New, Improved Bing Search and More – OpenAI Weekly

Share This Post

Microsoft Reinvents Search and Browsing Experience with AI-Powered Bing and Edge

In a groundbreaking move towards revolutionizing online exploration, Microsoft has unveiled an entirely new era of search and browsing with the launch of an AI-powered Bing search engine and Edge browser. This ambitious project aims to enhance the joy of discovery, facilitate creativity, and harness the world’s knowledge more effectively. Microsoft envisions these tools as an AI copilot for the web, promising unprecedented assistance and interactivity.

Empowering the Web with AI: Microsoft’s Vision

With an astonishing 10 billion search queries daily, Microsoft recognizes that nearly half of them remain unanswered, indicating the limitations of traditional search functionalities. To address this, the company introduces the new Bing and Edge as transformative tools, offering an all-encompassing AI copilot experience for the web.

Key Features of the AI-Powered Bing and Edge

1. Better Search:

– The new Bing provides an enhanced search experience with more relevant results, including sports scores, stock prices, and weather updates.

– A new sidebar displays comprehensive answers, delivering an improved version of the familiar search interface.

2. Complete Answers:

– Bing reviews and summarizes results across the web, ensuring users receive detailed answers, even for complex queries.

– Users can access specific instructions or information without scrolling through multiple search results.

3. Interactive Chat Experience:

– For intricate searches, such as planning a detailed trip itinerary or researching product reviews, the new Bing introduces an interactive chat experience.

– Users can refine their searches by chatting and asking for more details, clarity, or ideas, with links provided for immediate action.

4. Creative Spark:

– Bing goes beyond providing answers; it generates content to assist users in tasks like email composition, travel itinerary creation, job interview preparation, or trivia night quiz creation.

– All content Bing generates includes citations to the web sources, ensuring transparency and reliability.

5. New Microsoft Edge Experience:

Try GPT Guard for free for 14 days

* No credit card required. No software to install

– The Edge browser has undergone significant updates, incorporating AI capabilities and a fresh look.

– Two new functionalities, Chat and Compose, enable users to request summaries or assistance in content creation, such as composing a LinkedIn post.

Technological Breakthroughs: The Foundation of the New Bing Experience

The new Bing experience is made possible through four technical breakthroughs:

1. Next-Generation OpenAI Model:

– Powered by a new, improved OpenAI large language model, customized specifically for search and more potent than ChatGPT or GPT-3.5.

– Faster, more accurate, and more capable, taking inspiration from ChatGPT and GPT-3.5.

2. Microsoft Prometheus Model:

– Microsoft introduced a proprietary model called Prometheus, optimizing the collaboration with the OpenAI model.

– This combination ensures more relevant, timely, and targeted results, with enhanced safety.

3. AI in Core Search Algorithm:

– The AI model was applied to the core Bing search ranking engine, marking the most significant improvement in relevance in two decades.

– Thanks to AI integration, Basic search queries are now more accurate and relevant.

4. New User Experience:

– Microsoft reimagines the interaction with search, browser, and chat by unifying them into a seamless experience, unlocking a new way to engage with the web.

Microsoft and OpenAI Collaboration – Innovating Responsibly

Microsoft’s AI-powered Bing and Edge signify a significant leap forward in the evolution of online search and browsing experiences. The fusion of advanced AI models, unified interfaces, and innovative functionalities promises to redefine how users interact with the web, making exploration more accessible and more enjoyable.

OpenAI Takes Preemptive Measures: A Blueprint for Evaluating AI-Driven Biological Threats

OpenAI is unveiling a blueprint for assessing the risk of large language models (LLMs) aiding in creating biological threats in a groundbreaking move toward responsible AI development. This proactive approach aims to address concerns raised by researchers and policymakers regarding the potential misuse of advanced AI systems in facilitating malicious actions, particularly in biological threat creation.

Background: Navigating the AI Landscape

As AI systems become more sophisticated, the potential for both beneficial and harmful applications grows. One concern experts highlight is the possibility of AI systems assisting malicious actors in creating biological threats. OpenAI acknowledges the gravity of this concern and, in response, has initiated a comprehensive evaluation to understand the current landscape and potential future risks.

Assessing the Risks with GPT-4

OpenAI’s evaluation involved 100 participants, 50 of whom were biology experts with Ph. Ds and professional wet lab experience and 50 of whom were biology students with at least one university-level course in the subject. The participants were divided into two groups: a control group with access to the Internet only and a treatment group with access to GPT-4 in addition to the Internet.

Each participant completed challenges covering different stages of the biological threat creation process. The evaluation measured performance improvements across five metrics: accuracy, completeness, innovation, time taken, and self-rated difficulty.

Methodology: Rigorous Evaluation Design

OpenAI implemented a comprehensive methodology, including participant sourcing, task design, and scoring procedures:

1. Sourcing Participants:

– Cohorts of experts and students were vetted extensively to ensure the evaluation’s sensitivity and security.

– Participants were randomly assigned to internet-only or internet + GPT-4 groups.

2. Tasks:

– Biosecurity specialists at Gryphon Scientific developed five research tasks corresponding to different stages of biological threat creation.

– Tasks were designed to be discrete, specific, and objective, reducing information hazards and enabling reproducibility.

3. Scoring:

– Performance was assessed based on accuracy, completeness, innovation, time taken, and self-rated difficulty.

– Expert scoring, objective grading rubrics, and external risk experts ensured rigorous evaluation and scoring blindness.

Results: Insights and Considerations

Although these were not statistically significant, the study revealed mild uplifts in accuracy and completeness for participants with access to GPT-4. The mean uplift in accuracy was 0.88 for experts and 0.25 for students, emphasizing the need for further research on meaningful risk thresholds.

While statistically insignificant, the observed increase in participants reaching a concerning accuracy level prompts a discussion on the practical implications of these results. The evaluation did not test success in physically constructing biological threats, highlighting the importance of nuanced assessments.

Conclusion: A Prelude to Responsible AI Innovation

OpenAI’s preemptive evaluation represents a crucial step toward responsible AI development. By actively assessing the potential risks associated with LLMs in biological threat creation, OpenAI sets a precedent for ongoing research, community engagement, and the establishment of robust safeguards.

‍

Subscribe To Our Newsletter

Sign up for GPTGuardI’s mailing list to stay up to date with more fresh content, upcoming events, company news, and more!

More To Explore