Justina Petraitytė

Excited to finally share my chat with Uri Goren from argmax. Uri and I have been friends for a very long time, and he's a super smart guy, I think it shows that we're having a lot of fun in the recording. We're talking about old things and new! Old is 🍪 cookies going away for all of us, to get more privacy while browsing the internet 🛜. But for advertisers, that's a challenge since everyone still wants personalized ads, and there is much less information to use without cookies. Uri and I talk about how ML might be able to get the best of both world - more private, and more personal ads - how ads will look like in the "cookie-less" era. We dive into the challenges and emerging solutions like 🎲 probabilistic models and contextual features. Check out the episode for insights on balancing ad personalization with privacy! BTW, this episode is shorter than my usual ones, so if you like it, please let me know. Experimenting with different formats #DigitalAdvertising #MachineLearning #Privacy https://lnkd.in/dY9aDyqj

20

Joanna Maciejewska (@AuthorJMac) highlights a key point about AI: let's let technology handle chores so we can focus on creativity! 🎨✨ "You know what the biggest problem with pushing all-things-AI is? Wrong direction. I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes." Check out the original tweet here: [Joanna Maciejewska on X](https://x.com/AuthorJMac) #AI #Creativity #Innovation #TechTalk #WorkLifeBalance

7

The feedback about my recent Gen AI workshop.

14

What makes all the diffusion based voice conversion models from research papers ridiculously slow to train is that almost none of them use latent space, they all use a continuous time model (you train random time steps from 0 to 1), and you have to add random noise to the inputs or else you won’t have enough data per speaker to generalize. Also, some of them train the encoder and diffusion model together. Not much you can do about adding random noise to the inputs, but you can use latent space, use a discreet time diffusion model (only train on 1000 time steps), and always train the encoder by itself first, then the diffusion model by itself, and then a bit of both together. Ridiculous * X training speed up.

1

That's awesome! Mistral's new features for building custom LLMs sound incredibly developer-friendly. The ability to fine-tune and create custom agents makes it a powerful tool for personalized AI solutions. Thanks for sharing the video link—can't wait to check it out! 🚀🔍 #CustomLLMs #AIInnovation #DeveloperTools #MachineLearning #MistralLLM

1

WE ARE GETTING TO … The development of the #AI #foundational #LLM models is converging to state where you can run seriously intelligent model on your mobile device. That is going to hit … … mainly #Google and then mobile operators. Imagine the amount of traffic that can be saved, if you had your all-mighty advisor off-line.

2

I just published a blog on how to fine tune LLMs and create custom datasets. In this blog, I cover: - What fine-tuning is and its benefits - When to fine-tune a model - Methods for fine-tuning, including full fine-tuning and parameter-efficient fine-tuning - How to prepare your custom dataset Check it out here: https://lnkd.in/gDcrrBGP

15

1 comentário

Exploring the new PaliGemma model was enlightening! Check out this versatile notebook for fine-tuning vision-language models:(https://lnkd.in/evH9YckT) Let's brainstorm more use cases and expand our AI toolkit! #AI #LLM #Visionmodels

3

2 comentários