
CEO and Founder of @abstract3d
Stuttgart, Germany
Joined July 2015
- Tweets8.8K
- Following93
- Followers10.6K
- Likes105K
Quite the coincidence that your new "BiteTheBytes" website looks like an Abstract3D.com clone. From the header layout/icons/lang sel -- down to our little block shooter. lol 🙄
Our design is non-standard/hand-rolled so the clanker copy is obvious.
I get it, you need inspiration, but you have to look outside of your own field.. or at least, outside your own country.
"yet"
here is what "AAA" looked like in ... 2013. Fable is cool and all but it's not producing visuals or gameplay anywhere close to this yet.
youtube.com/watch?v=mpAYoMno…
168
When done right, a 4-bit quant introduces little perplexity, esp. for large models ~2%
Using 4-bit for activations and partials also has a very small impact, still ~2%
There's a thin line between model collapse and those results. Needs ultra-accurate engineering.
Can I ask a dumb question?
Is a 4-bit quantized 70B model actually smarter than a full-precision 8B model?
3
218
This is the most unhinged code I've seen in many years. "Open-source"? This might as well be binary.
tbh looks like disassembly using Ghidra or so.
We're living in truly remarkable times.
The ironic part is that this slop will ruin datasets for future AI models. Careful curation of datasets will be the moat of the future.
mattiaslind.gumroad.com/l/mp…🔶Introducing mPainter(Alpha)!
An open source texturing application with blender at its core, plus a bunch of custom goodies like a new brush-engine etc.
Work with layers, and import the blender materials you already have in a familiar texturing interface😊#b3d #blender #texture #app
That tweet blew up so I'm following up. Below you can see GI running on M4 Pro at a 3000x1500 res. Super clean + long distance!
Really happy with the perf & quality, as the constraints and challenges to be solved are endless. Also, handles light leaking quite well.
Have been working on a novel GI pipeline for RSX the whole year. It's rather clean and converges almost instantly. It's been an insanely ambitious project as this 1. runs entirely without bindless 2. no hardware RT 3. must generalize to gigantic viewing distances 4. must run on WebGPU
GI vs NO-GI below
Have been working on a novel GI pipeline for RSX the whole year. It's rather clean and converges almost instantly. It's been an insanely ambitious project as this 1. runs entirely without bindless 2. no hardware RT 3. must generalize to gigantic viewing distances 4. must run on WebGPU
GI vs NO-GI below
Check out the follow up with a video here: teapawt.net/mannyn83/status/211737…
That tweet blew up so I'm following up. Below you can see GI running on M4 Pro at a 3000x1500 res. Super clean + long distance!
Really happy with the perf & quality, as the constraints and challenges to be solved are endless. Also, handles light leaking quite well.
2
283
@_anderberg you can check out the cascade transitions here!
2
1
246
He's an actor -- don't forget that.
Ben Affleck reveals he writes Python, understands convolutional neural networks, worked extensively with GPUs, and used his celebrity status to get private looks at Google and OpenAI’s video models
“I’ve always been kind of into computers since I was young. Then, when film started to move from analog film to digital, I became more interested in that aspect of it. The visual-effects workflow for many years has included machine learning, so I can write pretty shitty Python scripts and stuff like that.
“With convolutional neural networks, which were the precursors to what the transformer can do, which is much more computation simultaneously, you would do things like look at what’s called a tensor. That’s the numerical translation of a visual image in numbers, like the batch number, the frame number and the red, green and blue values of each pixel in each frame. It’s just that simple. That numeric is called a tensor.
“You’d use a convolutional neural network to identify patterns that reveal what’s called edge detection or feature extraction, which is identifying patterns well enough to know, this is where the window ledge is, so we can more easily take the green-screen image out and replace it with something.
“That was familiar to me early on because, prior to Artists Equity, I had a small visual-effects company. I’ve worked with GPUs a lot too. The visual-effects guys said, ‘Hey, you should see. There are a couple: Google and this other company, OpenAI, are doing really interesting stuff with transformers in video.’
“I’ve learned that I can actually just call up and go, ‘Hey, it’s Ben Affleck. Can I come see what you’re doing?’ Sometimes people say yes, to my astonishment.”
324
Opus 5.5 is an incredible leap forward, usage limits are perfect for daily work.
A few weeks back, Opus 5 was slow and incompetent and using Codex with a x20 subscription felt unbounded when it comes to usage. Astra was fast and competent. Now Astra is slow and burns the whole plan in a single day.
1
287
This was like therapy for a whole generation. Now everybody has to hold back and look what's going on in the world.
Replying to @Pirat_Nation
Remember the good old Xbox 360 days in MW or Halo lobbies where everyone said anything and no one complained? good times.
people had fun and talked shit and then did it again the next day or week.
327
Right, Codex 20x sub -- 5hr usage is ~30% of your usage. Pretty odd, given that Codex used to give significantly more usage than Claude.
$20 Claude pro sub, running opus 5.5 high for almost 2 hours nonstop to use up a 5hr limit, and each 5hr limit is about 15% of weekly usage
This is way more than what you get with $20 ChatGPT plus, and opus 5.5 is so much better than Sol 6
Anthropic wins so hard here
3
338
Opus 5.5 sub usage is excellent, a 20x sub will likely match a work week.
Astra is unusable lately, a few hours of work and 40% usage is gone.
181
Would be useful if Codex would show the number of credits used in the task, when the task consumed credits not sub usage. Eg
Worked for 1h 10m, consumed 310 credits
@thsottiaux
1
255























