Mistral AI, a Parisian startup, is making waves in the AI community with its latest updates – Le Chat and Mistral Large, establishing itself as a significant European player in generative AI. While Mistral is mostly famous for its small language models (SLMs), this time, it is in the headlines with a large model. Created by experts who formerly worked at Meta and Google's DeepMind, Mistral AI has quickly drawn attention with its rapid fundraising achievements. Since its start in May 2023, the company has successfully raised $528 million, secured a €2 billion valuation, and a freshly inked partnership with tech giant Microsoft.
Mistral Large
Mistral Large is making headlines not just for its state-of-the-art reasoning and wallet-friendly appeal but also for its strategic alliance with Microsoft. Mistral is officially the second after OpenAI to offer a commercial language model on Azure.
It's designed for complex reasoning tasks across multiple languages, from understanding/transforming texts to generating code. It stands just a bit below GPT-4, positioning itself as an alternative to the big ChatGPT and Claude 2, with a focus on both performance and affordability.
Reasoning and pricing
The key highlights of Mistral Large are its advanced reasoning skills and its budget-friendly nature. It's priced 20% lower than GPT-4 Turbo, charging $8 per million input tokens and $24 per million output tokens, making it an attractive option compared to its competitors.
Speaking about performance, based on the benchmark below, GPT-4 is the only model outperforming it. However, this should be taken with a grain of salt until the model is tested on various use cases.
Going global with multilingualism
Mistral Large understands and speaks French, German, Spanish, and Italian, and is accessible to a more global audience. Here's a comparison chart with Mixtral 8x7B and LLaMA 2 70B (Mistral Large outperforms both).
Math and coding skills
Beyond linguistic capabilities, Mistral Large demonstrates proficiency in mathematical and coding tasks, outperforming other LLMs in several benchmarks. This makes it an attractive choice in more technical fields.
While benchmark achievements are impressive, the real test for Mistral Large is in its application to everyday tasks. For those considering its use, consider looking for independent evaluations and comparisons, especially in practical scenarios like coding, to fully assess its capabilities.
Accessing Mistral Large
Mistral Large is now accessible on La Plateforme and Azure, making it easy for users to explore its features. It's also been integrated into Mistral AI's beta assistant, Le Chat, providing a glimpse into its practical applications.
Here’s a step-by-step guide to show you how to deploy Mistral Large on Azure and start using it immediately with LangChain.
Le Chat
Le Chat, a user-friendly chatbot, is Mistral AI's second surprise for us. To dive into Le Chat, you must first sign up on the platform. After setting up your account, be prepared for a short wait—usually one to two hours—as you queue up for access.
Once in, you're greeted by a straightforward interface. Le Chat is currently free and has a context window of 32k tokens, offering users the choice between three models based on their needs: Mistral Small, Mistral Large, and Mistral Next, a concise prototype.
- Mistral Large: for those seeking depth in reasoning,
- Mistral Next: for concise, straight-to-the-point solutions,
- Mistral Small: for fast, efficient answers.
It's worth noting that Le Chat operates without internet access, ensuring your interactions are focused solely on the model's capabilities. Also, its knowledge stops in 2021.
Le Chat in action
You can join the Discord channel of the Mistral community if you’re up to some hot discussions on Le Chat.
And here's a demo of Le Chat's sequential and logical thinking. The first response correctly reasons that a marble is likely to fall out of the cup, unlike peanut butter. It also keeps track of the action sequence and explains its logic at each step.
Le Chat with Mistral Next:
Users’ feedback
We dived into the Discord chats about Le Chat, and here's the lowdown in simpler terms:
System hurdles: People trying out Le Chat have run into a few issues, like crashes and slow responses, probably because many people are checking it out simultaneously. Everyone's pretty patient, though, and they're sharing tips on getting around these problems, hoping things will get smoother.
UI can be improved: People have mentioned that Le Chat's interface isn't as easy to use as it could be, especially when you compare it to ChatGPT's cleaner setup. They're throwing out some ideas for making it better, like being able to rename chats to keep things organized.
Open-source or not?: There's a bit of chat about whether Mistral's models, like Mistral Large, should be open for everyone to tweak or if they should keep them under wraps for commercial use. It's part of a bigger conversation about how companies in the AI game decide to "share their toys."
Knowledge cut: Everyone's really into finding out which Mistral models can do what, especially since they're based on info up to 2021. Lack of access to fresh data is concerning for some users.
In a nutshell, users are genuinely excited about Le Chat but are hoping for some tweaks to make it even better.
Ethical use
As tools like Mistral Large and Le Chat become more integrated into our lives, the importance of ethical considerations grows. The responsibility to use these models ethically falls on both creators and users. Issues such as bias, misinformation, and misuse are critical areas requiring the development of conscientious AI practices.
Le Chat or ChatGPT
Deciding between Le Chat and ChatGPT? It comes down to your specific needs, coding preferences, and whether cost or advanced reasoning is your priority. Each platform excels in its own domain—Le Chat Mistral offers a diverse array of models, making it versatile for various conversational requirements, all within an intuitive and customizable interface.
ChatGPT, on the other hand, is renowned for its robustness and depth, resulting from extensive training on vast datasets. It ensures users navigate its features with ease. Ultimately, the right choice for you hinges on what you expect from a chatbot and how you intend to integrate it into your conversational AI projects.
Wrapping up
Mistral's here to bring AI to a whole new, European level. The fact that the topic was so hot that the servers crashed because of the number of users proves that Mistral's here to change the LLM game. With its diverse offerings and impressive performance, Mistral AI's recent successes signal a transformative shift in the European AI landscape.
