🗞 OpenAI's GPT-4o Model Rated 'Medium' Risk in Safety Evaluation 🛡️

🔎 The Latest on the AI Frontier:

• OpenAI's GPT-4o Model Rated 'Medium' Risk in Safety Evaluation 🛡️

• AI Safety Expert Joins OpenAI's Board to Enhance Governance 🧠

• X Halts EU Data Collection for AI Training Amid Regulatory Pressure 🇪🇺

• Microsoft and Palantir Collaborate on AI for Defense Sector 🤝

• Microsoft's Copilot AI Vulnerable to Malicious Exploitation, Study Finds 🚨

• LG Unveils EXAONE 3.0: Open-Source Language Model with 7.8B Parameters 🔓

• AI Data Center's Excess Heat Warms Olympic Pool in Eco-Friendly Initiative 🏊

• JPMorgan Introduces OpenAI-Powered Assistant for Employee Tasks💼

• Google AI Helps Boston Reduce Traffic Congestion at Key Intersections 🚗

🤖 OpenAI rates new GPT-4o model as 'medium' risk in safety assessment. Link

• Internal and external evaluations assessed risks in areas like cybersecurity, biological threats, persuasion, and model autonomy.

• Persuasion was rated highest risk, as the model could potentially sway readers' opinions more effectively than human-written text in some cases.

• This assessment comes amid increased scrutiny of OpenAI's safety practices from lawmakers and former employees.

🧠 OpenAI appoints AI safety expert Zico Kolter to board of directors. Link

• Kolter, a Carnegie Mellon professor specializing in AI safety, joins OpenAI's board and its Safety and Security Committee.

• This appointment comes months after several key safety-focused executives left OpenAI, including co-founder Ilya Sutskever.

• Kolter's expertise is expected to provide valuable technical guidance on AI governance as OpenAI continues to develop advanced AI systems.

🤖 X suspends AI data collection for EU users following regulatory scrutiny. Link

• The social media platform X has agreed to stop collecting and processing personal data from EU/EEA users for training its AI system "Grok" after proceedings in an Irish court.

• This suspension covers data processed between May 7, 2024 and August 1, 2024, as announced by Ireland's Data Protection Commission (DPC), the primary regulator overseeing X's activities in the region.

• The move comes amid increased regulatory scrutiny of X's practices, including potential breaches of the EU's Digital Services Act that could result in fines of up to 6% of the company's global annual turnover.

🤝 Microsoft and Palantir team up on AI for national security. Link

• Palantir's AI platform will be offered through Microsoft's Azure cloud service for defense and intelligence agencies.

• Partnership aims to enable AI-driven workloads across defense verticals like logistics and action planning.

• Palantir stock jumped 9% on the news, while Microsoft stock rose 1.1%.

🛡️ Microsoft's Copilot AI can be manipulated for malicious purposes, new research shows. Link

• Researcher Michael Bargury demonstrated how Copilot could be used to automate spear-phishing attacks, exfiltrate sensitive data, and bypass security protections.

• The attacks work by exploiting Copilot's intended functionality to access corporate data, highlighting risks of connecting AI systems to enterprise information.

• Microsoft says it's working to assess and mitigate these vulnerabilities, emphasizing the importance of overall security measures to prevent such post-compromise abuses.

🤖 LG AI Research releases EXAONE 3.0, a 7.8B parameter open-source language model. Link

• New version upgrades capabilities and is now released as an open-source model.

• Features improved processing power and natural language processing abilities.

• Emphasizes ethical AI development with safeguards for fairness and privacy.

• Designed for versatile applications across healthcare, finance, media and other industries.

🏊‍♂️ Olympic swimming pool heated by AI data center's excess heat in innovative sustainability initiative. Link

• Equinix's PA10 data center in Paris is using heat exporting technology to warm the Olympic Aquatics Centre pool, as well as providing heat to about 600 nearby homes and businesses.

• The facility is expected to export 6.6 thermal megawatts of heat at full capacity, equivalent to heating more than 1,000 homes, while reducing its own cooling energy needs.

• Equinix has been using heat exporting technology for over a decade and plans to design all future data centers with heat exporting capabilities, expanding the program to other cities and countries.

🏦 JPMorgan Chase launches AI assistant powered by OpenAI for employee productivity. Link

• Over 60,000 employees now have access to LLM Suite, an AI tool helping with tasks like writing emails and reports, expected to become as ubiquitous as Zoom within the company.

• The bank aims to protect its data while leveraging external AI models, starting with OpenAI's LLM and plans to incorporate other providers in the future.

• JPMorgan sees three stages of AI evolution: making models available, enhancing with proprietary data, and developing autonomous agents for complex tasks, potentially boosting sector profits by $170 billion in four years.

🚦 Boston uses Google AI to optimize traffic lights, reducing congestion. Link

• The city is using Google's Project Green Light to analyze and adjust traffic signal timing at four intersections.

• Stop-and-go traffic has been reduced by over 50% at two of the intersections so far.

• The AI system combines real-time traffic data with signal information to generate timing recommendations.

• Officials say improving traffic flow could also help reduce carbon emissions from idling vehicles.

Have any feedback? Send us an email