InnoPower.AI Daily AI Newsletter – 10/15/2025

OpenAI’s CEO, Sam Altman, has announced plans to ease certain restrictions on ChatGPT, notably to permit the AI to generate erotica for adult users. This strategic shift indicates a more open approach towards content generation, aiming to cater to the adult segment of its user base.

MIT assistant professor and J-WAFS researcher Ali Aouad proposes the use of an algorithm to revolutionize food assistance programs in the Global South. By leveraging digital platforms, the algorithm aims to optimize food subsidies to enhance nutrition among the most vulnerable populations. This innovative approach could significantly impact public health and economic stability in these regions.

OpenAI, a prominent figure in the AI sector, faces a significant challenge as highlighted by the Financial Times; it must leverage its $13 billion to achieve a $1 trillion valuation within five years. This comes as several of America’s most valuable companies begin to depend heavily on OpenAI for the fulfillment of substantial contracts, marking a crucial period for the AI firm in proving its financial and operational capabilities in the tech industry.

Watershed Bio, co-founded by an MIT alumnus, has introduced a solution that enables researchers without software engineering skills to conduct large-scale data analyses in the field of biology. This innovation aims to speed up biological research by simplifying the process of data handling and analysis, making it accessible to a broader range of scientists.

Visa has launched the Trusted Agent Protocol to enhance security in AI-driven commerce by enabling merchants to distinguish between legitimate AI shopping assistants and malicious bots. This cryptographic verification method aims to accommodate the booming AI shopping traffic without compromising security, addressing the challenges posed by a 4,700% increase in AI-driven site visits to U.S. retailers in the past year.

The MIT Schwarzman College of Computing and the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) have initiated a collaborative research program aimed at propelling AI innovation and its application in addressing critical scientific and societal challenges. The partnership will bring together faculty and students from both institutions, emphasizing the acceleration of AI solutions in various domains.

In 2025, amid the rise of AI agents tackling narrow tasks, EAGLET emerges as a new academic framework designed to enhance AI agents’ performance on long-horizon tasks. Developed by Tsinghua University, Peking University, DeepLang AI, and the University of Illinois Urbana-Champaign, EAGLET integrates a global planner into LLM-based agents, significantly reducing planning errors and hallucinations, thereby improving task efficiency and completion rates without the need for retraining or manual data labeling.

Researchers at MIT have developed and open-sourced a technique named SEAL (Self-Adapting LLMs), which enables large language models (LLMs) to self-improve by generating their own synthetic data for fine-tuning. This progress, significantly detailed in an updated paper and code released on GitHub, marks an advancement over traditional models that require manual updates and optimization, emphasizing the shift towards more dynamic, self-sufficient AI systems.

Researchers at the University of Illinois Urbana-Champaign have proposed a method for retraining parts of artificial intelligence models, specifically large language models (LLMs), to minimize costs and prevent the phenomenon known as ‘catastrophic forgetting.’ By concentrating on retraining narrow components of LLMs, like the multi-layer perceptron, the method aims to maintain prior learning while preventing performance degradation in untargeted tasks.

Mozilla has integrated Perplexity’s AI-powered answer engine into its Firefox browser, offering users conversational, cited responses instead of the traditional list of search links. This update, following positive user feedback from tests in select markets, will soon be available on mobile platforms as well.

Google has introduced Gemini, a new feature for scheduling one-on-one meetings via Google Calendar. This tool enhances efficiency by enabling users to embed available time slots in emails, which turns into a calendar invite as soon as the recipient selects a preferred time. The automation aims to streamline the meeting scheduling process.

Google Meet introduces a new AI-powered makeup filter, offering users the option to digitally apply makeup, such as lipstick, during video calls. This feature is designed to enhance user experience during virtual meetings by allowing them to look their best with minimal effort.

A newly developed AI tool, named SpectroGen, functions as a ‘virtual spectrometer’ by generating spectroscopic data across various modalities such as X-ray or infrared. This innovation enables rapid assessment of material quality, simplifying and speeding up quality control processes in manufacturing and research settings.

Dfinity Foundation launched Caffeine, an AI platform that enables the building and deployment of web applications through natural language prompts, eliminating the need for traditional coding. This platform, distinct from other AI coding tools, operates on a decentralized infrastructure tailored for autonomous AI development and promises data integrity during updates through its unique programming language, Motoko.

Ray Kurzweil, a notable technologist and author, expressed his strong optimism about the future of technology, highlighting potential advancements in AI and longevity research upon receiving the Robert A. Muh award. Kurzweil’s speech emphasized not just the ongoing progress but also the significant positive impacts he anticipates these advancements will bring to society.