In a world obsessed with speed, a new social network called Roost is betting that the future of connection lies in patience. It grew to 300,000 users almost by accident — and now it's forcing the industry to reconsider the cost of constant engagement.
What to know
- Roost is a social network created by a former Google engineer that intentionally slows down communication, acting as an alternative to the always-on, fast-paced culture of platforms like X and Instagram.
- The app grew from a side project to 300,000 users, surprising even its developer.
- On the same day, OpenAI announced a new voice mode that can speak and listen simultaneously, a key capability for live translation and more natural human-computer interaction.
- AI chip maker SambaNova raised $1 billion at an $11 billion valuation in a Series F round, just months after Intel was rumored to be trying to acquire it for about $1 billion.
- Several fast-growing AI startups continue to accelerate revenue, though details on specific companies were not disclosed.
- A separate analysis argued that large language models like ChatGPT and Claude lack the ability to understand movement through space and time — a critical gap for achieving artificial general intelligence (AGI). That gap may be filled by gaming data.
The Slow Revolution: Roost’s Accidental Success
Roost is not another fast-scrolling feed. It is a deliberate counterpoint to the dopamine-driven design of mainstream social media. The app forces users to operate at the speed of a carrier pigeon — messages take time to arrive, interactions are not instant, and the experience resists the urge for constant refresh.
Its developer, a former Google engineer, did not set out to build a movement. According to reports, the project began as a side experiment. But the user base swelled to 300,000 people, suggesting a latent hunger for a quieter, more intentional digital space. The growth was organic, fueled by word-of-mouth from people tired of the noise.
Roost represents a growing pushback against the idea that faster is always better. In an era of algorithmic amplification and notification overload, it offers something rare: permission to be offline.
OpenAI’s Voice Leap: Listening While Speaking
On the same day Roost's story broke, OpenAI released a new voice model that can listen and speak at the same time. This is a fundamental technical leap for live translation and real-time conversation. Current voice assistants typically operate in turn-taking mode: one speaks, then the other. OpenAI’s model breaks that pattern, enabling more natural, overlapping dialogue.
This capability aligns with the broader industry push toward more human-like interaction. Voice is becoming the next interface frontier, and OpenAI is positioning itself to lead it. The new model could power everything from instant translation earbuds to customer service bots that sound genuinely conversational.
But the move also raises questions. If AI can converse fluidly in real time, what does that mean for authenticity and trust? The technology is advancing faster than society has norms for it.
SambaNova’s $1B Bet on Alternative AI Hardware
While software models grab headlines, the hardware race is just as intense. SambaNova, an AI chip startup, closed a massive $1 billion Series F round at an $11 billion valuation. This comes months after Intel was rumored to be considering an acquisition for roughly $1 billion — a stark contrast that underscores how quickly valuations in the AI chip space can swing.
SambaNova designs specialized processors optimized for AI workloads, competing with giants like Nvidia and AMD. The funding spree signals that investors see long-term value in hardware that can handle the unique demands of generative AI and large-scale inference.
The SambaNova raise is a reminder that the AI boom is not just about algorithms — it's about the silicon that runs them.
The AGI Debate: Why LLMs May Not Be Enough
A separate report from TechCrunch questioned whether large language models (LLMs) can ever achieve artificial general intelligence. The argument: models like ChatGPT and Claude are excellent at processing text but poor at understanding how objects move through space and time — an essential skill for generalizing intelligence.
Gaming data may hold the key. Video games create rich, physics-grounded environments where actions have consequences in a 3D world. Training AI on such data could bridge the gap between language understanding and world understanding.
This critique is not new, but it gains weight as companies pour billions into scaling LLMs. If the path to AGI requires a fundamentally different training paradigm, current leaders may need to pivot.
Looking Ahead
Roost's quiet growth suggests a market for deliberate, low-friction social experiences that prioritize human connection over engagement metrics. Whether it remains a niche or sparks a broader trend depends on its ability to retain users without scaling into the very noise it seeks to escape.
Meanwhile, OpenAI's simultaneous voice model and SambaNova's hardware raise the bar for what AI can do — in real-time interaction and raw compute. The convergence of these developments points to a future where technology feels faster, more natural, and yet paradoxically, we may be craving the opposite. Roost might be early, or it might be a signal that the pendulum is beginning to swing back.
