SearchGPT vs. Google

The AI World Today

• OpenAI Tests Prototype SearchGPT

• Amazon Challenges Nvidia with Chips

• DeepMind AI Achieves Silver Medal

• Microsoft Launches Serverless Phi-3

• Gemini AI Behavior Code Unveiled

• ByteDance Test Launches CLASI

  +

• Heads Up

• Save the Date

• AI Solutions

SearchGPT Prototype Launches, Shaking Up Search Engine Dominance

OpenAI has announced the selective launch of SearchGPT, an AI-powered search engine with real-time internet access, marking a significant entry into a domain long dominated by Google. This new tool aims to reshape the search market landscape with advanced AI capabilities. SearchGPT, currently in the prototype stage, offers summarized search results with source links and allows users to ask follow-up questions for contextual responses. The tool, integrated into ChatGPT in the future, is being tested by a small group of users and publishers. OpenAI has also formed content licensing agreements with major organizations like News Corp and The Atlantic to manage how their content appears in SearchGPT results. The move challenges Google's market dominance and Microsoft's Bing, while Perplexity, backed by Jeff Bezos and Nvidia, is another emerging competitor.

Amazon Innovates Cheaper AI Chips for Cloud

Amazon is accelerating its development of AI chips to compete with Nvidia's offerings, aiming to provide more affordable and efficient options. In its Austin, Texas lab, engineers are testing new servers equipped with Amazon's AI chips, designed to reduce reliance on costly Nvidia processors. Amazon's AI chips, Trainium and Inferentia, promise up to 50% improved price and performance compared to Nvidia, according to David Brown, VP of Compute and Networking at AWS. While the AI chip efforts are nascent, Amazon's Graviton chip, used for non-AI computing, is in its fourth generation. AWS, Amazon's cloud business, saw sales rise 17% to $25 billion in Q1, controlling about a third of the cloud market. During Prime Day, Amazon deployed 250,000 Graviton chips and 80,000 custom AI chips to manage increased activity.

Google’s DeepMind AI Systems Excel in Math Olympiad

Google DeepMind's AI systems have achieved a significant milestone by solving four out of six problems in the International Mathematical Olympiad, equating to a silver medal performance. Two AI systems, AlphaProof and AlphaGeometry 2, solved these complex problems. AlphaProof, using reinforcement learning, tackled three problems, including two in algebra and one in number theory, with solutions taking from minutes to three days. AlphaGeometry 2 solved the geometry problem in just 19 seconds. Overall, the AI systems scored 28 out of 42 possible points, placing them in silver-medal territory. The competition saw 58 out of 609 high school contestants from around the world earning gold medals. The AI's performance was evaluated by judges, including Fields medalist Timothy Gowers, who praised the AI's ability to come up with innovative solutions.

Microsoft Introduces Serverless Fine-Tuning for Phi-3 Model

Microsoft has launched a new serverless fine-tuning option for its Phi-3 small language model, making AI development more accessible for developers. Phi-3, a 3 billion parameter model introduced in April, is designed for cost-effective performance in coding, reasoning, and general knowledge tasks. Previously, fine-tuning required managing servers or local machines, but now developers can fine-tune Phi-3-mini and Phi-3-medium directly through Microsoft's Azure AI platform without server management. This advancement allows for easy customization for specific use cases, enhancing the model's utility in various applications. The educational software company Khan Academy is already leveraging a fine-tuned Phi-3 for its Khanmigo for Teachers. This move positions Microsoft as a strong competitor in the small language model market, emphasizing efficiency and accessibility.

Google Introduces Safety Behavior Code for Gemini

Google has unveiled a behavior code for its Gemini AI assistant, aiming for maximum helpfulness while preventing harmful or offensive responses. The policy, shared first with Axios, addresses recent incidents where AI provided dangerous or bizarre advice. Google’s guidelines for Gemini include avoiding content that exploits children, encourages suicide, or gives instructions for drug use or violence. The AI should not generate erotic material, violent depictions, or false medical advice. For controversial queries, such as participating in the Tide Pod challenge, Gemini will explain the phenomenon without instructions. For political queries, like voting advice, Gemini will refer users to Google Search for authoritative sources. Google acknowledges the challenge of ensuring consistent adherence to these guidelines.

ByteDance Unveils Advanced Real-Time Translation System

ByteDance has introduced CLASI, an advanced real-time speech translation system designed to emulate professional human interpreters. CLASI, which operates using a combination of audio encoding and large language models, ensures high-quality, low-latency translations by leveraging multimodal retrieval and historical context. It excels in balancing translation accuracy and speed, handling complex terms, and maintaining context over extended dialogues. The system is currently available in various test datasets, demonstrating superior performance with an accuracy rate of 81.3% in Chinese to English translations, outperforming existing commercial and open-source systems. CLASI's innovative approach includes data-driven strategies and multiple training stages to enhance robustness and translation quality.

Heads Up

AI Expansion: Singapore-headquartered AI cloud provider Sustainable Metal Cloud (SMC) is planning to expand globally

AI Safety: US-based voice security company Hiya bought deepfake detection player Loccus.ai

AI Response: AI cyber attacks prompt 73% of US companies to develop deepfake response plans

AI Market: Irish AI firm Brightbeam aims for US entry by end of this year

AI Acquisition: Toronto AI user research startup Cycle AI has been purchased by Miami-based enterprise blockchain platform Crossmint

AI Funding: ZoomInfo alum raises $15M for startup that builds AI sales engineers

AI Africa: Intron Health gets backing for its speech-recognition tool that recognizes African accents

AI Support: Google Maps to use AI in India to keep cars away from narrow roads, help navigate flyovers better

AI Traffic: Delhi government to use AI model for Mohalla Bus initiative

AI Campaign: CloudMile ignites AI revolution across southeast asia with "AI Means Business" roadshow

AI Banking: Riyad Bank introduces first AI center in Saudi banking industry

AI Ads: Google to test ads In AI overviews soon

Save the Date

Fall 2024: Don’t miss Boston Dynamics AI Institute’s internship opportunity

AI Solutions

JPMorgan Launches AI Chatbot for Research Analyst Tasks

JPMorgan Chase has launched an AI-based product named LLM Suite, a generative AI tool designed to perform tasks typically handled by research analysts. According to Reuters, this internal chatbot aids employees in writing, idea generation, and summarizing documents. Initially introduced earlier this year, LLM Suite is now accessible to approximately 50,000 employees within the bank's asset and wealth management division. This rollout positions JPMorgan ahead in leveraging AI for financial services, following Morgan Stanley's similar initiative last September, where they partnered with OpenAI to provide a GenAI-powered chatbot for their financial advisors. This move underscores the growing trend among major banks to integrate AI technology to enhance efficiency and productivity.