AI News
1. Claude 3
Anthropic has introduced a new family of multimodal AI models named Claude 3, which includes three different versions: Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus. Among these, Opus is the most advanced model. The company claims that these models set new industry benchmarks and outperform competitors, including OpenAI’s GPT-4 and Gemini (The Verge).
2. TripoSR
Stability AI, in collaboration with Tripo AI, has released TripoSR, a cutting-edge 3D object generation tool from single images, boasting rapid output and low resource requirements. TripoSR, which outperforms existing models in speed and quality, is available under the MIT license for various uses, including commercial and research. The model is accessible for users with or without GPUs, underscoring its broad applicability across industries like entertainment and design (Stability AI).
3. Inflection-2.5
Inflection AI focuses on building an empathetic and colloquially adept AI with its Inflection 2.5 model, despite it still lagging behind GPT-4 in benchmarks. Inflection AI has enhanced the IQ aspects of its Pi chatbot, backed by Inflection 2.5, enabling it to cover a wider range of topics and showing substantial improvements across various benchmarks. The new model is integrated into the Pi chatbot, which is widely accessible and has shown significant user engagement, with Inflection AI emphasizing its efficiency in training compared to GPT-4 and its real-time web search capabilities (Venture Beat).
4. Read Aloud
OpenAI has introduced a “Read Aloud” feature to its apps and web platform, enhancing accessibility, particularly for the visually impaired, by allowing chats to be vocalized. This development reflects OpenAI’s commitment to inclusivity, informed by feedback from partnerships and user interactions, aiming to make advanced AI technologies beneficial for a broader audience (Forbes).
5. Chat with MLX
The MLX Chat app is a high-performance macOS application that connects local documents to a personalized large language model, supporting efficient document interaction and search without data leaving the device. It leverages advanced machine learning techniques, including retrieval-augmented generation and MLX for Apple silicon, enhancing user interaction with personalized queries and instructions (GitHub).
Other Innovations
1. Brain stimulation
The study introduces theta burst ultrasound stimulation (TBUS) for brain entrainment and modulation of neuronal plasticity in mice, demonstrating that TBUS can induce long-lasting changes in neuronal activity through two distinct protocols. These protocols, targeting the motor cortex, result in bidirectional modulation of plasticity, dependent on specific molecular pathways and protein synthesis. The research highlights the potential of TBUS as a noninvasive method for enhancing motor skill learning and persistent brain function modulation (Science Advances).
2. Obesity drug
Novo Nordisk has reported promising early results for its new oral weight loss drug, amycretin, showing a 13.1% weight loss after 12 weeks, potentially outperforming its injectable drug Wegovy. Amycretin, a GLP-1 and amylin receptor agonist, is hoped to offer benefits with a different action mechanism and easier daily oral dosing compared to weekly injections of Wegovy. The study’s early phase 1 results suggest a safe and tolerable profile for amycretin, with Novo Nordisk continuing to explore additional obesity and weight loss treatments (FierceBiotech).
Articles
1. Can AI Solve Science? (Wolfram)
The article explores the potential and limitations of artificial intelligence in advancing scientific knowledge. Wolfram argues that while AI can significantly assist in scientific endeavors by providing new tools and interfaces, such as linguistic interfaces via large language models (LLMs), it cannot solve every scientific problem due to the inherent computational irreducibility present in many scientific processes. He delves into the ways AI can and cannot contribute to different scientific workflows, including prediction, explanation, and creation, and highlights the importance of computational reducibility in making science possible.
2. A generative AI reset: Rewiring to turn potential into value in 2024 (McKinsey)
The McKinsey article discusses the need for a generative AI reset in 2024, emphasizing that capturing its potential value requires significant organizational and technological changes. Companies must learn from past digital and AI transformations, focusing on innovation, deployment, and improvement at scale. The article illustrates this through a case study of a telecommunications company and suggests strategies for integrating gen AI to gain a competitive edge, highlighting the importance of upskilling, data quality, and a centralized team approach to responsible scaling.
3. AI prompt engineering is dead (IEEE Spectrum)
The article highlights the varied applications of prompt engineering in the commercial sector and introduces new research suggesting that machines, rather than humans, should optimize prompts. Despite advancements in automated prompt optimization, the field’s future seems to involve human interaction, with evolving job roles to adapt to the changing AI landscape.
4. The many uses of mini-organs (MIT Tech Review)
The article discusses the versatile applications of organoids, which are miniaturized and simplified versions of organs produced in the lab. Researchers are using organoids for various purposes, including drug screening, understanding disease mechanisms, and even creating biocomputers. These mini-organs offer a more complex and human-relevant model than traditional cell cultures or animal models, providing a promising tool for biomedical research and potential clinical applications.
5. Artificial intelligence and illusions of understanding in scientific research (Nature)
Scientists are drawn to AI tools in research due to their potential to enhance productivity and objectivity by mitigating human limitations. However, there is a risk that these tools may foster illusions of understanding and contribute to the emergence of scientific monocultures, thereby reducing innovation and increasing susceptibility to errors in the scientific community.
