🗞 OpenAI Reclaims AI Chatbot Crown from Google 🏆

🔎 The Latest on the AI Frontier:

• OpenAI Reclaims AI Chatbot Crown from Google 🏆

• Open-Source AI Model Hermes 3 Debuts with Unexpected Traits 🧠

• Midjourney Streamlines AI Image Creation with New Web Editor 🎨

• Microsoft Cautions Users on AI Service Limitations ⚠️

• Google Expands AI-Powered Search to New Countries 🌍

• Study Reveals LLMs' Strengths and Weaknesses in Reasoning 🔬

• San Francisco Takes Legal Action Against AI 'Undressing' Websites 🚫

• California's AI Safety Bill Sparks Tech Industry Debate 🏛️

• Google Enhances AI Search with New Link Displays 🔍

• Colorado Schools Embrace AI Tools for Education 📚

• Chinese Startups Choose Gaming GPUs for AI Development 💻

🏆 OpenAI's new ChatGPT-4o model reclaims top spot on AI chatbot leaderboard from Google's Gemini. Link

• The latest ChatGPT-4o version scored 1314 points, 17 points ahead of Gemini-1.5-Pro-Exp on the LMSys Chatbot Arena.

• OpenAI's new model shows notable improvements in coding, instruction-following, and handling complex prompts.

• The AI chatbot race remains competitive, with new models from Google, Anthropic, and xAI expected to challenge for the top position soon.

🧠 Lambda Labs and Nous Research launch Hermes 3, an advanced open-source AI model. Link

• Based on Meta's Llama 3.1, Hermes 3 comes in 8B, 70B, and 405B parameter versions.

• The 405B version unexpectedly displays signs of an "existential crisis" when given certain prompts.

• Hermes 3 excels in creativity, reasoning, and instruction following, outperforming other open-source models in benchmarks.

• The model is designed for versatility and can be easily accessed through Lambda's API or deployed on their infrastructure.

🎨 Midjourney launches unified web editor, streamlining AI image creation workflow. Link

• New interface combines features like repainting, region variation, and zooming into one tool.

• Available to users who have created at least 10 images on the platform.

• Update aims to compete with emerging AI image generators and attract a broader range of creators.

🤖 Microsoft updates terms to warn users not to rely on its AI services for anything important. Link

• New Service Agreement says AI services "are not designed, intended, or to be used as substitutes for professional advice."

• Changes prohibit extracting data, reverse engineering AI models, or using Microsoft's AI to train other AI services.

• Microsoft clarifies rules around mass arbitration for disputes involving 25+ customers with similar claims.

🌎 Google expands AI-powered search overviews to 6 new countries with local language support. Link

• AI overviews now available in India, Brazil, Japan, UK, Indonesia and Mexico.

• Google is testing new ways to display source links, aiming to drive more traffic to external websites.

• In India, users can toggle between Hindi and English results and listen to AI-generated responses.

🧠 New research shows LLMs excel at inductive reasoning but struggle with deductive tasks. Link

• Study by UCLA and Amazon researchers evaluated GPT-3.5 and GPT-4's reasoning capabilities.

• LLMs showed near-perfect accuracy in learning rules from examples (inductive reasoning).

• However, they struggled to apply specific instructions, especially in unfamiliar scenarios (deductive reasoning).

• Findings have implications for real-world LLM applications, suggesting caution when relying on their ability to follow explicit instructions.

🚫 San Francisco sues 16 AI 'undressing' websites for creating non-consensual deepfake nudes. Link

• The targeted websites were visited over 200 million times in the first half of 2024 alone.

• The lawsuit accuses operators of violating laws against revenge porn, deepfake porn, and child pornography.

• City Attorney David Chiu calls for urgent societal action to address this "multi-faceted problem" of AI-enabled exploitation.

🏛️ California's AI safety bill sparks intense debate in tech industry. Link

• SB 1047 would require AI developers to provide "reasonable assurance" against catastrophic harms for systems costing over $100 million to train.

• Major tech companies and VCs strongly oppose the bill, citing concerns about innovation and start-ups, while supporters argue it's necessary for public safety.

• The bill has 70% support among Californians in polls, highlighting tensions between industry interests and public sentiment on AI regulation.

🔍 Google expands AI Overviews in Search, adding new link displays to address traffic concerns. Link

• AI Overviews are now available in 6 new countries, including the UK, India, and Japan, with local language support.

• New link displays within AI Overviews aim to increase referral traffic, showing positive results in early testing.

• While some websites may see reduced traffic, Google reports that AI Overviews are driving higher quality clicks to a more diverse range of sites.

✍️ Colorado schools embrace AI this fall, with some districts piloting innovative tools. Link

• The Colorado Education Initiative released a roadmap to help local school districts integrate AI into policies and curricula.

• Several districts are experimenting with AI tools like PowerBuddy and Khanmigo to assist teachers and provide personalized tutoring for students.

• While some districts are enthusiastically adopting AI, others like Denver Public Schools remain cautious due to data privacy concerns and potential overreliance on the technology.

🖥️ Chinese AI startups opt for GeForce RTX 4090 GPUs over sanction-compliant H20 due to cost-effectiveness. Link

• An eight-card Nvidia H20 AI server costs about 1.3 million yuan, while 65 RTX 4090 GPUs can be purchased for the same price.

• Only large companies like Baidu and Alibaba can afford the more expensive H20 AI chips, pushing smaller firms to consumer GPUs.

• Despite export restrictions, Chinese companies are finding ways to acquire full-fat RTX 4090s rather than the China-only RTX 4090D model.

Have any feedback? Send us an email