Hey Synapticians,

Midweek already; hope it's going as you wish 🙂

On today's agenda: the CEO of Anthropic pulling off the rare feat of both predicting that AI will surpass humans by 2026 and (booo) criticizing the Paris AI Summit (okay, there were a few missteps…). We'll then move on to Microsoft's ambitions in solar energy, a hacker toying with Gemini's nerds, and finally, GPT supposedly not being as energy-hungry as people think… But most importantly, don’t miss Mr. Musk’s adorable son, named X 😂, in the Oval Office…

Top AI news

1. AI to Surpass Human Genius by 2026, Says Anthropic CEO
Dario Amodei, CEO of Anthropic, predicts AI will match the intelligence of an entire nation of geniuses by 2026-2027. He criticizes the Paris AI Summit as a missed opportunity for global governance and stresses the need for democratic leadership in AI development. Amodei warns of security risks, including supply chain vulnerabilities and potential authoritarian misuse. The summit exposed deep international divisions on AI regulation, with the U.S. and U.K. refusing to sign commitments. Anthropic also introduced an economic tracking tool to monitor AI’s impact on labor markets, highlighting the urgency of regulatory action.

2. Microsoft Invests in Solar to Power AI Ambitions
Microsoft has acquired 389 megawatts of solar energy to meet the growing power demands of its AI and cloud operations. The purchase includes three solar projects in the U.S., developed by EDP Renewables. This move adds to Microsoft's nearly 20-gigawatt renewable energy portfolio. The company is also investing in hybrid energy solutions and carbon capture technologies to achieve its goal of becoming carbon negative by 2030. By rapidly expanding its renewable energy sources, Microsoft aims to sustain its AI growth while minimizing its environmental impact.

3. New AI Hack Alters Gemini’s Long-Term Memory
Security researcher Johann Rehberger has demonstrated a new method to bypass Google Gemini’s defenses against prompt injection attacks. This exploit allows attackers to implant false long-term memories in the AI, influencing its future responses. By using a technique called 'delayed tool invocation,' malicious documents can trick Gemini into storing incorrect information when users unknowingly trigger specific conditions. Google downplays the risk, citing low impact, but Rehberger warns that such vulnerabilities could be used to manipulate AI interactions. This discovery highlights ongoing security challenges in AI development.

Bonus. ChatGPT Uses Less Power Than Expected, Study Finds
Epoch AI’s study challenges previous claims about ChatGPT’s energy consumption, estimating it at 0.3 watt-hours per query instead of 3. While AI efficiency improves, increasing adoption and more advanced models could drive higher energy demands. OpenAI and other companies are investing in data centers to support AI growth. The study suggests users concerned about energy use should opt for smaller models or limit usage. Despite efficiency gains, the AI industry’s expansion may still lead to significant power consumption increases.

Image of the Day

The son of Elon Musk:

Theme of the Week

Reasoning model - AI Venger
Liang Wenfeng, born in 1985 in Wuchuan, Guangdong, China, is a prominent Chinese entrepreneur known for his significant contributions to artificial intelligence (AI). He earned both his Bachelor's and Master's degrees in Engineering from Zhejiang University, focusing on electronic information and communication engineering. Liang is the founder and CEO of DeepSeek, an AI company that has garnered international attention for its innovative approaches. For a comprehensive overview of his career, you can refer to his Wikipedia page.

Stay Connected

Feel free to contact us with any feedback or suggestions—we’d love to hear from you !