Mistral AI's LLM outsmarts GPT

Welcome back! This new, personal AI newsletter is my fresh start, building on the foundation of our previous conversations. I will be sharing 4 bullet points about AI research, concept, memes and hacks weekly.

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

AI Research

Machine translation is a crucial aspect of Natural Language Processing. Moderate-sized AI models like ALMA are good at translating languages but can be better. They usually need a lot of example sentences to learn, and their translations aren't always perfect. I came cross this paper that talks about CPO (Contrastive Preference Optimization) which makes these AI models better at telling the difference between okay and almost perfect translations. CPO uses a different way of teaching the AI, focusing on tougher examples. This makes the AI prefer making really good translations. As a result, the improved model, ALMA-R, translates better than GPT-4.

AI Meme

Mixtral 8x7B, an AI model by Mistral AI, is making waves by outperforming older models like GPT-3.5 in several tests. It's a part of the 'sparse mixture of experts' (SMoE) models, meaning it's big but still fast and doesn't cost much to run. It's proven itself on many benchmarks, especially in tasks like code generation and reading comprehension. And the best part? It's open-source, so anyone can use or improve it. Hold my parameters, rookie! Mixtral 8x7B is in the house

AI Concept

This week I will try my best to explain about machine learning. Machine learning is like teaching computers to learn from experience. Instead of being explicitly programmed for a task, machines use algorithms to analyze data and make decisions on their own. It's a way for computers to improve and adapt without constant human intervention. Imagine it as training a virtual brain to get better at specific tasks by learning from examples. You use algorithms as tools to automatically learn patterns from data to solve a specific problem and evaluate model's performance through various metrics.

AI Hacks

Time to monetise on AI Gold rush. 💸 Here’s what I would do if I were you.

1. Check Your Plan: Ensure you're a ChatGPT Plus, Enterprise, or Team subscriber. Select 'My Plan' to upgrade if necessary.

2. Explore Existing GPTs: Check for duplicates by exploring GPTs in the store.

3. Verify Your Builder Profile: Confirm your profile in the 'Settings & Beta' section to publish under your name or a verified domain.

4. Create Your Custom Chatbot: Use the 'Create a GPT' option in 'My GPTs' and follow the GPT Builder tool's guidance.

5. Publish Your GPT: Publish immediately by saving and setting visibility to 'Everyone,' or save privately and publish later.

6. Review in Store: After OpenAI's review, find your GPT in the store by searching for it by name or domain.

And that's all I have for you today! Remember, if you ever feel overwhelmed by the pace of AI advancements, just think of it as your computer trying to impress you with its homework. 😝 Stay curious, keep questioning.