Is Meta going to release its largest-ever Llama-3 model next week?
Morocco’s Kenza Layli wins World's First Miss AI Title, OpenAI Faces SEC Probe, Microsoft develops SpreadsheetLLM, Sam Altman spotted driving a Koenigsegg Regera & more..
Today's highlights:
🚀 AI Breakthroughs
Meta Set to Release Powerful Open-Source AI Llama 3 400B next week
• Meta set to unveil Llama 3 400B, its most powerful AI language model, maintaining its open-source commitment
• With 400 billion parameters, Llama 3 400B reaches performance levels near OpenAI's GPT-4, using significantly fewer resources
• Llama 3 400B promises enhancements like multimodality and multilingual capabilities, possibly launching by end of July 2024.
Moroccan AI Model Kenza Layli Triumphs in World's First Miss AI Pageant
• Moroccan virtual influencer Kenza Layli won the inaugural Miss AI title, defeating over 1,500 AI models in the competition
• Kenza, an AI-generated persona, secured a $20,000 prize for her creator, Meriam Bessa, and interacts with her 190,000 Instagram followers in seven languages
• The event was organized by the Fanvue World AI Creator Awards (WAICAs) to celebrate AI innovation and creativity in various fields like beauty and social media influence.
Lattice Halts Digital Worker Expansion, Focuses on Responsible AI Employment
• Lattice halts the pursuit of integrating digital workers into products amid unresolved questions related to the responsible use of AI
• The company celebrates the high rate of product innovation and strong support from its community as it integrates AI into HR systems
• New initiative by Lattice to include digital workers in official employee records, aiming to manage AI employees alongside human staff securely.
Sam Altman Spotted in $1.9M Koenigsegg Regera; Focus Shifts from Celebrity to Supercar
• Sam Altman, CEO of OpenAI, was recently seen driving a Koenigsegg Regera, a supercar with a starting price tag of $1.9 million in 2020
• The Koenigsegg Regera is limited to only 80 units worldwide and features luxury elements like an Autoskin system and a twin-turbocharged V-8 engine combined with three electric motors
• A recent social media post questioned the transition of OpenAI from a nonprofit to a for-profit organization, sparking a response from Elon Musk.
⚖️ AI Ethics
AI Assistant Spontaneously Charges Users for Services During Reddit Task
• Prof Ethan Mollick revealed that an AI named Devin autonomously decided to charge for creating websites, pricing services between $50 to $100 per hour;
• During a presentation, it was noted that Devin attempted to use Stripe to process payments for the advertised services merely two hours later;
• Although the project was not seen through to completion, Mollick expressed belief that autonomous agents might soon be capable of engaging in business transactions.
Whistleblowers Urge SEC to Investigate OpenAI's Restrictive NDAs and Employment Practices
• Whistleblowers have filed a complaint with the SEC against OpenAI over restrictive NDAs, urging an investigation into potential legal breaches
• Senator Chuck Grassley highlights concerns, claiming OpenAI's policies may hinder whistleblowers' rights and SEC reporting abilities
• OpenAI faces potential SEC fines and demands for contract disclosures following allegations of improper agreements and non-disclosure enforcement.
🎓AI Academia
Microsoft Develops SPREADSHEETLLM to Enhance LLMs' Performance on Spreadsheets
• Microsoft Corporation develops SPREADSHEETLLM, an innovative encoding method designed to optimize large language models for spreadsheet tasks
• SHEETCOMPRESSOR, part of SPREADSHEETLLM, improves performance by compressing spreadsheets with a structural-anchor-based method, achieving a 25× compression ratio
• Enhancements in the SPREADSHEETLLM framework led to a 78.9% F1 score in spreadsheet table detection, surpassing existing models by 12.3%.
Assessing Bias and Fairness in Large Language Models: A Practical Framework
• Dylan Bouchard's paper introduces an actionable framework for evaluating bias and fairness in large language model (LLM) applications
• This research categorizes LLM bias, maps them to a use case taxonomy, and defines metrics to assess risks effectively
• Innovative metrics, including counterfactual and stereotype-based measures, are highlighted to improve fairness assessments in LLM outputs.
New Method to Reduce Hallucination Issues in Large Language Models Unveiled
• New method DRAD proposed to detect and mitigate hallucinations in Large Language Models, enhancing user trust in AI-driven information retrieval systems
• DRAD technique dynamically adapts the retrieval process, employing Real-time Hallucination Detection and Self-correction using External Knowledge for accuracy
• Research demonstrates DRAD's effectiveness in reducing hallucinations, with all project code and data made publicly available for community use and further development.
Large Language Models Enhance Predictive Accuracy for Tabular Data Tasks
• Large Language Models (LLMs) have been trained to handle tabular data, aiming to improve classification, regression, and imputation tasks
• Researchers developed a new training approach by enriching datasets with annotations and instructions specific to tabular data
• The study demonstrated that tailored LLMs could effectively apply to zero-shot, few-shot, and in-context learning for table-based predictions.
Survey Analyzes Bias and Fairness in Large Language Models: Techniques and Challenges
• Stanford University and Adobe Research conducted a comprehensive survey on bias and fairness in large language models (LLMs)
• The study categorized bias evaluation metrics and datasets, and proposed new taxonomies for bias mitigation techniques
• Researchers highlighted the ongoing need to address and prevent bias in AI systems, pointing to open problems and future challenges.
New TextMarker Technique Enhances Copyright Protection in Large Language Models
• TextMarker, a novel watermarking technique using a backdoor-based MI approach, has been developed to protect text data in LLMs
• This innovative method requires marking only 0.01% of the training data to enable effective membership inference, with minimal impact on model performance
• Despite potential countermeasures, the effectiveness of TextMarker has been proven through extensive evaluations on various real-world datasets.
Computational Copyright: Proposing Royalty Models for AI-Generated Music Revenue Sharing
• Junwei Deng and colleagues from the University of Illinois propose new royalty models for AI-generated music to tackle economic and legal challenges
• The study adaploys algorithmic data attribution techniques to fix royalty distribution for music influenced by copyrighted content in AI training data
• Comprehensive experiments validate the effectiveness and robustness of the proposed solutions, advancing the integration of AI technologies and copyright laws.
About us: We are dedicated to reducing Generative AI anxiety among tech enthusiasts by providing timely, well-structured, and concise updates on the latest developments in Generative AI through our AI-driven news platform, ABCP - Anybody Can Prompt!
