Keeping pace with the rapidly evolving world of AI can be daunting. Until an AI solution can take over this task, here’s a concise summary of the latest developments in machine learning, alongside key research and experiments we haven't previously highlighted.
This week, Google made a big splash by unveiling Gemini, its new flagship multimodal AI model. However, the much-anticipated "lite" version, Gemini Pro, hasn’t lived up to expectations—especially since there are claims of a staged product demo. We'll hold off on our judgement of Gemini Ultra, the complete version of the model, until it begins integration into Google apps and services in early 2024.
Shifting gears from chatbots, a significant event to note is Mistral AI’s recent funding round, which yielded €450 million (approximately $484 million) at a valuation of $2 billion. Mistral, co-founded by veterans from Google DeepMind and Meta, first introduced its Mistral 7B model in September, which it claimed outperformed other models of similar size. Remarkably, the startup secured one of Europe's largest seed rounds before even launching a product.
However, it’s important to highlight that Mistral’s leadership team raises concerns about diversity. My colleague Dominic aptly noted that the co-founders fit a homogenous and privileged demographic, echoing criticisms from The New York Times regarding AI leadership representation.
Despite these concerns, investors view Mistral—and Germany's Aleph Alpha, its occasional competitor—as pivotal players in establishing Europe’s presence in the burgeoning generative AI sector. To date, most high-profile generative AI ventures have originated in the U.S., such as OpenAI, Anthropic, Inflection AI, and Cohere.
Mistral's achievements reflect broader discussions about AI sovereignty. The European Union (EU) is striving to keep pace with technological advancements while implementing regulations to steer development. As Germany’s Vice Chancellor Robert Habeck recently articulated, “The thought of having our own sovereignty in the AI sector is extremely important. But if Europe has the best regulations but no European companies, we haven’t achieved much.”
This week also brought into focus the ongoing debate between entrepreneurship and regulation, as EU lawmakers reached an agreement on policies to mitigate AI risks. Recently, lobbyists, including those from Mistral, have been pushing for a regulatory exemption for generative AI models, though EU lawmakers have resisted granting this exemption—at least for now.
The stakes are high for Mistral and its European counterparts, as industry analysts—and U.S. legislators—watch closely to see how upcoming AI regulations affect investment trajectories. Will Mistral rise to challenge OpenAI amid these regulatory frameworks, or will they create a chilling effect? It remains uncertain, but we are eager to find out.
Here are additional notable AI stories from the week:
A New AI Alliance: Meta aims to expand its reach in the AI landscape by teaming up with IBM to form the AI Alliance, promoting “open innovation” and “open science” amidst the ongoing competition in AI.
OpenAI's Expansion into India: Reports indicate that OpenAI is collaborating with Rishi Jaitly, former Twitter India head, to navigate talks with the Indian government on AI policy and establish a local team.
Google Launches AI Note-Taking App: Google’s AI-enhanced note-taking application, NotebookLM, is now available for U.S. users aged 18 and older. The app features integration with Gemini Pro to enhance document understanding and reasoning.
OpenAI Faces Regulatory Scrutiny: The Competition and Markets Authority in the U.K. has initiated an inquiry into the close ties between OpenAI and Microsoft to determine if they are operating in a “relevant merger situation,” while the FTC also investigates Microsoft’s investments in OpenAI.
Addressing AI Bias: Anthropic suggests that addressing biases inherent in AI models may require a gentler approach—like politely asking systems to avoid discrimination. Devin reports on this intriguing take.
Meta Introduces New AI Features: This week, Meta's generative AI experience, Meta AI, launched new functionalities, including an image creation feature for group chats and support for Instagram Reels.
Funding Boost for Respeecher: Ukrainian synthetic voice startup Respeecher, known for replicating iconic voices for popular franchises, continues to thrive despite regional challenges and increasing competition in the space.
Liquid Neural Networks: A new MIT spinoff aims to construct general-purpose AI systems using liquid neural networks. The company, Liquid AI, recently secured $37.5 million in a seed funding round.
Emerging Trends in Machine Learning: Researchers at EPFL are leveraging orbital imagery to improve tracking of oceanic plastic pollution, employing enhanced machine learning techniques. Meanwhile, biologists face challenges in documenting various animal movements, but a collaborative project with Imperial College London is using synthetic data to bolster their research efforts.
You can read their paper published in Nature Communications for more details.
However, not all generated imagery is reliable. A study by researchers at the University of Washington revealed that the open-source image generator Stable Diffusion 2.1 exhibits biases in its representations, particularly regarding race and gender.
The search for ways to navigate and improve biased AI models is a pivotal conversation among academics and students alike. Yale English professor Ben Glaser provided an optimistic perspective on integrating AI tools like ChatGPT in educational contexts, highlighting the importance of critical thinking and creativity.
In healthcare, the increasing use of AI tools in radiology raises questions about reliability. MIT is exploring automated processes to determine when doctors should trust AI-assisted tools, aiming to enhance patient care without relying solely on AI.
Additionally, we are witnessing a surge of interest in AI models that go beyond text and images, such as Microsoft's new MatterGen, which generates materials tailored to specific properties—streamlining the discovery process.
In summary, the landscape of AI is evolving rapidly, with significant developments across various sectors. Stay tuned for continuous updates in this dynamic field.
Ranking
