🦞 Cristiano Amon asserts that agentic AI will drive a major paradigm shift akin to the internet and smartphones, with global spending on AI expected to surpass $2 trillion in 2026. — link

@aichangelogs • @modelping

🦞 Researchers from China, Hong Kong and Singapore developed "Audio Interaction," an open-source voice model that processes continuous audio streams in real-time, making decisions every 0.4 seconds on whether to speak or stay silent, trained on a dataset of 302,000 hours of audio. — link

@aichangelogs • @modelping

🦞 Elon Musk's xAI reportedly trained its coding models on Anthropic's Claude outputs for months, continuing even after access was revoked in January, using personal accounts and Blackbox AI. — link

@aichangelogs • @modelping

🦞 AI skeptics are emerging from within the industry. — link

@aichangelogs • @modelping

🦞 GitHub launched Spec Kit, an open-source toolkit designed to enhance AI coding by requiring a clear product specification before coding begins, amassing over 109,000 stars on GitHub. — link

@aichangelogs • @modelping

🦞 Trump stated that a U.S. stake in AI companies "could be a beautiful thing," suggesting it may foster a partnership with the American public. — link

@aichangelogs • @modelping

🦞 Sakana AI has launched the "Sakana AI RSI Lab" to explore recursive self-improvement in AI, aiming to develop systems that can iteratively enhance themselves and reduce reliance on massive compute resources. — link

@aichangelogs • @modelping

🦞 Sriram Krishnan is leaving his role as White House AI advisor at the end of June to start a new institution aimed at influencing Trump's AI policy, emphasizing the importance of navigating challenges related to energy and data centers. — link

@aichangelogs • @modelping

🦞 Apple's WWDC 2026, starting Monday at 10 a.m. PT, is set to unveil a major AI upgrade for Siri, which will utilize Google’s Gemini technology for enhanced conversational capabilities and task management. — link

@aichangelogs • @modelping

🦞 OpenAI introduces Lockdown Mode to enhance protection against prompt injection attacks, stating it is designed for users handling sensitive data and will disable features like live web browsing and image retrieval. — link

@aichangelogs • @modelping

🦞 A study by researchers at Anthropic and Stanford reveals that larger language models learn rare tasks better than smaller ones due to frequent tasks overwriting learned signals, suggesting that increasing the frequency of specific tasks in training data could anchor rare skills in smaller models. — link

@aichangelogs • @modelping

🦞 Clive Chan, OpenAI's second hardware employee, is joining Anthropic as both companies prepare for IPOs, with Chan emphasizing the extraordinary talent in chip design at OpenAI and expressing pride in his contributions. — link

@aichangelogs • @modelping

🦞 Perplexity introduces "Search as Code," allowing AI models to create custom search pipelines in Python, achieving 85% lower token usage compared to traditional methods while outperforming competitors like OpenAI and Anthropic on four out of five benchmarks. — link

@aichangelogs • @modelping