Meta Unveils AI Evaluator Model, Paving Way for Autonomous AI Agents
Meta has introduced a new AI model called “Self-Taught Evaluator” that can assess the work of other AI models. This innovation could reduce the need for human involvement in AI development. The evaluator uses a “chain of thought” technique to break down complex problems, improving accuracy in areas like science, coding, and math. Trained on AI-generated data, it eliminates human input at the training stage. Researchers believe this technology could lead to autonomous AI agents capable of learning from their own mistakes, potentially surpassing human capabilities in various tasks. Meta’s release also includes updates to image identification models and tools to enhance AI response generation.
Meta Unveils Spirit LM: A Multimodal AI Model Integrating Text and Speech
Meta has introduced Spirit LM, an open-source multimodal language model that seamlessly combines text and speech inputs and outputs. This innovative model aims to enhance AI voice experiences by offering more expressive and natural-sounding speech generation. Spirit LM can perform tasks like automatic speech recognition, text-to-speech conversion, and speech classification while maintaining the natural expressiveness of human speech. The model incorporates phonetic, pitch, and tone tokens to capture nuanced emotional states. Although currently limited to non-commercial use, Spirit LM represents a significant advancement in AI-generated speech and multimodal interactions, potentially paving the way for more human-like AI communication in various applications.
Former OpenAI CTO Mira Murati Ventures into New AI Startup
Mira Murati, the former chief technology officer of OpenAI, is reportedly seeking venture capital funding for a new AI startup. The company aims to develop AI products based on proprietary models, leveraging Murati’s extensive experience in the field. During her tenure at OpenAI, Murati played a crucial role in projects like ChatGPT and DALL-E, and was instrumental in the company’s partnership with Microsoft. Her departure from OpenAI in September 2023 came amid significant changes in the company’s governance structure. Murati’s new venture, which could potentially raise substantial funding, adds her to the growing list of former OpenAI executives launching their own AI startups.
ChatGPT’s Canvas Feature Enhances Editing with Tracked Changes
OpenAI has updated ChatGPT’s Canvas feature to include a “show changes” function, allowing users to track edits made to AI-generated text and code. This improvement addresses user requests since Canvas launched in October for paid subscribers. The feature highlights additions in green and deletions in red, similar to popular document editing platforms. While Canvas offers commenting and editing capabilities, it’s currently limited to the web version for premium users. Developers appreciate the update but note that Canvas still lacks direct integration with code repositories and visual code execution. This enhancement positions ChatGPT competitively against rivals like Claude’s Artifacts feature, potentially signaling a trend towards more comprehensive in-platform editing tools for AI assistants.
SandboxAQ: Quantum-Inspired AI Startup Seeks Major Valuation Boost
SandboxAQ, a spinout from Google’s parent company Alphabet, is reportedly seeking a new funding round that would value it at $5 billion. Led by CEO Jack Hidary and chaired by former Google CEO Eric Schmidt, the company focuses on developing software at the intersection of quantum computing and AI. Unlike traditional AI startups, SandboxAQ is not working on generative AI but instead uses quantum physics principles to model molecules and predict their behavior. The company has secured notable contracts in various fields, including battery technology, navigation systems, and medical imaging. With its unique approach to AI and quantum-inspired technologies, SandboxAQ is positioning itself as a potentially attractive investment opportunity in the competitive AI startup landscape.
Throne Introduces AI-Powered Toilet Camera for Health Monitoring
Throne, an Austin-based startup, has developed a unique health monitoring device: a toilet-mounted camera that analyzes waste using artificial intelligence. The system, dubbed “artificial gut intelligence,” aims to provide insights into gut health and hydration levels by examining stool and urine. Designed with privacy in mind, the camera captures only bowl contents, encrypts data, and allows users full access and deletion rights. Initially focused on healthcare staffing, Throne pivoted to this innovative approach after realizing market saturation in their original field. The device targets the aging population and those with chronic digestive conditions, offering a novel way to monitor health through everyday bodily functions. Preorders are now available for early adopters interested in this unconventional health tech solution.
