Gemini vs GPT4

Happy Friday, everyone. Folks already started walking around in public with Apple Vision Pro. Hope this won’t turn to walking dead series 😁. Google announced something big last week.

Here are the 4 interesting AI things that I learned and enjoyed this week.

AI Research

Microsoft AI has introduced LLMLingua, a new model designed to cut down on the number of tokens needed without losing the prompt's meaning. Advancement in GenAI technology led to the use of longer prompts, resulting in more computing power. LLMLingua tackles this by using a special compression algorithm and instruction tuning to keep responses on point without the heavy resource drain. More on compression algorithm in today’s AI concept segment. If ChatGPT needed to process a detailed story summary, LLMLingua could shrink the prompt's size without losing key details, enabling quicker and more cost-effective responses.

AI Meme
I know I’m late but I can’t skip talking about this. Google's Gemini is stepping into the ring with GPT-4, ready to mix it up with its sidekick Bard. It's got three versions: Ultra for the heavy lifting, Pro for everyday tasks, and Nano for on-the-go. Gemini's not just about words; it can handle pictures and sounds too, aiming to outshine GPT-4 with versatility. It's promising big things like homework help and creating art, but it's still proving itself. We all know how the last fake demo went 🤣 Google might be hyping Gemini up, but time will tell how it really stacks up.

AI Concept

In our last edition we talked about perceptrons. This week we will be exploring compression algorithm. LLMLingua makes LLMs like ChatGPT work faster by making the prompts shorter. It does this through a compression algorithm, which is like summarizing a long story into a few key points without losing what's important. Then, instruction tuning adjusts how the LLM answers, making sure it still gives useful replies based on the shorter question. For example, if we have a prompt like “What are the global impacts of climate change on weather patterns, sea levels, and biodiversity?” It compresses it into something more concise like “Summarize climate change impacts.“ This shorter prompt still covers the key aspects such as weather, sea levels, and biodiversity, allowing the AI to provide a comprehensive response without needing the longer explanation.

AI Hacks

To develop web app in 5 minutes using GPT-Engineer. Let's walk through building a fitness tracking app with GPT-Engineer, simplified into clear steps.

1. Sign Up: Go to the GPT-Engineer site, enter your email, hit "Get Access," and fill out any extra info to boost your chances of getting in. For example, you're aiming to create a fitness tracking app.

2. Give Your Command: Once you're in, head to the prompt section and say, "Create a fitness tracking web app." GPT-Engineer will then cook up the basic code you need to start.

3. Generate Code: Hit the "Create" button to get the code rolling. You'll see your fitness app's code pop up in an editor, where you can check out how it's structured and what it does.

4. Refine Your App: Use the editor's prompt bar to tweak your app. For instance, to add a feature to track water intake; you can add that command here.

5. Launch: When you're happy with how your fitness app looks and works, press "Publish" to share it with the world. Now, your fitness tracking app is live and ready for users.

⏭️ Stay curious, keep questioning.