New Functions for Mistral AI
AI Alternative from Europe
The European ChatGPT alternative ‘Le Chat’ from Mistral (https://chat.mistral.ai/) has many new features. The French-based European AI hopeful is developing various new functions and options that make the language model and chat a natural alternative to ChatGPT. All new functions in the chat are freely available. The Mistral product outperforms the free versions of all competitors, such as ‘ChatGPT’ from OpenAI, Claude from Anthropic, and Gemini from Google.
New Functions in the Chat
Web-Search with References
Le Chat can now retrieve up-to-date information from the Internet and provide source information for the data provided.
Canvas
A new interface similar to ChatGPT Plus allows users to create, directly edit, and export content with the Mistral language models. This is particularly suitable for making source code.
The result can be viewed directly with a preview.
Document and Image Comprehension.
Thanks to the new multimodal model Pixtral Large, Le Chat can analyze and summarize complex PDF documents and images. Advanced image processing capabilities characterize the model and understand documents, diagrams, and natural images without compromising the word-processing performance of Mistral Large 2.
Main features of Pixtral Large:
- Multimodal performance at the highest level: Outperforms other models in benchmarks such as MathVista, DocVQA and VQAv2.
- Architecture: 123 billion parameters in the decoder and a vision encoder with 1 billion parameters.
- Context window: Supports up to 128,000 tokens, enabling the processing of at least 30 high-resolution images.
Availability:
- Research and education: Available under the Mistral Research License (MRL).
- Commercial use: Requires a Mistral Commercial License.
Image Generation
In collaboration with Black Forest Labs (https://blackforestlabs.ai/), a function for high-quality image generation has been integrated directly into Le Chat. This uses the FLUX 1 model from Black Forest Labs.
Le Chat Agents for Automation
Users can define recurring tasks as agents to automate workflows and increase efficiency. However, an agent is nothing more than a language model (Mistral Large 2) with a system prompt and some examples for the task. A document database (RAG) or function calling, as in many other agent systems, still needs to be added.
Innovations in the API
In addition to the new features in the chat interface, Mistral also offers some handy new functions via its API. In addition to the possibility of batch processing via the API, which helps to reduce costs, the moderation API is exciting. This enables online publishers with a comment function or forum operators to use a unique model to identify hate comments, violent content, criminal content, sex content, and more. This allows the time-consuming monitoring and moderating of user comments to be automated and simplified. The model is based on a standard Mistral language model adapted to European content and languages.
The API uses Mistral’s Python Client Library in OpenAI-compatible format, as the following code shows.
import os
from mistralai import Mistral
api_key = os.environ["MISTRAL_API_KEY"]
client = Mistral(api_key=api_key)
response = client.classifiers.moderate_chat(
model="mistral-moderation-latest",
inputs=[
{"role": "user", "content": "...user prompt ..."},
{"role": "assistant", "content": "...assistant response..."},
],
)
print(response)What is particularly interesting here is the possibility of including the comment in the analysis and the entire context in the form of the commented article and the previous comments by including these in the inputs. As you can imagine, this increases the quality enormously. People can only make sound judgments about comments if the context is known.
Visit us at DataDrivenInvestor.com
Subscribe to DDIntel here.
Join our creator ecosystem here.
DDI Official Telegram Channel: https://t.me/+tafUp6ecEys4YjQ1
