Bonjour! (Hello)

If you are looking for an AI that is extremely good at French, it can only be Mistral AI. So what is Mistral AI and is it more powerful than ChatGPT?

In this article, we will travel to faraway France, to learn about the extremely successful startup, reaching a valuation of 6 billion dollars in 14 months after closing a Series B round of funding with 640 million dollars.

Oh, just a quick note, we have:

• Mistral AI is a company
• Their free botchat is called Le Chat or as many of you will find it Le Chat Mistral.

If you simply want to learn how to use their botchat, click on Le Chat to jump straight down, but if you want to learn more:

Allons-y! (Let’s go!)

Learn About Mistral AI

What is Mistral AI?

Mistral AI is a company founded in France in April 2023 by Arthur Mensch, Guillaume Lample and Timothée Lacroix. Mistral AI has assembled an impressive team. Among them:

• Mensch used to work at Google DeepMind
• Lample and Lacroix come from Meta AI in France (formerly Facebook)

All three of them are over thirty and they met when they were all students at the prestigious École Polytechnique in France.

Mistral AI’s mission is to democratize AI, making it more accessible and useful for developers and businesses. The name “Mistral” is a nod to the strong, cold wind that blows through France – a fitting metaphor for the powerful impact the company wants to have in the AI space.

The outstanding development of Mistral AI

In just its first year, Mistral AI has achieved important milestones:

• Raised €105 million in June 2023, with support from investors such as Lightspeed Venture Partners, Eric Schmidt, Xavier Niel and JCDecaux
• Secures additional €385 million in funding by December 2023, with participation from Andreessen Horowitz, BNP Paribas and Salesforce
• Reaching a valuation of over $2 billion by December 2023
• Announced a partnership with Microsoft in February 2024 to bring its language models to the Azure cloud platform.
• According to the latest news from TechCrunch, Mistral AI reached a valuation of $6 billion in 14 months after closing a Series B funding round of $640 million .

These formidable achievements have helped Mistral AI quickly establish a strong position in the competitive AI industry. Their technology has been used by quite a few large corporations such as Orange, CMA CGM and BNP Paribas, as well as AI-focused startups such as Hugging Face, MongoDB, Cloudflare and OctoAI.

With a rapidly growing valuation and support from high-profile investors and partners, Mistral AI is well positioned to shape the future of AI in Europe and beyond.

Mistral AI’s journey is really interesting and worth following, to see how far this startup will jump when artificial intelligence technology is boiling like a furnace at present 🙂 .

Use cases of Le Chat Mistral

Honestly, I don’t think Le Chat Mistral fits the normal user’s use cases very well ==’ So I don’t know how to write it in a reasonable way.

But if you are a business that wants to build a service based on AI technology, Mistral AI has many outstanding advantages, especially in terms of cost. So, I will list the main use cases for those who want to develop a service for reference.

Content Creation and Marketing

• Create engaging blog posts, articles, and social media content
• Create compelling product descriptions and advertisements
• Support brainstorming ideas for content and campaigns
• Provide writing suggestions and editing support

Customer Service and Support

• Provide 24/7 support through intelligent chatbots, handling common queries and issues
• Classify and route customer requests to appropriate personnel
• Support employees by eliciting feedback and providing relevant information
• Analyze customer feedback and sentiment to identify areas for improvement

Software Development

• Create code snippets and function templates
• Context-based code autocompletion
• Explain complex code and provide documentation
• Identify and suggest fixes for errors and bugs
• Transcoding between different programming languages

Research and Data Analysis

• Summarize research articles and long articles
• Extract important information and data from reports
• Analyze sentiment and opinions on large datasets like social media posts
• Generate hypotheses and research questions based on existing literature
• Support writing and editing research articles

Education and Training

• Create tests, assignments, and assessments that are tailored to each student
• Provide smart tutors and instant feedback
• Answer student questions and assist with homework
• Create summaries and study guides from course materials
• Support language learning through conversation practice

These are just a few examples of the many ways Mistral AI can do this. Being open source means you can download Mistral AI models for free and train your own with your own datasets.

As for calling API directly from them, you can research and refer to it yourself.

Mistral AI models provide

Mistral AI Open Source Models

One of the key things that sets Mistral AI apart is their commitment to open source for open technology and empowering the AI community.

That’s right, because their first models were developed based on Meta’s LLaMA 🙂

The company has released several powerful open source models under the Apache 2.0 license, which allows anyone to freely use, modify, and distribute the software.

Mistral AI’s main open source models include:

• Mistral 7B : A transformer model with 7.3 billion parameters, outperforming larger models like LLaMA 2 13B. It uses Grouped-Query Attention (GQA) mechanism to increase efficiency.
• Mixtral 8x7B : A Sparse Mixture-of-Experts (SMoE) model that uses only 12.9 billion running parameters out of a total of 45 billion. It demonstrates strong cross-lingual performance, being fluent in English, French, Italian, German, Spanish, and code. Mixtral 8x7B outperforms models such as LLaMA 70B and GPT-3.5 in tests.
• Mixtral 8x22B : Currently the highest performance open model from Mistral AI. This SMoE 22B model uses only 39 billion operational parameters out of a total of 141 billion. It shares the multilingual and encryption capabilities of the 8x7B version.
• Codestral 22B : A lightweight framework specially optimized for code generation tasks. Proficient in over 80 programming languages, Codestral 22B is very efficient compared to larger code frameworks.

From these open source models, many different companies and groups have “pulled” this model and created very cool “products” like Dolphin 2.5 with the ability to chat without cover, without any limitations.

In short, it will curse you in the face 🙂 If you want, you can try it on Hugging Face , click on it and you are responsible, I don’t know anything.

Mistral AI’s commercial models

In addition to open source products, Mistral AI offers a range of optimized commercial models designed for high performance. These models are available via API and through cloud platform partners such as Azure.

The main commercial models include:

• Mistral Large : Mistral AI’s flagship model, second only to GPT-4 in performance across many benchmarks. It excels at complex reasoning tasks and has strong multi-language capabilities and good coding skills.
• Mistral Small : Designed for workloads that require low latency and cost efficiency, Mistral Small shares the multi-language and encryption capabilities of the Large model.
• Mistral Embed : A specialized model that converts text into high-dimensional vector representations, enabling advanced semantic analysis for various natural language processing tasks.

These commercial models provide businesses and developers with advanced AI capabilities with the convenience and scalability of API-based services. Mistral AI’s partnerships with major cloud providers like Microsoft Azure further extend the reach and accessibility of these powerful tools.

If you want to try it, you can visit their API page and get free credits to try it out. As for whether anyone has used it or how much revenue they have, I can’t find any data so I’ll just leave it at that.

Le Chat: Chatbot by Mistral AI

What is Le Chat?

Le Chat or Le Chat Mistral is a conversational assistant developed by Mistral AI. It serves as an interaction point to connect with Mistral’s various language models in a user-friendly conversational interface.

One of the standout features of Le Chat is its customization. Users can choose which Mistral AI model will provide feedback to the assistant:

• Mistral Large : This option uses Mistral’s most advanced model, ideal for complex and in-depth discussions.
• Mistral Small : Optimized for speed and efficiency, this model is suitable for handling simple queries and providing quick responses.
• Mistral Next : A prototype model designed to provide concise and accurate answers.
• Mistral Codestral: This is the ultimate model for coding, you can call it for free and use it for free in chat too.

Le Chat can understand context, nuance, and even emotion in user requests, allowing for more natural and meaningful interactions, but it’s best in English, I don’t know French so I can’t rate it yet.

Although still in beta and still in development, Le Chat has demonstrated impressive language understanding and generation capabilities. It can assist users with a variety of tasks, from creative writing and brainstorming to answering questions and providing explanations on various topics.

Advantages and disadvantages of Le Chat Mistral and Mistral

In short, it has quite a few disadvantages when compared to ChatGPT, but it also has its own advantages.
Advantage:

• Free of charge
• All 4 of their paid models are completely free to use.
• Each model is optimized for its own task.
• Has good reasoning ability and is similar to GPT 3.5
• There is a free model you can download and train yourself for personal use.
• Low cost compared to other AI models on the market
• Response speed is quite high, suitable for many different tasks
• Can remember more than ChatGPT version GPT-3.5, this advantage is great
• Vietnamese is better in the latest version. Before, I got the result that I could mine coins in prehistoric times to get rich 🙂

Disadvantages:

• Inference performance lags behind current state-of-the-art models like GPT-4o, Claude 3 or Gemini 1.5.
• Sometimes the answers are confusing
• The ability to respond in Vietnamese is moderate, maybe better than some other open source models.
• Le Chat’s interface is chat only and that’s it. They have not announced a model with multimedia capabilities.
• You also cannot customize your botchat to suit the task you need to perform like in ChatGPT.

Although Le Chat Mistral still has many shortcomings, just like ChatGPT in the beginning, they will gradually upgrade to catch up with OpenAI’s ChatGPT.

How to use Le Chat

How to use Le Chat

Using Le Chat Mistral is quite simple, you just need to register an account, confirm and use it. Here are the step-by-step instructions:

1. Visit the Le Chat Mistral website at chat.mistral.ai .
2. Click the ” Sign up ” button to create an account. You will need to provide an email address and create a password.
3. After creating an account, log in with your credentials.
4. Once logged in, you will be taken to the main chat interface. Here, you can start interacting with Le Chat Mistral.
5. Choose the Mistral AI model you want to use in Le Chat. You can choose from:

• Mistral Large : for complex reasoning
• Mistral Small: for speed and efficiency
• Mistral Next: a prototype designed for short answers
• Mistral Codestral: this is the optimal model dedicated to writing code

6. Type your message or question into the input field at the bottom of the chat window and press ” Send ” or press Enter .
7. Le Chat will process your input and generate a response based on the selected model. The chat history will be displayed in the chat window.
8. You can continue the conversation by typing new requests or questions. Le Chat will maintain the context of the conversation, allowing for more natural and coherent interactions.

Keep in mind that while Le Chat Mistral is a powerful conversational AI, it is still in beta and may have limitations. It does not have real-time internet access, so its knowledge may be outdated in some areas. Also, like all AI models, it may occasionally produce biased or inaccurate responses.

How to write prompt for Le Chat

Just like the AIs based on the Transfomer platform, creating prompts for Le Chat Mistral revolves around a few key ideas: be specific, contextual, and break down tasks. Here are some more detailed tips for writing prompts:

• Be clear and specific : Give Le Chat clear instructions or questions. The more specific you are, the more focused and relevant the response will be.
• Provide context : If you are asking about a specific topic or scenario, give Le Chat some background information. This will help the model understand the context and generate a more appropriate response.
• Break down complex tasks : If you have a multi-step task or a complex question, break it down into smaller, more manageable pieces. This will help Le Chat process it and provide more accurate responses. If you don’t want to do that, throw in Claude 3, which handles these complexities without breaking them down.
• Experiment with different wording : If you’re not getting the response you want, try rephrasing your prompt. Sometimes a slight change in wording can produce a more satisfying response.
• Best languages : Le Chat Mistral is fluent in English, French, Spanish, German, and Italian. Feel free to interact with it in any of these languages for more natural conversations.
• Provide examples : If you want Le Chat to generate content in a specific format or style, provide examples to guide the model. This is especially helpful for tasks like writing emails, creating social media posts, or writing code.
• Use the right model : Choose the Mistral AI model that best suits your needs, as I noted above, you will need to choose the right model then Le Chat will return the best answer, unlike the self-identifying model like ChatGPT.

In short, you have to practice and experiment many times to get the best formula when chatting with Le Chat (if you are patient enough).

Try writing code with Mistral

For fun, I’ll write a short, obscure paragraph to see if Mistral does what I want:

Please help me create an HTML interface with basic CSS to create a website that translates the word "Hello" into 10 different languages ​​including Vietnamese and Chinese.

Le Chat Mistral’s answer:

Hello

Vietnamese Chinese French Spanish Italian German Swedish Dutch Norwegian Danish Portuguese Russian

You can also go to my chat to see the results, to see how the code runs, you just need to create a .txt file on the computer screen, paste the code in, save it, change the file extension from txt to html and double click to open (just in case you are not a coder).

Mistral and GPT-4 Comparison Table, Le Chat Mistral and ChatGPT

Comparison table between Mistral and GPT-4 models

Features	Mistral Large	Mistral Small	GPT-4
Efficiency	Behind GPT-4 on industry standards	Optimized for speed and efficiency	Best performance on standards
Ideal use case	Complex reasoning tasks, creating synthetic content, producing code	Simple tasks like sorting, customer support, content creation	Complex tasks across multiple domains
Context window	32k tokens	32k tokens	Up to 32k tokens
Language ability	English, French, Spanish, German, Italian, code	English, French, Spanish, German, Italian, code	Multilingual, preferably English
Expense	$8 per 1M tokens input, $24 per 1M tokens output	More cost-effective than Large for workloads that require low latency	$30 per 1M tokens input, $60 per 1M tokens output
Availability	Mistral API, select cloud platforms	Mistral API, select cloud platforms	OpenAI API

Comparison table between Le Chat Mistral and ChatGPT

Features	The Mistral	ChatGPT
Basic model	Choose between Mistral Large, Small, or Next prototype	GPT-3.5 line
Multilingual support	Fluent in English, French, Spanish, German, Italian	Mainly English
Customize	The basic Mistral model can be selected based on needs.	Limited customization options
Real-time information	No access to real-time data, may have outdated knowledge	Limited real-time data access
Expense	Free beta access, paid enterprise version expected	Paid access via ChatGPT Plus or API
Unique Features	Choose the model for reasoning or speed, concise mode with Mistral Next	Consistent performance, extensive knowledge
Potential limitations	Still in beta, may produce biased or inaccurate responses	Can also create biased or erroneous responses

These tables highlight some of the key differences and similarities between Mistral AI’s products and OpenAI’s GPT-4 and ChatGPT models.

While the Mistral models are very capable and offer advantages in cost efficiency and customization; the GPT-4 and ChatGPT remain the industry leaders in overall performance and flexibility.

Apprehend

While Mistral AI is not yet able to compete 1-1 with OpenAI, they are still a formidable opponent with their open source model. You can also consider using Mistral AI and develop your own AI applications for example.

With Mistral AI’s remarkable progress, its abundant funding from investors, and its open source approach, it becomes a serious competitor in the growing AI landscape that we can look forward to. Looking forward to the day when we can use a better version of GPT-4o completely free, use as much as you want :3

FAQs about Mistral AI and Le Chat