saretup

> Scalable Perceptual Generative Model

If you wanna call it Bagel, just call it Bagel. No need to make up a justification.

Then why bother to blur it?

From my experience, YouTube recommendations on Apple TV suck compared to the home feed you get when you open YT on iOS. I was pleasantly surprised when I opened YouTube on mobile after a long time, it actually felt like what YouTube used to be with a wide variety of new and interesting videos (plus added shorts and posts).

I think Google needs to look into optimizing it better for those of us who prefer watching YouTube on TV.

That’s a really cool idea. I often used https://dannymator.itch.io/randomicon to come up with novel ideas, never thought of feeding random words to llm as a way of doing it.

That’s interesting. I always intuitively assumed x-axis was progress and y-axis was cumulative effort.

> "they’re very good coders" and "they make trivial mistakes or hallucinate" seems incongruous.

It seems that way if you judge them the same way you would human coders, but they’re different. They might be able to do things that veteran coders can’t without spending days on it, and fail at things that beginners can do in half an hour.

A car might not be able to traverse difficult terrain as well as a horse, but it doesn’t mean the car is not a good mode of transportation.

You sound more prejudiced against AI than they seem to be hyping AI in that sentence.

Yes, they’re very good coders; no, they aren’t perfect. Sometimes they make trivial mistakes or hallucinate, and other times they have unique insights about difficult problems or one-shot a lot of complex tasks.

If you don’t see the value of AI even at this point, you’re either lacking imagination and/or just too stuck in your own ways.

What I found while using these models:

For generating a new image, GPT 4o image gen is the best.

For editing an existing image (while retaining parts of the original image) such as adding text or objects in the original image, Gemini 2.0 image gen model is the best (GPT 4o always changes the original image no matter what).

The whole article feels like justifying a bunch of legal nonsense to get to the end result of removing the capped structure.

Every new format or protocol gets used to display someone’s resume at least once (http://www.rleonardi.com/interactive-resume/).

Congrats on getting there for MCP resume before anyone else :)

Not to mention the cost users paid to download 250 MB every 5 minutes.

That would be a waste of a data point. Consider joining the colors in a circular shape.

That’s because those are not personalized. The economy of scale allows for artists to make generic coloring book with high quality art, but it’s expensive for artists to create (and customers to buy) custom made coloring books personalized for the customers photos.

Does it mention it’s 100x more efficient anywhere? Or is it just an example you’re providing, in which case, why not 1000x?

I’m assuming that’s the Speedometer score

Now let’s hear from some dead people for balance.

4o isn’t really comparable to deepseek r1. Use o3-mini-high or o1 if you wanna stay near the state of the art.

I misunderstood it to mean they are selling any YouTuber’s email address for $10k

Anyone can get a 0% false positive by always inferring negative. What you wanna look at is the precision-recall curve.

https://en.m.wikipedia.org/wiki/Precision_and_recall

That’s the one I used when I moved from Spotify to YouTube music. Luckily my playlists contained less than 500 songs so the whole migration was easy and free.

If it is a serious game, I’m surprised they don’t examine the chestnuts first

Partially inspired? That’s literally a diep.io clone. Not that that’s necessarily bad, sometimes it’s easier to clone an existing game you like when learning a new game engine/language. But you should own it.

Firstly, English is a highly subjective category.

Secondly, Llama 3 usually adds first sentences like ‘What a unique question!’ or ‘What an insightful thought’, which might make people like it more than the competition because of the pandering.

While Llama 3 is singular in terms of size to quality ratio, calling the 8B model close to GPT4 would be an overstretch.

That could be pretty handy. Any plans to add support for Firefox Multi-Account Containers?

This is really nice. I tried it out on the app and wanna start using it more. The only issue stopping me is the privacy policy.

I understand why you have to retain the voice data, however, is there any way you can implement an opt-out feature?

I’m just not comfortable with my voice data lingering in some servers, ready to be used for training ml models.

> Watching movies in Vision Pro is great at first but most people will stop doing it after the initial novelty excitement wears off Watching TV/movies in virtual reality seemed like such an incredibly compelling idea that we (the Oculus team at Meta/Facebook) built an entire product around that idea — Oculus Go. Launched in 2018, Oculus Go was the biggest product failure I’ve ever been associated with for the simple reason that it had extremely low retention despite strong partnerships with Netflix and YouTube. Most users who bought Oculus Go completely abandoned the headset after a few weeks.

You can't possibly extrapolate the movie watching retention rate from Oculus Go to Apple Vision Pro. That's like saying most people won't use ipod based on the data we collected from our walkman.

> It looks worse vs the original instant response

I think that’s an unpopular opinion. The instant response looks janky, exponential smoothing looks well polished as long as it’s not too slow.

Even earth gets acid reflux sometimes

Is there an online tool/piece of code that can find words like this in a string?