Listen for yourself- ElevenLabs's new AI tool generates sound effects using prompts
++ -Launch of Codestral -Gemini updates -SEAL Leaderboards -ChatGPT Edu -OpenAI for Nonprofits -Property Graph Index & more..
Today's highlights:
🚀 AI Breakthroughs
ElevenLabs Partners with Shutterstock to Launch AI-Driven Text to Sound Effects Tool
• ElevenLabs' new AI Audio model, Text to Sound, can create sound effects, instrumental tracks, and character voices from text prompts, now accessible to all users.
• The Text to Sound Effects tool is designed to help film studios, video game developers, and content creators produce immersive audio content efficiently and cost-effectively.
• ElevenLabs partnered with Shutterstock to develop this sound effect technology, leveraging Shutterstock's extensive audio library for a more versatile and rich content creation tool.
Google I/O Follow-up: Gemini 1.5 Flash and Pro Enhancements, Google AI Studio Upgrades
• Gemini 1.5 Flash and 1.5 Pro stable releases include new billing options, available now in Google AI Studio
• Rate limits for Gemini 1.5 Flash increased to 1000 RPM with no daily cap, although 1.5 Pro rates remain unchanged
• Google has introduced mobile support and a light mode in AI Studio, enhancing accessibility and user experience.
Claude 3 Expands Tool Use Across Major Platforms for Enhanced AI Interactions
• Claude 3's tool use feature is now available on Anthropic Messages API, Amazon Bedrock, and Google Cloud's Vertex AI
• This feature allows Claude to extract data, convert requests into API calls, search databases, automate tasks, and orchestrate multiple agents
• Companies like StudyFetch, Intuned, and Hebbia utilize Claude's tool use for enhancing educational tools, automating data extraction, and improving workflow management.
Scale.com Enforces Rigorous Expert Evaluation and Methodology Standards
• Evaluations are conducted by rigorously selected experts using industry-specific methods to guarantee top-notch quality and credibility.
• Details on the evaluation methodologies can be reviewed by following the provided link to the host’s website.
• Models are included in the leaderboard only at their initial encounter, ensuring fairness and accuracy in ranking.
ChatGPT Edu Launched for Universities with Enhanced Security and GPT-4o Technology
• ChatGPT Edu, tailored for educational use, integrates advanced AI capabilities, including reasoning across text and vision, with enterprise-level security features.
• Several leading universities, such as the University of Oxford and Columbia University, have successfully implemented versions of this AI technology.
• The adoption of ChatGPT Edu by universities promises affordability and specialized tools for enhancing academia and campus operations.
OpenAI Launches Initiative to Boost Nonprofit Accessibility to Advanced AI Tools
• OpenAI launches OpenAI for Nonprofits to improve tool accessibility, offering ChatGPT at reduced costs for nonprofit organizations.
• Nonprofits can leverage ChatGPT for various tasks like drafting grant proposals and enhancing data analysis to boost productivity and impact.
LlamaIndex Expands Capabilities with New Property Graph Index Feature
• LlamaIndex introduces Property Graph Index, enhancing knowledge graph flexibility with advanced labeling and querying capabilities
• The new Index supports hybrid search functions and complex queries using Cypher graph query language for more dynamic data interaction
• Developers can utilize vector and symbolic retrieval for nodes, featuring integration with multiple extraction methods for customized graph construction.
Codestral Launches: A New Benchmark in AI-Powered Code Generation for Developers
• Codestrel, a new AI model specialising in code generation, supports over 80 programming languages including Python, Java, and JavaScript
• Featuring a 32k context window, Codestrel reportedly outperforms competitors in benchmarks like RepoBench for long-range code generation
• Codestrel is available for testing and integration through dedicated API endpoints and on platforms like HuggingFace under the Mistral AI Non-Production License.
SambaNova's Samba-1 Turbo Sets New Benchmark for AI Speed and Efficiency
• SambaNova Systems' Samba-1 Turbo achieved a record-breaking performance of over 1000 tokens per second in the latest Artificial Analysis benchmark
• Samba-1 Turbo operates the Llama 3 8B model efficiently on just 16 chips, significantly outperforming competitors in speed and cost efficiency
• Demonstrating a breakthrough with Dataflow technology, Samba-1 Turbo offers superior processing speed and lower latency compared to traditional GPUs.
Mumbai's AI Semiconductor Revenue Forecasted to Surge by 33% in 2024
• AI semiconductor revenue forecasted to surge by 33% to $71.25 billion in 2024, driven by increased AI integration in consumer and enterprise sectors
• Demand for AI-enabled chips in data centers and consumer electronics, including AI PCs, projected to significantly boost semiconductor sales
• Microsoft's launch of 'Copilot+ PCs' with AI-dedicated chips expected to further accelerate adoption and demand for AI-specific processors.
⚖️ AI Ethics
EU Unveils New Office for Stringent AI Act Implementation and Oversight
• The EU has inaugurated a new office to oversee the landmark AI Act, focusing on safe and ethical AI applications employing a risk-based approach
• This initiative will see more than 140 experts collaborating for effective AI governance, aiming to confirm the EU's role as a global leader in AI innovation and safety
• Stakeholders from business and academia will work with the new AI office to ensure AI development aligns with European societal values and boosts economic growth.
Apple to Reveal AI Strategy at WWDC: Balancing Innovation with Privacy Concerns
• Apple to unveil AI strategy at WWDC, incorporating both on-device and cloud computing in iOS 18 and other updates
• The company commits to strict privacy with "black box processing," enabling data security during cloud operations
• Plans include potentially enabling lightweight devices to offload processing to the cloud, though details remain unclear.
Meta Uncovers AI-Generated Disinformation Campaign Linked to Israeli Firm STOIC
• Meta discovered "likely AI-generated" content on its platforms, involving fake online personas targeting U.S. and Canadian audiences
• The deceptive content, linked to the Tel Aviv-based firm STOIC, utilized sophisticated generative AI tools to enhance disinformation efforts
• Despite advancements in AI technology, Meta maintains these innovations have not hindered its ability to detect and disrupt covert influence operations.
🎓AI Academia
GNN-RAG Hybrid Model Enhances Question Answering on Knowledge Graphs, Surpassing GPT-4
• GNN-RAG combines the language understanding of Large Language Models with the reasoning power of Graph Neural Networks for enhanced question-answering over Knowledge Graphs
• The GNN-RAG model excels on multi-hop and multi-entity questions, outperforming competing approaches by 8.9–15.5% points in answer F1 on KGQA benchmarks
• Code and detailed results of the GNN-RAG's performance, which matches or surpasses the GPT-4 with a 7B tuned LLM on WebQSP and CWQ benchmarks, are available on GitHub.
Generative AI Enhances IoT and Mobile Networking: A Comprehensive Tutorial and Review
• Generative AI (GAI) is set to revolutionize mobile networking by generating synthetic data that mimics complex data distributions from devices and applications
• A new paper surveys the importance of GAIs in enhancing network management, wireless security, and semantic communication in mobile networking
• Challenges in deploying Generative AI for more efficient mobile and wireless networking are highlighted, emphasizing the need for continued research and development.
New AI Tool Simulates Complex Patient Conversations for Medical Training
• A novel educational AI tool has been developed to simulate patient-provider conversations about sensitive topics using video-based, interactive avatars
• This AI simulation provides a cost-effective, highly-realistic setting for medical trainees to practice difficult healthcare conversations within diverse scenarios
• Future enhancements will focus on integrating actual patient histories into simulations and using AI for immediate feedback to improve learner outcomes.
New Study Reveals Tactics to Manipulate Large Language Models Undetected
• Recent studies reveal vulnerabilities in Large Language Models (LLMs), where attackers manipulate LLMs to extract or input specific, often harmful, information
• A novel manipulation tactic using strategic word insertions from other LLM vocabularies proves difficult to detect, often requiring only a single word to compromise model outputs
• This subversive technique was tested on popular models from the Llama2 and Flan-T5 families, demonstrating successful control over the output using different source models for the attack.
About me: I’m Saahil Gupta, an electrical engineer turned data scientist turned prompt engineer. I’m on a mission to democratize generative AI through ABCP—world’s first Gen AI-only news channel.
We curate this AI newsletter daily for free. Your support keeps us motivated. If you find it valuable, please do subscribe & share it with your friends using the links below!