Daniil Koltsov @dokoltsov

Seasoned Golang & Cloud Engineer | Exploring AI & Intelligent Apps | Crafting scalable solutions with a pinch of machine learning #GoLang #CloudComputing #AI Berlin Joined June 2021

Tweets

222
Followers

37
Following

451
Likes

88

Daniil Koltsov @dokoltsov

2 years ago

@zackhargett Don’t like this comment =)

0 0 0 5 0

View Details

Daniil Koltsov @dokoltsov

3 years ago

@mitchellh What's your shell? and why?

0 0 1 143 0

View Details

Daniil Koltsov @dokoltsov

3 years ago

Check out a beautiful explanation of decision trees in machine learning. Just read part 1 and part 2 because they are so beautiful. r2d3.us

0 0 0 64 0

View Details

Daniil Koltsov @dokoltsov

3 years ago

@ylecun @elonmusk 😁😁😁

0 0 0 278 0

View Details

Daniil Koltsov @dokoltsov

3 years ago

@KevinNaughtonJr As their desktops - yes, 100% Otherwise we all use linux, pretty much anyone who has internet. It is a backbone of any complex software system.

2 0 2 804 0

View Details

Daniil Koltsov @dokoltsov

3 years ago

Alan Turing invented the Turing Machine, laying the groundwork for modern computing. Without him, we might not have today's computers. #AlanTuring #100GreatestProgrammers

0 0 0 64 0

View Details

Daniil Koltsov @dokoltsov

3 years ago

@voooooogel Haha, it was once called ascii art : ) Thus you will follow only millennials =)

0 0 1 46 0

View Details

Daniil Koltsov @dokoltsov

3 years ago

@RogerGrosse @aidangomez Hehe 😜 true! Just googled wikipedia first edit - it was done on 15 January 2001.

0 0 0 269 0

View Details

Alex Friedman 🤠 @heyalexfriedman

3 years ago

Life’s too short to work on something you don’t believe in. Don't settle.

13 24 214 21K 21

View Details

HOW DIFFUSION MODELS CREATE IMAGES: HIGH-LEVEL OVERVIEW. V2 Diffusion models are tools (like Midjourney) used to generate images from random noise. Here's a simple explanation of how they work: 1. Stochastic Differential Equations: At the core of diffusion models are stochastic differential equations. These equations describe how to gradually add noise to an image, transforming it into a random pattern over time. This process is akin to blurring an image more and more until it becomes unrecognizable. 2. Reverse Process: The true magic happens in the reverse of this process. Using another set of SDEs, the model learns to reconstruct the original image from the noise. This is where the training data plays a vital role, as the model has learned the statistical properties of images and how to gradually form coherent structures from randomness. The training doesn't involve remembering specific images but understanding general patterns and features. 3. Navigating the Image Space: Diffusion models work in a complex, high-dimensional space known as latent space. Here, every point can be thought of as a potential image. The model starts with a random point (pure noise) and then uses mathematical rules to refine this point into a detailed image that makes sense. The model doesn't store any images. ‘Navigating the image space’ refers to a conceptual and mathematical process. 4. Interpreting Text Prompts Mathematically: When the model receives a text prompt, it converts the text into mathematical instructions. These instructions are then used to guide the image creation process in the latent space, ensuring the final image matches the description. 5. Generating Unique Images: The randomness of the starting noise means the images produced are different each time. The model makes calculated predictions at every step, influencing how the image evolves. It's similar to using a different set of building blocks and a new plan for each construction project. I hope this was helpful and you now understand diffusion models a bit more.

24 82 571 55K 486

View Details

Daniil Koltsov @dokoltsov

3 years ago

@burny_tech Bro, this thing is crazy as fuck! So creative! I have spent 15 minutes studying and reflexing on those characters 🤯🤓😍

0 0 4 268 0

View Details

Daniil Koltsov @dokoltsov

3 years ago

@bryan_johnson Damn man you are healthy and cool 😎 I wanna be 80% like you!) Doing 20% of what you do!) Paretto stuff you know)

0 0 2 1K 0

View Details

Daniil Koltsov @dokoltsov

3 years ago

@svpino Good prediction!) Vibes and optimism 🥳🥰

0 0 1 385 0

View Details

Jessica @schmafia

3 years ago

tying your value as a person to your work identity is a sure fire way to misery

3 2 19 2K 2

View Details

Nikhil Naik @nikhil_ai

3 years ago

Checkpoints for Diffusion-DPO trained SDXL and SD-1.5 are now available from @meihuadang and @StefanoErmon at the @StanfordAILab! SDXL-DPO obtains 70% win rate over SDXL and is smaller! (1/2) SDXL-DPO: huggingface.co/mhdang/dpo-sdx… SD1.5-DPO: huggingface.co/mhdang/dpo-sd1…

6 34 150 20K 77

View Details

Daniil Koltsov @dokoltsov

3 years ago

There is an artist on twitter, the one who created many of the concepts for marvel. It is a crucial question for him. Can’t even imagine how it is to be a world class artist invested 10k hours and being plagiarised by a schoolboy.

Yann LeCun @ylecun

3 years ago

Genuine question about image generation: If someone uses a generative AI tool to produce an image that is substantially similar to a copyrighted piece (a drawing, painting, movie screenshot, etc), who should be liable for copyright infringement? Should it be: A. the company

391 232 2K 646K 644

0 0 0 39 0

View Details

Daniil Koltsov @dokoltsov

3 years ago

@rowancheung March 30: Elon Musk and AI experts call for a 6 month pause developing AI systems more powerful than GPT-4. April 17: Elon Musk starts xAI. Elon Musk: “now everyone stop developing, we need more time”

0 0 4 516 0

View Details

Daniil Koltsov @dokoltsov

3 years ago

@natolambert Thanks for sharing, Nathan!) This is a cool and useful tool!)

0 0 0 73 0

View Details

Daniil Koltsov @dokoltsov

3 years ago

Cool! Nathan releases a whole pipeline of creating voiceovers for podcasts and youtube vids for free. I heard fireship.io does this as well. Thanks 🤩 for sharing Nathan!🥳

Nathan Lambert @natolambert

3 years ago

Merry XMas - I'm releasing the code needed to generate the AI audio + image-videos I use to narrate my blogs and distribute as podcasts/youtube. Having your own voice is much better than using the meh, default Substack voice. I'll explain more about how to use it sometime early

3 6 21 4K 21

0 0 0 24 0

View Details

Daniil Koltsov @dokoltsov

3 years ago

Python, a language that has surged in popularity due to its simplicity and power, was first released by Guido van Rossum in 1991. It's now a cornerstone in fields from web development to AI. #Python #GuidoVanRossum