I noticed my fan got quiet when I closed Chrome, so I measured its RAM and CPU usage 250 times per second while opening 54 tabs, then did the same in Safari
More charts here:
I made an app that can turn itself into ANY app, just by giving it a new name. It's kind of like phone apps for desktops; compact, light-weight, unbelievably fast. - thread
ok, so I added cropping to the Upscalo app, and now I'm zooming in on the fabric fibers of this tiny 488px Nike shoe photo by upscaling the upscaled image over and over. It feels like a dreamy nano microscope
Some folks got FURIOUS that the toggle in the previous tweet might not match the actual weather outside. So anyway, here's what might be...the world's first live streaming toggle
Watching AI "type" word by word gets kind of boring fast, and it may set the wrong expectations as in "these are facts." What else could it look like?
1. The AI choosing among millions of tokens in a blur, then sharpens as it picks a random token
Just like that, I now have an app I'm pretty sure I'm going to use again tomorrow.
I didn't consult the API docs once. I didn't even pick the API in the first place. I did help gpt4 out a few times.
Like a good coworker.
Here's the entire app
@itsPaulAi
Let's play it out
Company A fires half their staff yet keeps delivering 100% of their old capacity = they save money
Company B keeps all their staff, but can now deliver 200% of their old capacity = they win the competition
So seems like the strategic move here is to raise the
OK, let's put this AI-generated owl under a 6,400% AI microscope? The hardest part is handling the huge image in Upscalo's UI. 100x100px 100kb -> 8,000x8,000 80Mb. Should I ship it or too big?
@OmniFaces
You may need a code-specific intro
You are an AI programming assistant.
- Follow the user's requirements carefully & to the letter.
- First think step-by-step — describe your plan for what to build in pseudocode, written out in great detail
- Then output the code in a single
I made a little macOS upscale app with CoreML today. Surprised how fast it is! Do apps like this exist? (Check out the wet sponge look on that BigMac's bread)
Well, I actually needed both upscaling and background removal, so I put that in there too. It's still so fast it almost feels wrong. Anyway, let me know if you want to try it out
5 attempts at making 10 seconds pass faster using
@framer
motion
1. We're creating a PDF. When you print something, the paper is a real-world physical progress animation
Made a quick Whisper prototype a while back and forgot how fast it is:
62x realtime, typing 8,000+ words per minute....and
@ggerganov
has a branch that is at least THREE times faster than that. Completely offline
I made a podcast app. It’s about episodes, not shows. Self-googling episodes. Bookmarking. Mixing with your own music. Take a look in this thread, and here
Wait, there's more to this. It seems psutil may not be able to see all of Safari's child processes, like each renderer. Here's a quick dump of ALL my mac's thermometers (a mac packs a ton of them). The peak is 54 open tabs, then quit, cool off. X is minute of the hour, Y is deg C
Language models that code are indeed "infinite app stores"
Now we just need infinite product managers to give us the apps we _really_ need. Wait, could it .. ? 👇
LLMs are infinity app stores in a box.
With time you will not need to download others apps, you will just build your own highly customized app for your specific purpose simply by describing your problem or task.
The apps and GUI will be created as a byproduct.
#AI
There's a few questions about the process. Here's an unedited screencast from start to end building a simple todo app in ~8 minutes. Note the starting prompt.
Full:
Hundreds of real humans helped train this neural network to be able to predict where users would most likely look at a design. It's like running thousands of real sessions ... per second. You can try it out - let me know what you think
I'm pulling frames from the webcam through a warp kernel on the GPU, then desaturate, blend, opacity etc. No idea how the kernel stuff works, just stole it from
@FlexMonkey
's book
Getting closer (and literally too close to the top edge)
Iterations on the path takes dozens of minutes, not seconds. I'm sure there's a smarter way? Here's how I've done it so far
I made a Google Sheets add-on that takes a bank transaction → Finds it in Gmail → Makes a PDF → Saves the PDF in Drive → Adds a link, and lets you drag it to Finder.
Not knowing if the data was at all available, I asked it to show available streaming services.
What was cool was the way it told me to update the code: instead of giving me the entire app over and over, it chose sections and even commented unrelated code out
Really, why is it so hard and even unsafe to convert 3D models?
@fursund
had an idea: Let's just do it in the browser? No uploads, no server, 50+ formats. So we did, with Wasm and nextjs (and made it free, no ads). It's on Product Hunt right now, link below.