7 OCTOBER | LONDON 2024
SEPTEMBER 12TH - 14TH
The O2, LONDON
The CogX Blog
Thought leadership on the most pressing issues of our time
France's ChatGPT competitor is going all in
By the CogX R&I team
November 20, 2024
French genAI startup Mistral AI is taking the fight straight to OpenAI with an arsenal of cutting-edge updates announced earlier this week.
Credit: CogX
The upgrades include web search with citations, a new Canvas tool for content manipulation, and advanced document processing features, all available for free during the beta phase.
The Canvas feature enables users to edit and transform various content types, including web pages, data visualisations, and PowerPoint presentations, through text and voice commands. Users can make real-time modifications without the tedious process of regenerating responses while maintaining version control of their drafts.
Le Chat's capabilities are powered by the newly announced Pixtral Large, a 124-billion-parameter multimodal model that reportedly outperforms several leading models, including Anthropic's Claude 3.5 Sonnet and Google's Gemini 1.5 Pro on key benchmarks. For image generation, Mistral has integrated Black Forest Labs Inc.'s Flux Pro model.
Credits: Mistral AI
Additionally, Le Chat now supports automated workflows for complex tasks like invoice processing and expense reporting through "agentic AI," a feature currently unavailable in ChatGPT.
The update represents a significant step in Mistral's evolution since its founding in April 2023 by former Google DeepMind and Meta employees.
Now read the rest of the CogX Newsletter
A Conversation with Craig Mundie: Navigating 'Genesis: Artificial Intelligence, Hope, and the Human Spirit.'
The conversation around AI often swings between extremes. On one side, we hear promises of AI solving humanity's greatest challenges. On the other, we face warnings about existential threats, job displacement, privacy invasion, and the spread of misinformation at unprecedented scales.
This vision of the future of AI technology has often taken the shape of a zero-sum game, in which there can only be one outcome: either machines are used to enhance human workers, or they will ultimately replace them.
The question of whether AI will be a net negative, however, assumes that AI will outpace all human qualities and abilities — a notion that's far from certain.
In this exclusive Q&A with Craig Mundie, we explore how humans and AI might evolve together in ways that could enhance rather than diminish our human dignity.
… want to keep reading? Check out the full Q&A here on the CogX Blog
Also in the news
Perplexity AI is now a shopping assistant: Perplexity's AI-powered search engine now displays product recommendations in the form of cards. Some of these cards will feature a "Buy with Pro" button, allowing Pro subscribers to purchase items with a single click directly. The best part? All orders placed through this button come with free shipping.
AI agents design potent nanobodies to combat COVID-19: Stanford researchers have introduced the Virtual Lab, an AI system where LLM agents work together to solve complex scientific problems. In a recent study, the Virtual Lab was used to design 92 novel nanobodies targeting SARS-CoV-2 variants. Experimental validation has shown promising binding profiles for these nanobodies, demonstrating their potential as promising therapeutic agents against COVID-19.
Antropic released prompt improver. This AI-powered tool automatically refines existing prompts, applying best practices and techniques like "chain-of-thought reasoning" to boost accuracy of their AI model, Claude. Think of prompts as instructions for your AI. The better you write them, the more accurate and reliable your AI's outputs will be. Poor prompts can lead to mistakes, making it risky to use AI in critical business workflows.
OpenAI released desktop app integration: The latest update to the ChatGPT desktop app allows users to directly access third-party applications like VS Code, Xcode, Terminal, and iTerm2. This move, inspired by GitHub Copilot's success, aims to streamline the process of using AI-generated content.
Other fun releases:
Alibaba’s new AI models: Alibaba Cloud has unveiled a suite of new AI coding models, including a 32B parameter model that rivals the performance of closed-source models like GPT-4 and Claude 3.5 Sonnet on certain benchmarks. The best part? This powerful model is completely open-source, making it accessible to developers worldwide.
Nvidia introduced Edify 3D: By leveraging advanced techniques like multi-view diffusion and transformers, Edify 3D can transform text or image prompts into high-quality 3D models complete with UV maps and 4K textures.
Ever wondered if AI could outperform your doctor?
A small study designed to assess the impact of ChatGPT on medical diagnosis has yielded surprising results. The AI chatbot outperformed human physicians, achieving a higher diagnostic accuracy rate than the doctors themselves.
Source: Midjourney
The research, involving 50 physicians, showed that AI outperformed physicians by a striking 16 percentage points, achieving a 90% success rate compared to the doctors' 74%.
Surprisingly, when doctors were given access to ChatGPT as a diagnostic aid, their accuracy only marginally improved to 76%, falling far short of the AI's standalone performance.
The study's lead researcher, Dr. Rodman, expressed astonishment at these findings. "I was shocked," he told The New York Times. The core challenge wasn't technological but human. Physicians largely dismissed AI suggestions that contradicted their initial diagnoses, with many struggling to effectively use the AI's capabilities.
"They didn't listen to AI when AI told them things they didn't agree with," Rodman noted, explaining that most doctors were wedded to their own diagnoses and couldn't be convinced a chatbot knew more than them.
In case you missed it
Hopkins researchers trained surgical robots to mimic human surgical skills using video-based imitation learning.