egeres

Very excited about this!

I wonder if these fusion techniques could help to run better local AI by streaming tokens from multiple machines and combining them

Sarah Cadigan-Fried has designed some very cool modern pixel art fonts worth checking! https://www.soft-type.com/

There's also a growing interest on integrating DFlash: https://github.com/ggml-org/llama.cpp/issues/21978, I can't wait to see how it will compare against MTP

Yes, it's active and even had a new release last week

Makes me wonder if in the far future it will be ever possible to do something crazy like 7nm processes at home

I wasn't able to pull some images and I lost 1h trying to diagnose network problems in my setup, but it didn't occur to me that "la liga" was the root cause . My workaround was to add "registry-mirrors": ["https://mirror.gcr.io"] in my /etc/docker/daemon.json

Thank you and your brother for all the amazing work, it's really inspiring to others <3

Maybe this is far fetched, but you could get an LLM-based auto-research system to extract these potential relationships

It also give you the possibility of filtering out which ones are worth cracking and which ones not

They have terrible support for banking apps and any app that needs play integrity

Related to this, I highly recommend anyone to install github.com/ActivityWatch/activitywatch, it's an amazing tool to keep track of your computer use completely locally. I think there are lots of possibilities with data analysis/AI aimed to improved one self's life

If it's an open source project the landing page should have a direct link to it so that developers can get a broader sense of what this product is

This Christmas I bought my mom a new computer because her old one (W10) was falling apart. It took some time but I managed to convince her to give ubuntu a try instead of moving to W11. After a week of complaints and stubbornness, she got really surprised about the lack of annoying prompts asking for updates, dark patterns to switch to edge, promotions etc. Now she has fully adapted and for her basic needs (browsing, reading pdfs, editing spreadsheets) she's basically set for life

It's shocking to read the headlines about the latest direction windows is taking and how user unfriendly is becoming

To me it's very clear that the icons have that "stable diffusion trying to make pixel art" style. I think this needs an extra layer of code that gets the generated image and turns it into actual pixel art

It stays around 26Gb at 512x512. I still haven't profiled the execution or looked much into the details of the architecture but I would assume it trades off memory for speed by creating caches for each inference step

Incredibly fast, on my 5090 with CUDA 13 (& the latest diffusers, xformers, transformers, etc...), 9 samplig steps and the "Tongyi-MAI/Z-Image-Turbo" model I get:

- 1.5s to generate an image at 512x512

- 3.5s to generate an image at 1024x1024

- 26.s to generate an image at 2048x2048

It uses almost all the 32Gb Gb of VRAM and GPU usage. I'm using the script from the HF post: https://huggingface.co/Tongyi-MAI/Z-Image-Turbo

Amazing to see software like this without sign-in requirements or paid subscriptions!

Super interesting project, at first I thought it would be a naive implementation of YOLO but I wasn't aware about retro-reflections. The papers he linked in the GH discuss very interesting ideas

Imagine this inside miegakure (https://miegakure.com/)

Unfortunately, that last 10% of games are AAA competitive multiplayer that account for a massive user base who are still dependent on windows to play them (battlefield 6, fortnite, any of the call of duty games from the last 8 years, league of legends, GTA online, apex legends, rainbow six siege...)

It's a completely different level, he has my favorite combo, incredibly detailed videos with fresh and complex engineering ideas

All three (GB10, GB200 and GB300) are part of the Blackwell family, which means they have Compute Capability >= 10.X. You could potentially develop kernels to optimize MoE inference (given the large available unified memory, 128Gb, it makes the most sense to me) with CUDA >= 12.9 then ship the fatbins to the "big boys". As many people have pointed out across the thread, the spark doesn't really has the best perf/$, it's rather a small portable platform for experimentation and development

I'm pretty sure they just want to be coherent with the which has that "steel scrubber finish" on the hardware

(photo for reference: https://www.wwt.com/api-new/attachments/5f033e355091b0008017...)

I wonder how this will affect the large cinema production companies (Disney, WB, Universal, Sony, Paramount, 20th century...). The global film market share was estimated to be 100B in 2023. If the production cost of high FX movies like Avengers Infinity War goes down from 300M$ to just 10K$ in a couple of years, will companies like Disney restrain themselves to just release a few epic movies per year? Or will we be flooded with tons of slop? If this kind of AI content keeps getting better, how will movies sustain our attention and feel 'special'? Will people not care if an actor is AI or real?

Recently, I found out a couple of friends were looking for jobs and I convinced them into creating a joint notion database for companies, job positions, networking events etc... So far I think it's been positive, it makes the job search less isolating and it certainly helps to see further away in the market

I suspect they took some inspiration from blender's changelogs, which are an astonishing example of how much better we could market open source software: https://www.blender.org/download/releases/4-5/

This article and james clear go along the idea that: behavioral change is preceded by an identity one

Personally I disagree. as I see it, the harder "something" is to achieve, the less true that statement is. For example, identifying with a particular political ideology requires less material effort to align your behavior than identifying as a genius or olympic athlete

On top of that, I have the subjective perception that the world keeps getting more and more complex and difficult to fully understand. There are way too many loops of information to be deep into

On android I'm able to extract some of this data by directly accessing the sqlite files in the databases directory of my rooted device, (with sudo-level permission obviously). I find that much more comfortable than having to use APIs with temporary tokens