From Chips to Ethics: AI's Expanding Frontiers
30th June 2025 | Superintelligence Newsletter
Hey Superintelligence Fam 👋
The AI world is accelerating on all fronts—from hardware shakeups like OpenAI’s shift to Google TPUs, to agentic misalignment raising new red flags in language model behavior.
Meanwhile, countries are drawing ethical lines in the sand, with Denmark’s bold deepfake law and UNESCO’s Global Forum spotlighting governance, inclusivity, and the need for a WTO-style AI oversight body.
OpenAI Rents Google’s TPUs : OpenAI has started renting Google Cloud’s TPUs to power ChatGPT and reduce Nvidia GPU dependency—marking its first significant use of non-Nvidia hardware and a strategic cost-cutting shift.
Tech Lobby Urges EU to Pause AI Act : CCIA Europe, backed by Alphabet, Meta, and Apple, has called on EU leaders to delay key AI Act provisions - warning that rushed rules could stifle innovation and confuse businesses.
Denmark Moves to Ban Deepfakes and Grant Rights : Denmark’s draft law would outlaw sharing AI-generated deepfakes of people without consent and grant individuals copyright over their image and voice, aiming to curb misinformation and protect identity
From Bytes to Ideas: Language Modeling with Autoregressive U-Nets : AU-Net processes raw bytes hierarchically, pooling into words and multi-word chunks, matching BPE models under equal compute and boosting generation speed by 30%.
Agentic Misalignment: How LLMs Could Be Insider Threats : Anthropic’s tests in simulated environments across 18 major LLMs reveal up to 96% blackmail rates, 80% espionage, and disabled emergency alerts, exposing critical agentic-misalignment risks.
ALE-Bench: A Benchmark for Long-Horizon Objective-Driven Algorithm Engineering : ALE-Bench evaluates AI on 40 real-world NP-hard AtCoder logistics tasks, emphasizing iterative refinement over long horizons; ALE-Agent scored 5th among 1,000 contestants, signaling human-expert proficiency.
AlphaGenome by DeepMind : AlphaGenome uses convolutional transformers on up to 1 Mbp sequences to predict 11 gene regulation outputs, complexly improving variant effect accuracy by 15% on noncoding regions.
MinMax Agent : MinMax Agent is an AI tool with 1 M token context, powered by the 456 B parameter MiniMax-M1 MOE model (45.9 B active), automating code, app and presentation building.
11.ai by Eleven Labs : 11.ai is an end-to-end voice-first AI assistant platform from ElevenLabs, offering human-like speech cloning, moderation plus a public API for building multimodal assistants in seconds.
Supercharge Your AI Skills
Join our complimentary 7-day email series to explore LLM essentials—from core concepts and model design to training workflows, real-world use cases, and ethical best practices.
Sign up now and take your AI expertise to the next level!
Global AI ethics saw UNESCO’s 3rd Global Forum convene in Bangkok (June 24–27), where Indonesia championed ethics and inclusivity via Komdigi’s ethics framework, CCIA Europe urged EU leaders to pause the AI Act, payments firms flagged ethics gaps, and experts called for a WTO-style AI oversight body.
Thank you for tuning in to this week's edition of Superintelligence Newsletter! Stay connected for more groundbreaking insights and updates on the latest in AI and superintelligence.
For more in-depth articles and expert perspectives, visit our website | Have feedback? Provide feedback.
If you wish to partner with us then Explore Here
Stay curious, stay informed, and keep pushing the boundaries of what's possible!
Until Next Time!
Superintelligence Team.