Hacker News new | past | comments | ask | show | jobs | submit | ttoinou's comments login

But rendering engines have been optimized for years and this is a research paper. Probably this technique will also be optimized for years and provide a 10x speedup again

Sure, but algorithmic complexity beats linear factors, so unless they somehow manage to get from O(N²) to O(log N) for triangle count, this technique cannot ever even come close to established traditional approaches; no matter the linear improvement.

Nice. It required teachers that are good though. Maybe not anymore with AI

Amazing thank you. I thought about that years ago but since I never saw anyone do it I thought something must be wrong if nobody does it.

Are you… me ? I have 6 browsers for 6 different independent profiles. Instead of using browser profiles I need different browsers for different features (no google tracking, youtube, business services, web development etc.) I was wondering if I was the only one crazy

Would be practical to have a unified way to install extensions in all of them


Thanks for the suggestion! I'll think how it can be added in a sane manner

Cool. Do you also use tons of browsers ? How many total tabs do you usally have ? I can easily have 300 tabs, including 30 useful

Lol I don't think I ever had 300 tabs open. Must be a nightmare to navigate haha. I believe I also used to have 6 browsers, but with Donut it now fluctuates at around 10 profiles, depending on what I need to do.

In terms of tabs, I rarely have over 30 tabs open. The moment I feel like I'm getting overwhelmed with the number of open things, I close all tabs that I don't absolutely need.


I don't think AI LLM are very good at Rust, no ? I tried multiple times with Sonnet 3.5 last year to produce web interface with Tauri and it got in infinite loops with async functions. On the other hand setting up shaders in Rust worked out of the box. But I couldn't ask too much before the AI was looping over bad code over and over.

Do this in C++ and I'm pretty sure you won't have any issue


I've had excellent experience with several models writing Rust. Wonder if there's just a particular issue with Tauri? I'm primarily writing code on top of the Candle ML framework.

Any learned tricks to getting good performance out of LLMs with Rust in particular that would differ from using LLMs to generate Python code?

Which models are the best for Rust? How are you finding Gemini 2.5 Pro?


Your comment is def why we come to HN :)

Thanks for the detailed info


Thought the SAME thing

HN comments vote need more than 40 minutes to stabilize

Maybe even if it was a possibility before, it wasn't used and now airlines have enough data and market power to actually make different prices for groups

(By the way, if it's about inflating prices for individual, then it's not really volume discounting... it just appears this way on the outside)


Huh if this becomes mainstream there's an opportunity to make a social media website to purchase in groups and make friends for the flights

Good idea in principle. In practice this could invite unscrupulous actors, or people who flake out at higher rates than close family -not that families can’t flake out, but I’d imagine it’d be a lower incidence.

There are already websites, like Going, for getting flight deals. As a solo traveler who doesn’t have to coordinate with anyone and can pack light, I can jump on the deals when they come up and save a lot more than what a regular price group rate probably is. Looking at my upcoming trip, I got it for 50% off the current pricing, for solo or a couple.

Coordination with others also makes booking take longer and tends to fix dates and locations, which makes it hard to grab a deal when they come up.


I "used" Scott Cheap Flights in 2018 for a few years and never found a good deal

You have to invert the order of how most people plan travel.

Typically people plan in this order:

1. Where to go

2. When to go

3. Check airfare

Flip it on its head:

1. Keep an eye out for cheap flights from your home airport

2. Pick one of those destinations

3. Choose when to go

Not exactly inverted, but the flight goes from last to first.

I had Ireland in my head for my next trip, but then Italy showed up last week. That sounded pretty good, so I checked the dates and found a flight that fit in my schedule, and booked it.

It’s not great for getting a flight around a conference, wedding, or some other event you are planning around. But when you know you want to take a vacation to somewhere and sometime this year, it can cut flight costs in half.

A single flight ends up paying for the cost of the subscription 10x over, and then some.

Also, it could be somewhere near where you want to go. I had tickets to Croatia about 5 years ago, but hand to cancel due to the pandemic. I didn’t know anything about Croatia when I booked, but I figured the worst case scenario was I catch a quick ferry or flight over to Italy once I’m there. Once you’re over there it’s cheap to go one country over. That flight to Croatia was $289, that cheaper than a flight to Nashville for me, which is only about an 8 hour drive.

I did the same thing when I booked a flight to Sweden. I didn’t know if I’d like it there, so from there I booked a connection to Copenhagen for next to nothing. I spent a week in Copenhagen, but then ultimately did go back to Sweden for week, which I ended up loving. I’m glad I didn’t spend the whole trip in Copenhagen.

Another deal I got that stood out was to Tokyo. I think I paid around $550 give or take. A coworker of mine has family there and goes on a regular basis, he was floored by that price. He always pays over $1k, and usually closer to $1,500.

Ultimately it’s just an alert service for flights that are abnormally low. If you have a specific destination in mind, Google Flights is pretty good at showing when the cheapest time to travel is, giving a booking date of today. Of course it fluctuates over time, which is where the alerts come in.

It’s saved me thousands. Though I probably would have only taken about 20% of the trips I have without it.


Thanks yeah that was my strategy I'm flexible with travels, I remember setting Paris and Lisbon after to see what flights I could have. Maybe it was focused on USA. Maybe their emails was getting my SPAM folder, IDK what happened but I never found what I was looking for.

I got my own flight tickets by looking on my own though.

550 USD for Tokyo is so cool, wow


How knowledgable do you need to be to tweak and train this locally ?

I spent two days trying to train a LoRa customization on top of Flux 1 dev on Windows with my RTX 4090 but can’t make it work and I don’t know how deep into this topic and python library I need to study. Are there scripts kiddies in this game or only experts ?


Just use https://github.com/bghira/SimpleTuner

I was able to run this script to train a Lora myself without spending any time learning the underlying python libraries.


Well thank you I will test that

SimpleTuner is dependant on Microsoft's DeepSpeed which doesnt work on Windows :)

So you probably better off using Ai-ToolKit https://github.com/ostris/ai-toolkit


OneTrainer would be another “easy” option.

The open-source model is not released yet, but it definitely won't be any easier than training a LoRA on Flux 1 Dev.

Damn, I’m just too lazy to learn skills that will be outdated in 6 months

[flagged]


I'm selectively lazy for sure. I'm working full time right now. Like, all the time except for sleep

It sounds you’re not lazy enough, that’s not healthy longer term.

It's normally easy to find it ore configured through comfyui.

Sometimes behind patreon if some YouTuber


> I spent two days trying to train a LoRa customization on top of Flux 1 dev on Windows with my RTX 4090 but can’t make

Windows is mostly the issue, to really take advantage, you will need linux.


Even using WSL2 with Ubuntu isn't good enough ?

Nah, that’s fine. So is Windows for most tools.

The main thing is having 1. Good images with adequate captions and 2. Knowing what settings to use.

Number 2 is much harder because there’s a lot of bad information out there and the people who train a ton of Loras aren’t usually keen to share. Still, the various programs usually have some defaults that should be acceptable.


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

Search: