Daily AI Newsletter: GPT creators will be getting paid soon

In today’s email:

👂🏻 Researchers gave AI an 'inner monologue' and it massively improved its performance
📈 ChatGPT now helps you backtest Simple Trading Strategies.
👀 A look at the rise and fall of Emad Mostaque and Stability AI
🧰 9 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

OpenAI’s app store draws investors and students seeking artificial aids

OpenAI has introduced an app store where users can create and access custom versions of ChatGPT, attracting interest from investors and students. Some apps, such as those aiding in academic paper summaries or image design, have gained popularity. However, concerns arise as some GPTs may contravene OpenAI's guidelines, especially those offering unverified financial, legal, or medical advice or promoting academic dishonesty.

Despite the innovative approach, the uptake of these custom GPTs remains modest, with such apps accounting for a small fraction of overall site visits. OpenAI is working on enhancing the GPT store, acknowledging the need for improvements based on user feedback. The initiative aims to emulate the success of digital platforms like Apple's app store, providing a new revenue stream for GPT creators and potentially transforming how users interact with AI technology.

Notable organizations, including AllTrails, Khan Academy, and Kayak, have developed popular GPTs, showcasing the diverse applications of this technology. Canva, an Australian tech company, launched a GPT that assists users in creating social media graphics, exemplifying the potential for GPTs to cater to specific user needs and integrate AI more seamlessly into various services.

Researchers gave AI an 'inner monologue' and it massively improved its performance

Researchers have introduced an innovative training method called QuietSTaR for AI systems, enhancing their reasoning abilities by incorporating an "inner monologue." This approach, which prompts AIs to deliberate before responding, mimics human cognitive processes and has significantly improved performance in reasoning and mathematics. Unlike traditional models like ChatGPT, QuietSTaR enables AI to generate multiple internal rationales before producing an answer, improving its predictive capabilities and learning from interactions.

The QuietSTaR technique was applied to an open-source language model, Mistral 7B, demonstrating a notable boost in its reasoning test scores and nearly doubling its math performance. This method differs from previous AI training strategies by being adaptable across various types of large language models (LLMs), showing promise in bridging the gap between AI and human-like reasoning.

While AI systems like ChatGPT and Gemini rely on neural network architectures, they often struggle with common sense reasoning and contextual understanding. QuietSTaR represents a significant step forward, offering a more dynamic learning process that allows AI to better anticipate and adapt during interactions, suggesting a potential pathway toward more advanced and nuanced AI capabilities in the future.

Introducing DBRX: A New State-of-the-Art Open LLM

Databricks introduces DBRX, a groundbreaking open large language model (LLM) that outperforms existing models in various benchmarks, including language understanding, programming, and mathematics. Utilizing a fine-grained mixture-of-experts architecture, DBRX demonstrates superior efficiency in training and inference, offering faster performance and reduced parameter count compared to other models. Available for use via APIs, DBRX also supports custom training on Databricks' platform, integrating advanced tools and technologies for comprehensive data processing and model optimization.

DBRX's architecture, featuring a high number of smaller experts, enables more effective learning and prediction, supported by technologies like rotary position encodings and grouped query attention. Its performance on standard benchmarks outshines that of well-known models like GPT-3.5 and CodeLLaMA-70B, especially in programming and mathematics. Additionally, DBRX's training process benefits from Databricks' extensive toolkit, enhancing efficiency and effectiveness in model development.

Databricks emphasizes the collaborative effort behind DBRX, acknowledging contributions from various teams and the broader academic community. By making DBRX openly available, Databricks aims to foster innovation and shared progress in the field of AI, encouraging enterprises and the open community to build upon and enhance the model's capabilities for diverse applications.

Google will now let you use AI to build travel itineraries for your vacations

Google is integrating AI into its travel services, launching an update to its Search Generative Experience (SGE) that lets users in the U.S. craft travel itineraries using AI. This new feature, still in an experimental phase, provides suggestions for attractions, dining, and accommodation based on web information and user-contributed data. Although these itineraries are non-transactional, users can export them to Gmail, Docs, or Maps, indicating Google's exploration of AI's potential in travel planning.

Additionally, Google is enhancing Google Maps with new features that allow users to access curated lists of recommendations in certain U.S. and Canadian cities. These lists, sourced from publishers and user contributions, aim to guide users to top-rated and unique local experiences, showcasing Google's commitment to enriching user interactions with its mapping service.

Furthermore, Google is introducing tools for users to personalize their own lists in Maps, enabling them to organize and share their favorite spots. This development reflects Google's broader strategy to deepen its engagement with users' travel planning processes, potentially reshaping how people organize and share their travel experiences.