EU Probes Microsoft-Mistral Deal, Mistral Large Released, Meta Readies Llama 3, & DatologyAI's Seed Round
The EU is scrutinizing Microsoft's partnership with Mistral, Mistral launched their new cutting-edge model, Llama 3 is slated for July, & DatologyAI will curate AI training datasets
EU Scrutinizes Microsoft’s €15 Million Investment in Mistral
The European Commission is closely examining Microsoft's €15 million investment in French startup Mistral AI, coinciding with the release of Mistral's new large language model. More on that below.
This partnership allows Mistral to leverage Microsoft Azure's infrastructure for AI development, in exchange for integrating their AI models within Azure's ecosystem. However, this deal, like others in the rapidly evolving AI sector, has drawn regulatory scrutiny over concerns that such collaborations between Big Tech and AI innovators might stifle competition and concentrate power. The EU is delving deeper into the dynamics of these partnerships in an effort to safeguard innovation and regulatory capabilities in the AI landscape.
The question, though, is what happens if the EU blocks this deal. If EU startups are prevented from partnering with hyperscale cloud providers like Microsoft, while American startups can, what will that mean for these European startups longer term?
Mistral AI Unveils Mistral Large
Speaking of Mistral, they also launched Mistral Large, their most sophisticated language model yet, boasting exceptional reasoning and multilingual prowess in English, French, Spanish, German, and Italian.
This flagship model, now accessible via La Plateforme and Azure, shines in text understanding, transformation, code generation, and sets new benchmarks in AI, ranking as the world's second-best API-accessible model.
Alongside this, Mistral introduced Mistral Small, optimized for latency-sensitive applications. Both models feature advanced capabilities like a 32K token context window for detailed document recall, precise instruction following for customized moderation policies, and function calling for robust application development.
Mistral's partnership with Microsoft Azure marks a significant step in making cutting-edge AI models widely available, offering seamless integration for developers and businesses aiming for tech stack modernization and application development at scale. The question, once again, is whether the EU will let that partnership stand.
Meta Preps Llama 3 for Nuanced AI Interactions
Meta Platforms plans on launching Llama 3 in July, promising to enhance user engagement with more nuanced responses to complex queries.
This latest iteration of its AI language model aims to tackle contentious questions more effectively by providing context rather than avoiding them, marking a significant improvement over its predecessor, Llama 2. The move is part of Meta's broader strategy to refine AI interactions, especially in light of Google's recent pause on Gemini's image-generation feature due to inaccuracies.
Llama 3's development includes efforts to understand and clarify user intentions behind sensitive questions, with Meta also planning to bolster its model with internal oversight on tone and safety. This approach underscores the AI industry's shift towards more responsible and context-aware AI technologies.
DatologyAI Raises $11.65 for Automatic Data Curation in AI Training
In a recent stride towards revolutionizing AI model training, DatologyAI announced an $11.65 million seed round. This startup is at the forefront of developing technology designed to automatically curate AI training datasets, addressing the critical challenge of biases and inefficiencies in these massive datasets.
Models are only as good as their data, and nailing the right training data among billions of examples is quite challenging. With advanced tooling, DatologyAI aims to refine the data preparation process, crucial for training sophisticated models like OpenAI’s ChatGPT and Google’s Gemini.
They have not revealed their client base so far, and currently have 10 employees.