cause you're using an LLM doesn't mean you're "vibe coding".

I regularly use LLMs at work, but I don't "vibe-code", which is where you're just saying garbage to the model and blindly clicking accept on whatever is spit out from it.

I design, think about architecture, write out all of my thoughts, expected example inputs, expected example outputs, etc. I write out pretty extensive prompts that capture all of that, and then request for an improved prompt. I review that improved prompt to make sure it aligns with the requirements I've gathered.

I read the output like I'm doing a deep code review, and if I don't understand some code I make sure to figure it out before moving forward. I make sure that the change set is within the scope of the problem I'm trying to solve.

Excluding the pieces that augment the workflow, this is all the same stuff you would normally do. You're an engineer solving problems and that domain you do it in happens to involve software and computers.

Writing out code has always been a means to an end. The productivity gains if you actually give LLMs a shot and learn to use the tools are real. So yes, pretty soon it's going to become expected from most places that you use the tools. The same way you've been expected to use a specific language, framework, or any other tool that greatly improves productivity.

When I write my prompts, I literally write an essay. I lay constraints, design choices, examples, etc. If I already have a ticket that lays out the introduction, design considerations, acceptance criteria and other important information, then I'll include that as well. I then take the prompt I've written and I request for the model to improve the prompt. I'll also try to include the most important bits at the end since right now models seem to focus more on things referenced at the end of a prompt rather than at the beginning.

Once I do get output, I then review each piece of generated code as if I'm doing an in-depth code review.

I just tried on Chrome and Firefox on desktop, and the iOS YouTube app. Both show me a message saying "Recommendations are off. Your watch history is off, and we rely on watch history to tailor your Shorts feed. You can change your setting at any time, or try searching for Shorts instead."

I'll also clarify, sometimes if you click on a shorts video that you searched for manually, a few related videos will be queued, but then the feed will try up and the watch history message will display again.

Do you have left over watch history from years ago you've never cleared, and maybe shorts is enabled since it uses that...?