Varvara Novozhilova’s Post

This is eye-opening. We are getting closer to a world where nothing can be trusted online: I recently came across DeepFacelive. It's a tool that can perform real-time face-swapping during live broadcasts and video calls. • Video calls: Change your face in real-time during a video call or use a pre-recorded video • Live-streaming: Adjust your facial expressions or entire appearance • Facial animator: Control static facial images and manipulate them through your own face We've seen another example pop up in April 2024 with Microsoft's hyper-realistic talking head AI: VASA-1. It can generate lifelike talking faces of virtual characters all from a single static image and audio clip. My takeaways: • Proof of authenticity will be an absolute must to solve what is coming • I believe real-world interaction is more important than ever • Zoom meetings are about to get a whole lot more interesting If you enjoy insights like this, follow me Alex Banks for more on AI.

Google has published a 45-page guide to prompt-engineering, generally sharpened for dozens of different use-cases in different roles. The basic recommendations are pretty basic, but still very valuable. Things to consider in a prompt: - Persona: Who you are - Task: What you need the LLM to do (write, summarize, change tone, etc.) - Context: As much context as possible. This could be a written description or a link to an existing document in Google Drive. - Format: Paragraphs, talking points, specifying character limits, etc. More tips from Google: - Break the assignment into parts. - Specify limitations. - Assign a role. - Ask for feedback. - Consider tone. - Say it in a different way. Examples from different roles (you can find many more in the PDF): Customer Service Draft templates for three different types of customer communication. Create a template for apology emails, order confirmation messages, and thank you notes for loyal customers. Keep each template to one paragraph and use a friendly tone. Executives I’m the COO of a mid-sized company. I am hosting a quarterly town hall meeting with the entire company. There has been a lot of positive progress and momentum this quarter, but I am worried that multiple teams have low morale due to some unforeseen setbacks. I want to brainstorm and practice how I will respond to potentially tough questions. Help me write challenging questions that employees may ask at the upcoming town hall about our earnings, leadership changes, and vision of the company. Generate potential answers for each question that use a confident but firm tone. HR Write a list of 20 open-ended interview questions for [open role URL] that I can use to screen candidates. Marketing Create compelling copy for a landing page promoting a new [ebook/webinar/free trial and details] designed for an executive target audience. Highlight key benefits and encourage conversions with persuasive calls to action. Project Management Summarize this call transcript in a short paragraph. In bullet points, highlight the action items, decisions made, and owners for each item based off of [call transcript]. Sales Draft a customized script for me to follow during my sales call with a prospect. The call will happen over a video call and is set to last 30 minutes. Make sure to add the following in the script: How [insert company products/solutions] can help address potential customer’s pain points, how [company]’s delivery system guarantees seamless and timely delivery, competitive pricing and volume-discount table, and space for a customer reference in the [customer’s industry] industry. I highly suggest checking if your usecases are in there and trying them - https://lnkd.in/ddj4rJRq

Last week we launched a new project called Repeater 🚀 This is an application for Russian-speaking people who want to learn English and remember new words without pain. In the first version it's just a bot in Telegram, but we plan to create mobile apps. How does it work? You just text the bot a new word you came across in a book, article, movie, etc. and it will reply with a simple sentence containing that word. Every two hours the bot will send you different sentences. It needs to create a context for new word because you need to encounter new word several times before your brain remembers it. You can also look up the meaning of this word in English. Or you can translate it to Russian directly from the bot. I'll be happy if you try our bot. It's free. The next version will include other types of content like videos with words and pictures. Have fun! https://repeater.one ⚡️

Machine Learning jargon buster: 5 machine learning concepts explained in emojis 1. Supervised Learning: 👩🏫📚 → 🤖🎯 A trainer (👩🏫) provides a model (🤖) with labeled examples (📚), so it can make predictions or classifications (🎯) on new, unseen examples. 2. Unsupervised Learning: 🤖👀 → 🍎🍌🍎🍌 → 🤖🤔 → (🍎🍎), (🍌🍌) A model (🤖) explores items (🍎🍌) without labels. It observes (👀) the item's features. On its own (🤔), it finds similar items (🍎🍎), (🍌🍌). 3. Deep Learning: 🤖(🧠🔗🧠)👀 → 🖼️📚🎶 → 🤖(🎨🗣️🎵) A model (🤖) has a digital brain (🧠) composed of interconnected nodes (🔗), resembling the neurons of the human brain. These nodes are structured into multiple layers (🧠🔗🧠) and collaborate to process information. This multi-layered digital brain (🧠🔗🧠) learns from images (🖼️), text (📚), and audio (🎶). It finds (👀) patterns in data and performs tasks like image recognition (🤖🎨), speech recognition (🤖🗣️), and music generation (🤖🎵). 4. Reinforcement learning: 🤖🎮 → 🚶♂️🕹️ → 🎯❓ → (✅🍬)(❌🚫) → 🤖🎯🏆 Robot (🤖) plays a game (🎮). The robot tries different paths and actions (🚶♂️🕹️) and receives feedback (🎯❓) in the form of rewards (✅🍬) or penalties (❌🚫). Through trial and error, the robot learns which actions lead to the best outcomes, ultimately achieving success (🎯🏆) in the game. 5. Transfer Learning: 🤖📚🐱 → 🤖🎓🐶 A model (🤖) shares knowledge (📚) from one domain (🐱) to another (🐶), using pre-existing expertise (🎓) to tackle new tasks. Add some yours? :)