Meta launched its new AI model, Llama 3, on April 18, 2024. It was the company's most capable openly available LLM to date.
Meta says it has amazing knowledge and capabilities that can outperform some of the best AI models, such as Gemini 1.5, Claude 3 Sonnet, and Mistral. I'll talk more about the comparison later.
First, I will discuss what it is, what technology they have used, and what it can do, reviewing and comparing it with Gemini and Claude.
So, Meta AI is like other AI chatbots like Gemini, Claude, and ChatGPT. It has the same chat box where you can type in natural language, and it will provide the answers.
I am not a tech person who understands complex terms, but I simplify them as much as possible so everyone can understand them.
The new Llama comes in two sizes, 8B and 70B parameters. These are the variables that the model learns and the training process to improve its performance and make better predictions.
Another significant change in Llama 3 is the new tokenizer that expands the vocabulary size to 128 256 from 32K tokens in Llama 2. This is also same for Gemini 1.5 Pro, which comes with a 128K token context window.
Llama 3 is pre-trained on over 15 trillion tokes collected from publicly available sources, seven times larger than Llama 2.
If you want to learn more about Llama 3 technology, please refer to Hugging Face and Meta.
Real-time search results:
Meta AI can now do real-time searches from Google and Bing. Whenever you ask it something, it can browse the internet and give you the most accurate results.
Image generation and gifs
Meta AI has the new high-res image generation on the fly as you type. So, when you type, it will automatically create images.
Coding
Meta AI is also your coding campaign. It can help with Writing code, debugging, explaining, supporting,g and providing examples with almost every Coding language.
Conversation
Meta AI can help you with conversations like writing a letter to your boss, asking about the best books to read, and discussing ChatGPT and other AI models.
I have used the Meta AI for almost one full day, and here is my review. I have categorized it into a few sections.
Searching the internet
I have asked a few questions about Meta AI and will see how accurately it surfs the internet and answers the quires.
I have asked a simple question, "What are the best time management apps for students for managing class schedule assignments and entertainment?" and here is the answer it has been given.
It has done a pretty job here.
Let me ask some complicated questions, like " Latest update of Boston humanoid robot." Here is the answer, and I have also made a small follow-up question.
It has done great work for it, too. But the answer it gave was short but accurate.
Conversations
So, here, conversations mean asking some creative questions or personal questions. I have requested Meta AI only if I want to start a conversation with you and how to get started. It says to ask me questions about hobbies and more.
I started asking some personalized questions about hobbies and gave some excellent answers; later, I made some follow-up questions that were answered well.
The problem is that it gave short answers unless I asked for long answers, but anyway, it worked well,d and the recommendations it made were also good.
Assignments
I am an engineering management student with some mathematical questions. Let's ask Meta AI to see how accurate the answers are.
To my surprise, it gave the correct answers to both questions. This was my first attempt, and it was also my first time trying other chats.
Let me ask some creative questions from my Class. I had a final project, and it was fuc* big and complex, with many calculations Let'sLet'ss compare how Meta AI and ChatGPT can do that.
I have previously done that project from ChatGPT 4 and am now trying it with Meta AI.
As Meta AI doesn't support file uploads, I need to copy the text from the project guidelines and ask the questions.
This is the ChatGPT answer. It has done a pretty decent job, but when it comes to calculations, ChatGPT 4 has struggled a lot. I copy the answers into Perplexity and get the calculations.
Here is a quick video of Meta AI. You can pause it and read it if you like. As I said earlier, if we don't ask Meta AI to provide long answers, it gives short ones, but as this was my presentation, I was fine. Meta AI has done the best job so far compared to ChatGPT. It has given detailed calculations for the project, as shown in the video. I like it.
Which one is better? ChatGPT has an advantage when it comes to long text content. With GPT 4, I could upload files, but it could have been more helpful when it came to calculations; it left me.
While Meta couldn't access any files, it wasn't easy to give instructions. Still, when it comes to overall responses, they were much better and more concise.
Email Writing
Any chatbot can do email writing, but what I am trying here is sending cold emails.
I will give any company name and founder name. And I will ask to write an email for the company.
You can see the response here. What are your thoughts on this?
I like Meta AI email. It remembers the founder's name and has also created niches for students, teams, and others—even me—without mentioning that it's pretty good. I have also tried it with another company, which has done the same.
Image generation
Meta AI surely has a big advantage over image generation. Meta AI can create high-res images on the fly as I type. Here, the prompt for image generation was "Imagine."
Regarding the final output,t I think ChatGPT did a fantastic job. Meanwhile, the Meta AI fly-as-you type has also been imposing.
Article Writing
I asked Meta AI and ChatGPT 4 to write a blog for todoist. I have not given either AI model any features, target audience, or other personalizations to work on.
Here are the articles from both.
Meta AI has written the article beautifully. It has given the title ideas and even a better structure. When finally writing a blog, Meta AI has researched the internet and given better blogs. When you see the conclusion, it has made three stories of its own for blog copywriting.
Whereas ChatGPT 4 has provided the types of content, it needs help sharing the title ideas. In the end, ChatGPT could have done better copywriting to generate leads.
Meta AI has done a great job. Regarding AI detection, ChatGPT and Meta AI have been identified 100% as AI-written text.
Meta AI has given me some better ideas for my blog than ChatGPT, which has given generic ideas.
Meta AI is performing better than expected overall. As it was the third version of the Meta AI, I didn't expect that.
I can now use Meta as an AI option. I have previously tried Gemini, which could be better; even I have been using the Gemini Advance version (pro version).
While Meta AI is not yet perfect or a game changer like never seen before, as Mark Zuckerberg said, it will provide good competition for other AI chats.
What are your thoughts?
