Latest news with #LeChat
Forbes
2 days ago
- Business
- Forbes
Mistral AI's Environmental Audit Puts Spotlight On AI's Hidden Costs
Mistral AI Mistral AI has quantified the environmental price of artificial intelligence with unprecedented transparency, releasing what appears to be the first comprehensive lifecycle assessment of a large language model. The French AI startup's detailed analysis of its Mistral Large 2 model reveals that training alone generated 20,400 metric tons of carbon dioxide equivalent and consumed 281,000 cubic meters of water over 18 months. This disclosure comes as enterprises face dual pressures - implementing AI to stay competitive while fulfilling sustainability commitments. The audit provides decision-makers with concrete data points that were previously hidden behind industry opacity, enabling more informed technology adoption strategies. The numbers from Mistral's assessment illustrate the resource intensity of AI. Training the 123 billion parameter model required energy equivalent to 4,500 gasoline-powered cars operating for a year, while water consumption matched filling 112 Olympic-sized swimming pools. Each individual query through Mistral's Le Chat assistant generates 1.14 grams of CO2 equivalent and consumes 45 milliliters of water, roughly equivalent to growing a small radish. Mistral AI More significantly, the analysis reveals that operational phases have a greater impact on the environment. Training and inference account for 85% of water consumption, far exceeding the environmental cost of hardware manufacturing or data center construction. This operational dominance means that environmental costs accumulate continuously as model usage scales up. Mistral's research identifies actionable strategies for reducing environmental impact. Geographic location has a significant influence on carbon footprint, with models trained in regions with renewable energy and cooler climates exhibiting markedly lower emissions. The study demonstrates a strong correlation between model size and environmental cost, with larger models generating impacts roughly one order of magnitude higher for equivalent token generation. These findings suggest specific optimization approaches. Enterprises can reduce environmental impact by selecting appropriately sized models for specific use cases rather than defaulting to larger, general-purpose systems. Continuous batching techniques that group queries can minimize computational waste, while deploying models in regions with clean energy grids substantially reduces carbon emissions. Mistral's disclosure strategy differs significantly from that of its competitors. While OpenAI CEO Sam Altman recently claimed ChatGPT queries consume just 0.32 milliliters of water per request, the lack of a detailed methodology makes meaningful comparison difficult. This transparency gap presents opportunities for companies willing to provide comprehensive environmental data, allowing them to differentiate themselves competitively. The audit establishes environmental transparency as a key differentiator in the enterprise AI market. As sustainability metrics increasingly influence procurement decisions, vendors providing detailed environmental impact data gain advantages in enterprise sales cycles. This transparency enables more sophisticated vendor evaluations that balance performance requirements against environmental costs. For technology executives, Mistral's audit provides decision-making criteria previously unavailable. Organizations can now factor environmental impact into AI procurement decisions, alongside traditional metrics such as performance and cost. The data enables more sophisticated total cost of ownership calculations that include environmental externalities. Looking ahead, environmental performance may become as critical as computational performance in selecting AI vendors. Organizations that establish environmental accounting practices now position themselves advantageously as regulatory requirements expand and stakeholder scrutiny intensifies. The Mistral audit demonstrates that detailed environmental measurement is feasible, potentially making opacity from other vendors increasingly untenable in enterprise markets.
Android Authority
6 days ago
- Android Authority
I fired ChatGPT for a week and hired a European AI instead
Nathan Drescher / Android Authority I've cut my ties to a bunch of American tech. Gmail became Fastmail, Keep turned into Obsidian, and I replaced Tasks with ToDoist. I figured AI would be the hardest one to give up. I've been using ChatGPT in an unorthodox way, not for chatting, but as a full-on project management system for scheduling, reminders, and budgeting. Then I stumbled on LeChat a few months ago. LeChat, from the company Mistral, is Europe's answer to the LLM arms race. I used it instead of ChatGPT to run my life for a full week, pushing it way beyond what it was built for. What I found was a mix of blazing speed and blunt honesty, as well as a bunch of limitations to my workflow. Would you consider switching to a non-American AI tool like LeChat? 0 votes Yes, I'm already using a non-American AI tool. NaN % I'm actively looking for alternatives. NaN % Maybe, I hadn't thought of it before. NaN % No, I'm sticking with American tech. NaN % Switching away from American AI Nathan Drescher / Android Authority I've been feeling uneasy about using ChatGPT for a while now. The leadership drama, opaque corporate structure, its close ties to some people in the US government…these already had me questioning my choice to stick with it. But worse than that was how much ChatGPT 4o feels like it is degrading recently. I don't use this LLM for chatting or writing emails. I use it to manage my day, but it was going off the rails by forgetting things I'd input like my diet, giving me randomly hallucinated events when I asked it what was next on my schedule, and other such shenanigans. LeChat felt different. It is built on Mistral's open models and adheres to Europe's strict privacy laws. It doesn't track me outside of the app like Gemini. Plus, its answers are practically instant. So I closed ChatGPT for a week and made LeChat my go-to assistant for everything I'd normally delegate to an AI. This AI is a breath of fresh air Nathan Drescher / Android Authority The first thing I noticed was how ridiculously fast LeChat is at answering. I'd get a full answer for my prompt the moment I hit enter. I thought at first it must surely be missing context or skipping details, but the responses were accurate and useful. This app is set to Ludicrous Speed all the time. It also didn't try to flatter me. During my test, I asked it to schedule a nighttime vampire hunt on my calendar. LeChat told me vampire hunting wasn't real and moved on. When I recreated the same task with ChatGPT, it played along and scheduled it for a Wednesday night. LeChat wasn't rude, just grounded, and that was refreshing. I didn't have to constantly prompt it to keep it short or just answer the question. Web search, calendar, and deep think LeChat also includes a built-in live web search, making it feel a bit like Perplexity. I didn't need to open a special chat or toggle anything. It just searched the web directly inside the thread when needed. I could also turn this off anytime. LeChat connects to Gmail and Google Calendar too, though I didn't test this since I've moved away from those tools. There are also 'Think' and 'Research' modes. The latter is supposedly aimed at summarizing and analyzing source material, possibly competing with Google's NotebookLM. These features aren't in my usual workflow, but I could see them being incredibly useful for students or researchers who need to quickly summarize and analyze large volumes of text. Here's where it couldn't keep up It wasn't all rainbows and unicorns. The biggest problem was memory. LeChat has none. There's no persistent knowledge of who I am or what I've said in other threads. ChatGPT and Gemini can both retain context, so if I ask them to schedule my son's dentist appointment, they remember who he is and even know the name of his dentist. I need to re-enter the information for each and every chat I start with LeChat. I would love to see LeChat incorporate a memory feature that retains key information across sessions. There's no persistent knowledge of who I am or what I've said in other threads. There are no push notifications either. Maybe that works if you're still using Gmail, but I use Fastmail. I couldn't get LeChat to remind me of anything unless I reopened the app and asked it what I had already told it to remember. It also couldn't create documents the way ChatGPT can. I'm talking about those in-line documents that sit outside the main context window, which are especially useful for quick to-do lists or structured notes. LeChat didn't offer anything like that. I understand these tools aren't designed to manage lives, but I've turned ChatGPT into a decent personal assistant. LeChat couldn't keep up. For one-shot answers, web searches, or brainstorming what to feed my picky daughter, it was great. But it fell short for ongoing support. I'm still glad I tried it Nathan Drescher / Android Authority There's something calming about a tool that doesn't try to be everything. Using LeChat made me realize some of what we've grown used to has grown bloated. LeChat is lean, responsive, and no-nonsense. Its speed is mind-boggling for anyone who has grown used to ChatGPT's laborious answers. It's the kind of AI that might appeal to people who don't want AI running their lives, just helping with their day. I'll probably keep using ChatGPT for the heavy lifting. I still haven't found an LLM that can manage projects like ChatGPT. But LeChat has earned a place in my app drawer. Not every tool needs to be a powerhouse. Sometimes it just needs to answer a question and move on, like LeChat.
Yahoo
6 days ago
- Business
- Yahoo
Asia Morning Briefing: The First AI vs BTC Environmental Impact Numbers are Here. And it Might Start a New Debate
Good Morning, Asia. Here's what's making news in the markets: Welcome to Asia Morning Briefing, a daily summary of top stories during U.S. hours and an overview of market moves and analysis. For a detailed overview of U.S. markets, see CoinDesk's Crypto Daybook Americas. Mistral AI recently offered a rare benchmark in the Artificial Intelligence industry's environmental disclosure, detailing the footprint of its flagship large language model, Mistral Large 2. Over 18 months, training and operating this model generated 20.4 kilotonnes of CO₂-equivalent emissions, consumed 281,000 cubic meters of water, and depleted 660 kilograms of antimony-equivalent materials, Mistral's report said. Notably, a single 400-token response from its chatbot, Le Chat, uses just 1.14 grams of CO₂, 45 mL of water, and 0.16 milligrams of mineral resources. But how does this compare to bitcoin's carbon footprint? After all, bitcoin's energy use has been the subject of significant debate and is often cited when establishing bans on bitcoin mining in jurisdictions. That makes AI inference seem downright frugal compared to Bitcoin's proof-of-work engine. On average, one Bitcoin transaction emits between 600 and 700 kilograms of CO₂, consumes more than 17,000 liters of water, and generates over 130 grams of electronic waste. Zooming out, the entire Bitcoin network emitted roughly 48 million tonnes of CO₂ in 2023, according to the Cambridge Centre for Alternative Finance. It also consumed over 2 billion liters of water and produced more than 20,000 tonnes of e-waste. However, the Cambridge Centre's numbers, although peer-reviewed, have been the source of considerable criticism and require important caveats. First, Bitcoin's electricity mix is not monolithic. According to a survey of miners conducted by BTC Investment fund Batcoinz as of March 2023, Hydropower (23.1%), wind (13.9%), and solar (5%) collectively account for more than 40% of Bitcoin's energy consumption. The difference between the numbers is because surveys done by Batcoinz include off-grid generation. Nuclear energy, often considered carbon-neutral, accounts for another 7.9%. Gas and coal together represent 44%, but Bitcoin's energy profile is more diversified than critics often assume. Second, LLMs may benefit from a cleaner grid by default. For example, nuclear energy comprises over 22% of the European Union's electricity generation, which reduces the CO₂ emissions associated with model training and inference in EU-based data centers such as Mistral's. That advantage isn't due to model architecture, it's grid geography. A U.S.-based training run drawing from coal-heavy regions would present a very different environmental profile. So while the marginal footprint of using an LLM is vastly smaller than processing a BTC transaction, both operate within infrastructure landscapes that significantly shape their true environmental impact. Training frontier models like GPT-4 or Gemini can still require millions of GPU-hours and heavy water consumption, depending on location. Still, Bitcoin's design, mining every 10 minutes regardless of demand, results in a fixed energy cost that scales with time, not usage. In contrast, AI's marginal cost scales with the frequency of model usage. That distinction makes the emissions from a chatbot reply easier to amortize than those from a block reward. As global scrutiny increases over the environmental costs of computation, transparency initiatives like Mistral's, provide important reference points. While proof-of-work is energy-intensive, the Bitcoin blockchain's halving mechanism steadily reduces the rate at which new coins are created, encouraging miners to become more efficient over time. Its environmental footprint should be weighed against the utility it provides in securing a decentralized, global financial network. Continued improvements in clean energy adoption and mining optimization will be key for both BTC and AI as they scale into core pillars of the digital economy. Market Movers: BTC: Bitcoin is trading at $119,500, struggling to maintain momentum after last week's all-time high of $123,100, as retail-driven sell pressure on Binance has pushed Net Taker Volume below $60 million and signaled growing bearish sentiment, according to CryptoQuant. ETH: Ether has pulled back over 3% to $3,696 after a multi-week climb toward $4,000, as technical indicators flash red and analysts question whether the rally can continue without a broader correction, despite ongoing institutional accumulation. Gold: Gold prices rose nearly 1% on Tuesday, with spot gold reaching a five-week high of $3,430.41 amid ongoing trade uncertainty and falling US bond yields, which continue to draw investor interest. Nikkei 225: Asia-Pacific markets opened higher after U.S. President Donald Trump announced a 'massive Deal' with Japan, lifting tariffs to 15% on Japanese exports, with the Nikkei 225 rising 1.71% at the open. S&P 500: US stocks closed mixed Tuesday, but the S&P 500 edged slightly higher to a record 6,309.62 as investors weighed earnings reports Elsewhere in Crypto: Ethereum Validator Exit Queue Nears $2B as Stakers Rush to Exit After 160% Rally (CoinDesk) Crypto Prediction Market Polymarket Weighs Launching Its Own Stablecoin: Source (CoinDesk) Tokenized equities face resistance from prominent Wall Street firm Citadel Securities in letter to SEC (The Block) Error in retrieving data Sign in to access your portfolio Error in retrieving data Error in retrieving data Error in retrieving data Error in retrieving data
Indian Express
18-07-2025
- Business
- Indian Express
Mistral upgrades Le Chat AI chatbot with features like Think Mode, Deep Research and more
Mistral, the France-based AI startup, is adding a bunch of new features to Le Chat, its AI chatbot that will help it compete against the likes of ChatGPT, Gemini and Claude. The new update adds a deep research mode, voice recognition and a redesigned image editor, to name a few. In a blog post, Mistral said that the new Deep Research mode turns Le Chat 'into a coordinated research assistant that can plan, clarify your needs, search and synthesize.' What it means is that users can now ask the AI chatbot in-depth questions, which it will answer by gathering information from credible sources and show the answer in an easy-to-understand structured report. The AI startup also rolled out its Vocal mode, which, as the name suggests, lets you talk to the chatbot just like you would talk to a real person. The new feature is powered by the company's voice-in model named Voxtral. Mistral also introduced 'Think mode', a new feature powered by Magistral, the company's reasoning model. The feature helps users by answering complex questions with 'clear, thoughtful answers.' The AI startup says its new mode is helpful for doing things like drafting a proposal in Spanish, understanding a legal concept in Japanese and doing things in your preferred language. If you use AI chatbots for different things, the newly rolled out Projects functionality might help you organise things. This works by grouping related chats into separate spaces, with each project featuring its own Library. As it turns out, the feature will also remember which tools and settings you are using. Users can not only upload files directly into a project, but can also help you pull content like conversations and documents from your Library. Le Chat is also getting a redesigned image editor, which, according to Mistral, can help users edit AI-generated images while preserving characters and detail. All of these features are now live, and you can try them out by heading over to ' or using the mobile app available on Android and iOS.
The Hindu
18-07-2025
- Business
- The Hindu
France's Mistral adds new AI features to rival U.S. chatbots
French start-up Mistral announced new features to its AI chatbot Le Chat on Thursday as it tries to carve out a place for itself in a global AI race dominated by the United States and China. Often presented as France's and Europe's AI champion, Mistral competes with American giants such as OpenAI and Alphabet's Google, but lags behind in terms of market share or revenues. "We're making Le Chat even more capable, more intuitive — and more fun — with a collection of powerful new features to help you research more thoroughly, express yourself more naturally, and keep your interactions — text, voice, and images — organized in context", said Mistral in a blog post. While Europe's AI champion Mistral's chatbot has been lagging in some advanced features offered by bigger American rivals such as OpenAI and Google, the latest update brings it closer to the options available in the market. Among the new features accessible to anyone using Le Chat, users can talk to Mistral's AI generative model instead of typing through its "Voxtral" model. It can also be used as a research assistant gathering "credible sources" with a "Deep Research" mode. AI models have all been struggling to sort fake news from authentic information. Valued at $6.2 billion by venture capitalists and feted by French President Emmanuel Macron, Mistral gained global exposure at an international AI gathering in Paris in February. Mistral, which formed partnerships with several top French companies such as TotalEnergies, CMA CGM or Orange, is intent on developing its corporate clientele and implementing its models on-premises.
