🙊Change Your Accent Using AI

Hey TEDchats Fam!

Happy Thursday! Imagine hopping on a call with a client, and instead of repeating yourself three times, they understand you instantly—no awkward pauses, no misunderstandings. Krisp’s latest AI tool is making that a reality, tweaking accents on the fly while keeping your unique voice intact. 

🧑‍💻 WEEKLY PROMPT

🍕 QUICK BITES

Krisp Unveils Accent-Changing AI for Calls

Krisp has launched an innovative AI feature that can change a user’s accent during calls, starting with Indian English to U.S. English. This tool aims to make conversations clearer by adjusting phonemes while preserving the speaker’s original voice. Tested in enterprise settings, the feature is already showing positive results, including increased sales conversion rates. While still in beta, Krisp has already expanded to include more accents—such as Filipino—and plans to support even more in the near future (watch demo videos).

OpenAI Unveils Major ChatGPT Image Upgrade!

ChatGPT just got a big visual boost. OpenAI has announced that GPT-4o can now generate and edit images natively, replacing DALL·E 3 with more detailed and accurate results. The feature is now live for Plus, Pro, Team, and Free users as the default image generator in ChatGPT, with access coming soon to Enterprise and Edu users. OpenAI says it trained the model using publicly available data and proprietary sources, while also implementing policies to prevent the mimicry of living artists’ styles (see OpenAI’s announcement).

AI vs Humans: A New Puzzle Test Sparks a $700K Challenge

A new test designed to measure AI’s ability to adapt to unfamiliar problems has left top models scratching their heads. The ARC-AGI-2 benchmark, created by the ARC Prize Foundation, challenges AI systems to solve complex visual puzzles without relying on brute-force computation. Even the most advanced models—including GPT-4.5 and Claude 3—barely scored above 1%, while human participants averaged 60%. The ARC Prize Foundation is now offering a challenge: build an AI that scores 85% on ARC-AGI-2 using just $2.50 per task, and win a grand prize of $700,000! (view challenge details)

🧰 AI TOOLBOX 

🤯 MIND SPARK

🧠 Google's AI Takes a Thinking Pause: Google has unveiled Gemini 2.5 Pro Experimental, its most advanced reasoning AI model yet. Designed to "think" before answering, it outperforms top competitors in key coding and reasoning benchmarks. With a massive 1 million token context window—soon expanding to 2 million—Gemini 2.5 Pro is built for complex problem-solving and agentic coding (see Google’s announcement).

🛡️ Fighting Deepfakes with Cutting-Edge AI: A new startup, GetReal, is tackling the dangerous rise of deepfakes and digital impersonations. Co-founded by Hany Farid, a pioneer in deepfake detection, GetReal has raised $17.5 million in funding to launch its forensics platform. With major backers like Forgepoint Capital and strategic investors like Cisco, the company aims to protect high-profile individuals and enterprises from malicious media manipulation (see GetReal’s announcement).

🛑 China’s AI Censorship Machine Gets Smarter: A leaked database reportedly reveals that China is using AI to supercharge its censorship efforts, training a large language model (LLM) to detect and flag sensitive topics automatically. The system, containing 133,000 examples of banned content, targets political dissent, corruption scandals, military discussions, and even subtle criticism disguised as proverbs (read more).

📱 Poe Introduces New AI Subscription Plans: Quora’s AI chatbot platform, Poe, has launched a new $5/month plan, making it one of the most affordable AI chatbot subscriptions available. The platform operates on a point system, with different AI models consuming varying amounts of points per message (see Poe’s announcement).

🔍 Claude’s Web Search is Powered by Brave: Anthropic has quietly integrated Brave Search into its AI chatbot, Claude, for live web search capabilities—closing the gap with other leading AI chatbots. This was revealed after “Brave Search” appeared in Anthropic’s subprocessor list, and users found identical citations in both Brave and Claude’s search results (see user forum).

🤖 Meta's Llama Model Monetization Revealed: Meta CEO Mark Zuckerberg previously downplayed plans to monetize its Llama AI models, but a recent copyright lawsuit suggests otherwise. The filing accuses Meta of training on pirated e-books and reveals it’s earning revenue through undisclosed hosting partners via revenue-sharing deals (see full lawsuit).

🎨 AI ART