I have solved this problem by forcing the model to compress every time by specifying token budgets for the documents

You mean word budget? I don't see how a token budget on the LLM helps with doc length. Or do we denominate the length of text docs in tokens now?

Can confirm that. Major scars and my first broken bone on my third time taking the bike. And I liked it a lot when I was younger.Never had the courage to buy myself a motorcycle but realized that I was going way faster than even my in my car with my bike. Love the thing

Trying to make AI do what you want just with instructions didn't work for me. In a new system I am designing, I've implemented a set of algorithmic checks to make sure that every to-do affects only the area it needs to and does not create deltas which are unintended. I developed a system of fingerprinting pieces of code to not do the checks every time as well for performance reasons. It's working great so far and I'm building more and more layers of algorithmic checks and the model is beginning to work extremely well with it. I am now able to fill out a limited to-do list and be absolutely certain that only the pieces that I wanted were changed and nothing was broken. Maybe it's nothing new for most of you, but it added so much more confidence in my results in this new system, which will be a hundred percent AI generated. I was very tired with the old way of doing this by relying only on a set of loops and instructions. In it was never giving me this feeling of confidence that I have in this new system I'm developing. I could write about it if there is interest

Do you have any more details about this or maybe a repo? I feel like I'm running into the same issues, especially with agents wanting to run insane amounts of test s and check and measuring for every little change.

I would love to see how you’ve implemented this if you can share a repo.

Not fully relevant to the comment, but I've noticed that most motion sickness issues derive from an FOV which is too low.Big companies set it lower for performance reasons I guess, but on PC I always loved to crank it. And it's not about sitting closer or about the distance in itself, it's about the ratio of the size of the screen to the vision field it occupies

For me it was more about the mismatch between what my eyes saw and what my inner ear felt.

Running around in first person while sitting or standing in a room took some real getting used to. But cockpit games or mostly stationary (in a reasonably small volume) standing-perspective games were fine.

Nice font, it reminds me the kind of fonts they are using when showing code in movies

I don't get it. Is this guy admitting of doing piracy? Is that like a thing you do?

Media distribution companies are like health insurance companies (though not to the same degree) in that they can't stop themselves from squeezing their customers to the absolute limit. Majority opinion and the definition of ethical consumer behavior can change very quickly under those conditions.

Reply to everyone, I wasn't criticizing the author, I was just trying to understand what's happening. I am an avid collector of Linus ISOs myself

You're posting on a site called "hacker news" and you draw the line at cartoon piracy?

It's a great way to watch whatever you feel like watching on whatever device, frankly I don't even know what are the alternatives.

If you continue to clutch your pearls like that, you'll break the strand.

who cares?

Looks nice but the presentation of the website is terrible.Looks like they are try-harding.Not happy with the copy at all.But the lines of of the car are nice to see.Please Porsche keep it together.

> Looks nice but the presentation of the website is terrible.Looks like ... all.But the ... see.Please ... together.

Please add spaces after periods. The presentation is terrible.

My recent strategy to design presentations has been to record meetings, summarize them and dump it all in onn agent. It can create something that is pretty much perfect very quickly. It's all about context and giving to agents as much of it. Of course you need to prompt them a lot to focus on specific areas you want, but the extra raw context really does help.

I am having great results by mapping the press to dictate onto my pedal.I have been vibecoding without using my hands for the past four months with great results.It's amazing using the computer without touching it.I can go on for hours with a couple of pedals.I use them to switch windows, dictate and press enter.That's really all you need.

What kind of pedals?

In my own vibe coded implementation during the recordings the live transcript gets sent for validation and allows me to take better decision while in the call itself.I'm using a a small local model for processing.Also like when it detects bullshit told by others in real times so I can immediately reply.Fun