AI just crossed a threshold. Conversations are now spoken. Virtual worlds are becoming real. And machines are learning to reason solo.
Claude now handles real-time voice and smart workspace commands. SpAItial pushes generative AI into photorealistic 3D worlds. A new study reveals unsupervised AI reasoning. Plus, the UK is piloting an AI transcription tool, eyeing the $57B GovTech market.
Itās a reinvention of what AI can be. And itās happening faster than anyone expected.
Anthropicās Claude debuts voice mode
SpAItial pioneers photorealistic 3D generative AI
INTUITOR study shows AI mastering self-guided reasoning
UKās Minute tool pilots AI transcription for public sector efficiency
Special highlight from our network
Not every founder wears a hoodie.
You know that hoodie math: raise fast, burn cash, hope for a buyout.
RAD Intel isnāt that. Weāve raised $41M+, grown 1600%, and brought in 7,000+ investorsāwith revenue, not just roadmaps.
Weāre up 20% in Q1. Locked in our NASDAQ ticker: $RADI in Q2. Our AI drives 3.5x stronger results for brands like Hasbro, Skechers, and MGM by predicting what content performs before it runs.
Adobe and Fidelity are backing us. So are insiders from Google, Meta, Amazon, and YouTube.
This is what real traction looks like pre-IPOāand the window to get in early is closing fast.
Shares are $0.60 until May 29. After that, the price moves.
Add high-upside growth to your portfolio before RADās share price changes.
Image Credit: Anthropic
Anthropic is launching Voice Mode for Claude on iOS and Android, finally joining OpenAI, Google, and others in enabling full spoken conversations with its AI. The feature is in beta and rolls out in English over the next few weeks.
Key Details:
Runs on Sonnet 4, Claudeās latest model, with real-time speech-to-text and five unique voice personalities.
Users can seamlessly switch between voice and typing mid-conversation.
Voice-integrated Google Workspace access is available for paid users ā including Gmail, Docs, and Calendar commands.
Free users get 20ā30 voice messages monthly, while paid plans unlock higher limits.
With Claude, ChatGPT, Gemini, and Pi all offering real-time voice, the difference comes down to latency, fluency, and how smart the underlying model actually is. Claudeās launch also throws into sharp relief just how far behind legacy voice assistants like Siri still are; no emotion, no memory, no real reasoning. The interface shift is here, and it speaks.
Image Credit: spAItial
Matthias Niessner, co-founder of Synthesia, is back with SpAItial, a new AI startup chasing what he calls the āholy grailā of generative models: building full, interactive virtual worlds from a single prompt. The company is developing Spatial Foundation Models to natively understand how the world looks, moves, and physically behaves.
Details:
SpAItialās models go beyond surface-level rendering, aiming to grasp geometry, physics, and material properties natively.
The founding team includes former leads from Google, Meta, and Synthesia, with deep expertise in neural rendering and telepresence.
Early prototypes show photorealistic 3D rooms built from short text prompts, with potential across gaming, construction, robotics, and VR.
AI has cracked 2D generation. Video is improving fast. But coherent 3D spaces, the kind that feel alive and responsive, remain unsolved. SpAItial is betting that whoever figures it out first wonāt just power the next gaming engine or metaverse tool, but set the foundation for how AI models understand and operate in physical space.
Image Credit: INTUITOR
In a bold new study, researchers from UC Berkeley and Yale have developed INTUITOR, a training method that teaches language models to reason using nothing but their own internal confidence. No labels, no answers, no external reward signals. Just raw self-belief.
How does it work:
INTUITOR taps into token-level confidence, letting the model judge how certain it is about each word it generates.
The model is then reinforced to pursue what it āfeelsā sure about, sidestepping the need for human-provided ground truth.
On math problems, performance matched supervised training. On code generation, it outperformed it.
Surprisingly, models began thinking more like humans: planning ahead, breaking problems down, and explaining their logic clearly.
This flips the script on how AI learns. Rather than mimicking human answers, the model builds its own internal compass. For tasks where there is no clear right answer or no one around to provide it, this kind of self-guided reasoning could be the key to navigating the unknown.
Image credits: Deposit Photos
The UK government has quietly launched Minute, an AI-powered speech-to-text and meeting summarisation tool developed by its in-house i.AI lab. This secure, browser-based tool delivers real-time transcription and custom features built specifically for public sector needs.
Key developments:
Minute processes all data locally in the browser, avoiding external servers to meet strict public sector security standards
The tool includes speaker recognition, editable transcripts, and API access for smooth integration with government IT systems
It is the first AI product used in a Prime Minister-led meeting, part of a broader £45B productivity drive across UK government
The global GovTech market is valued at $57B, positioning Minute as a major opportunity for AI startups and investors
Minute is already piloting across 25 councils, with early results expected in July 2025.This launch signals a critical moment for AI in government, offering a clear path for startups to succeed in a traditionally cautious sector with thousands of potential customers worldwide.
Weād love to hear from you!
Your opinion really counts and helps us make our newsletter even better for you. Weāre always working to make it more useful and interesting.
This super short survey is your chance to tell us what you think.
š Boost your business with usāadvertise where 10M+ AI leaders engage
š Sign up for the first AI Hub in the world.
š² Our Socials
Reply