- Ben's Bites
- Posts
- Mistral takes next steps to challenge OpenAI.
Mistral takes next steps to challenge OpenAI.
Mistral AI, the star-studded European startup, is shaking up the AI world. After their initial torrent releases, the company is onto serious stuff with a new language model, Mistral Large. That’s not it, it’s also launching a chat platform and bringing Microsoft on board as a partner.
What’s going on here?
Mistral AI drops cutting-edge AI models, and partners with Microsoft for infrastructure and distribution. Translation: they're about to get big(ger).
What does this mean?
Mistral claims that Large is the second-best AI model available via an API after GPT-4 and benchmarks follow. Large beats GPT 3.5, Gemini Pro 1.0 but they haven’t mentioned Gemini Ultra which is likely to be better than Large (Gemini Ultra is not available via API).
Mistral Large is a multilingual model. Apart from English, it handles French, Spanish, German, and Italian like a native. For developers, this model also comes with native function calling and JSON mode. As for pricing, it’s roughly 20% cheaper than GPT-4.
For bringing this model to users, Mistral is running two playbooks: a) Launch its own chat interface: Le Chat and b) Partner with Microsoft and provide Mistral on Azure.
On Le chat, you can chat with three models from Mistral: Large, Next and Small. Next is their prototype model with short and chatty responses. Mistral Small is an updated zippier package that outperforms Mistral’s older model Mixtral. Le Chat is in beta, and can’t browse the web yet.
The partnership with Microsoft is not just providing models on the Azure cloud. FT reports that Microsoft is investing some $$s to get a minor stake in this 10-month-old startup.
Why should I care?
Let’s get real here, if you are an average AI user (or developer), Mistral Large is not changing how you interact with AI. It’s at a similar pricing as GPT-4 but performs slightly worse. What it’s signalling is that the business of making foundational AI models is shifting from having only Open AI as an option to having other players in the game (who are catching up fast).
Also, Mistral is the only European startup playing in the big leagues with OpenAI, Google, Anthropic and others. Although a bit annoyed that their open-source approach and quirky releases that made them stand out are somewhat fading with this release. Hope they stick to their guns.
Reply
Jane Willie • 7 months ago
HIRE A LEGITIMATE CRYPTO RECOVERY COMPANY -->REVENANT CYBER HACKER
Revenant Cyber Hacker has done an incredible testimony that keeps on trending in the heart of everyone who has been able to benefit from its service of funds recovery or mobile remote spy authorization, just like having the key to someone's heart, getting back lost property or finance is a very joyful wonderous moment well deserve to be celebrated, we lose $861,899.11USD, we lent higher part of that money with the hope of paying it in a short while simply because we were told that our investment is insured and profit is guaranteed.
Email: revenantcyberhacker@gmail.com
website: https://revenantcyberhacker.net
TELEGRAM: @ revenantcyberhacker
WhatsApp: +1 (913) 820-0739
(OR ) +1(208) 425-8584
Ken Walls Walls • 11 months ago
BITCOIN LOST OR STOLEN? COIN RECOVERY SPECIALISTS TELLS YOU HOW TO GET IT BACK.
l was originally introduced to bitcoin by an Instagram acquaintance of mine. The notion appealed to me, so I decided to invest through the website. I made money from investments in the beginning, and I continued to do so until the website vanished. I couldn’t get any money out before the incident, but luckily my friend suggested Coin Recovery Specialists. Right on schedule, they helped me receive my entire money back. It was a pleasure to work with them and they did an excellent job. It may seem unbelievable, yet it is exactly what happened. They put a lot of effort into their work, and we were reimbursed. Consult Coin Recovery Specialists
Email: { CoinRecoverySpecialists@Gmail.Com }
san francisca • 11 months ago
This is super interesting but that chart should have more of an explanation - nothing indicating what the percentages are out of etc :)
Keep reading
Anthropic's primer on challenges in AI evaluation.
Nvidia makes it easy to run local LLMs
a16z wants to invest in consumer AI