The Ghost in the Machine: Google's Lost AI
Google, the titan of search, once held the keys to the generative AI kingdom, only to lock them away. Approximately a year before ChatGPT’s seismic public debut, Google DeepMind had already developed LM Chat, an internal AI chat system eerily similar to what would become OpenAI's flagship product. This wasn't a nascent experiment; it was a fully functional ChatGPT-equivalent, a ghost in Google's own machine.
According to Tibor Blahor Blaho, an AI product leader now at OpenAI who was then with Google DeepMind, the tech giant was "too nervous" to release LM Chat. The fear was palpable: a public-facing conversational AI could irrevocably disrupt Google's incredibly lucrative search monopoly. DeepMind, primarily a research arm, was actively "blocked from shipping products" that might cannibalize the company’s cash cow.
This internal paralysis, born from fear of self-disruption and a research culture not geared for product launch, proved to be a catastrophic strategic misstep. Google's hesitation created the colossal market opening that OpenAI seized with audacious speed and a relentless "bias to ship." Had Google acted, the AI landscape, and OpenAI's meteoric rise, might look vastly different today.
The OpenAI Playbook: Ship, Disrupt, Repeat
Where Google hesitated, paralyzed by its own success, OpenAI embraced a philosophy of relentless execution. Tibor Blahor Blaho, now an AI product leader at OpenAI, witnessed firsthand at Google DeepMind how LM Chat, a ChatGPT-equivalent, was shelved because the company was "too nervous" to release products that could disrupt its core business. OpenAI, by contrast, operates with a profound "bias to ship," fostering an empowering, bottoms-up culture where new ideas quickly translate into public offerings.
This willingness to disrupt oneself, even at the expense of a current cash cow, forms a core tenet of OpenAI’s strategy, directly informed by Google’s misstep. While Google protected its search revenue, OpenAI understands that the future of AI cares little for established revenue streams. They constantly reallocate resources, pursuing new research and ideas that might render existing products obsolete, a critical lesson from Google's past.
Yet, this rapid, almost chaotic innovation is carefully balanced with an unwavering commitment to product quality and simplicity. OpenAI avoids a "hodgepodge" of features by prioritizing user utility and delight. The ChatGPT iOS app, lauded for its performance and natural voice capabilities, stands as a prime example of how fast-paced development, fueled by a bottoms-up culture, can still deliver a highly polished, intuitive user experience.
Beyond Typing: The Dawn of Ultrafast AI
OpenAI’s latest gambit, Ultrafast Mode, for its GPT-5.6 Sol model, completely redefines AI interaction. Boasting speeds up to 14 times faster and generating 750 output tokens per second, this advancement, powered by Cerebras infrastructure, shifts the performance bottleneck entirely. AI generation no longer represents the constraint; network latency, tool calls, and human comprehension now become the limiting factors in real-time workflows.
This breakneck speed, coupled with OpenAI’s advanced voice interaction, transforms user experience. Prompt dictation, as product leader Tibor Blahor Blaho notes, becomes "much more efficient" than typing, fostering a seamless, "in-the-flow" creative partnership. No longer are you managing parallel agents; you are conversing with an instantaneous, intelligent peer, capable of tool use and natural dialogue.
OpenAI isn't just chasing speed; it’s pursuing a fundamentally new relationship with AI. The ultimate aim is an AI that deeply understands your evolving context, immediate goals, and even your attention, transcending the role of a reactive tool. This proactive partner anticipates needs, offers solutions, and becomes an indispensable, almost invisible, extension of your cognitive process.
The New Arms Race: Speed vs. Safety
The AI landscape now defines a clear strategic battle. OpenAI, driving for broad access and maximum utility, continues to aggressively ship innovations like Ultrafast Mode, pushing speeds to 14x and shifting bottlenecks from AI generation to system-level constraints. This mirrors Tibor Blahor Blaho’s vision for widespread, impactful AI.
Countering this velocity, Anthropic champions a safety-first ethos with its "Constitutional AI" framework, prioritizing ethical alignment and controlled deployment. This deliberate approach positions them as a crucial check on unchecked acceleration, creating a vital competitive tension in the market.
Underneath this divergence, the pace of AI evolution accelerates relentlessly. Recursive Self-Improvement (RSI) is no longer theoretical; models like Claude already write the majority of their own code, demonstrating an exponential leap in autonomous capability. This self-development compresses innovation cycles to an unprecedented degree.
Managing this explosive trajectory demands a dual approach. OpenAI’s "bias to ship" culture remains crucial for rapid iteration and real-world impact. Yet, as Tibor Blahor Blaho alludes to with his reset button, the industry must simultaneously maintain a hand on the controls—a constant balancing act between unfettered innovation and the critical need for oversight and responsible deployment. This tension defines the new arms race.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
Frequently Asked Questions
What was Google's LM Chat?
LM Chat was an internal conversational AI developed at Google DeepMind about a year before ChatGPT's public release. According to OpenAI's Tibo Sottiaux, who worked there at the time, Google was 'too nervous' to release it publicly, fearing it would disrupt their core search business.
How does OpenAI's 'bias to ship' philosophy work?
OpenAI's 'bias to ship' is a cultural principle emphasizing rapid product releases and iteration based on user feedback. It contrasts with more cautious corporate environments by encouraging close collaboration between research and product teams to get innovations into users' hands quickly.
What is OpenAI's Ultrafast Mode?
Ultrafast Mode is a new, high-speed tier for OpenAI's models, capable of processing up to 14 times faster than standard models. Tibo predicts this level of speed will become the default within one to two years, enabling real-time, in-flow AI interaction.
What is the 'reset button' associated with Tibo Sottiaux?
The 'reset button' is a playful reference to Tibo Sottiaux's role as the lead of the Codex team at OpenAI. He has the authority to 'reset' usage quotas for developers, an action he's taken to support the community and manage access to the powerful coding model.
How does OpenAI's strategy differ from Anthropic's?
OpenAI focuses on creating powerful, widely accessible general-purpose models. Anthropic differentiates itself with a primary focus on AI safety and predictability, using 'Constitutional AI' to appeal to regulated industries that require highly reliable and ethical outputs.