The Twenty Minute VCCohere Founder, Nick Frosst: How To Compete with OpenAI & Anthropic, and Sam Altman’s AI Disservice
At a glance
WHAT IT’S REALLY ABOUT
Cohere’s Nick Frosst: Pragmatic AI, Enterprise Focus, Not AGI Hype
- Nick Frosst, cofounder of Cohere and former first hire in Geoff Hinton’s Google Brain group, argues that today’s LLMs are transformative for work but fundamentally not a path to AGI as popularly described. He strongly criticizes Sam Altman’s and others’ rhetoric about near‑term existential risk as academically disingenuous and harmful to productive AI discourse and policy. Frosst explains Cohere’s strategy: tightly focused on enterprise use cases, efficient, small-footprint models, and agentic systems that operate safely on internal tools and data, rather than broad consumer products. He also explores talent wars, regulation, open vs. closed models, sovereignty, labor displacement, and why Cohere is betting on long‑lived, infrastructure‑like enterprise AI rather than hype-driven AGI narratives.
IDEAS WORTH REMEMBERING
5 ideasAGI and existential-risk hype is misleading and counterproductive.
Frosst argues that claims about near-term AGI and AI posing imminent existential threats were obviously wrong when made, distort policy debates, and crowd out discussion of real, near-term issues like labor shifts and inequality.
Enterprise-focused LLMs must be trained and architected differently from consumer chatbots.
Cohere optimizes for workplace augmentation (e.g., agents filing expenses, using internal APIs, reading documentation) rather than engagement or entertainment, so training data, evaluation, and product interfaces are all tailored to enterprise workflows and constraints.
Efficiency—especially small, capable models—is a core competitive edge.
Cohere trains models like Command-A to run on as few as two GPUs, enabling real-world deployment for customers constrained by infrastructure, and spends orders of magnitude less on compute than some rivals while still delivering production-grade models.
Benchmarks and leaderboards poorly capture real enterprise value.
Most popular evals (math reasoning, ARC AGI, etc.) rarely reflect what customers actually need; they can be gamed, and high scores don’t necessarily translate into easier deployment, ROI, or reliable behavior on business tasks.
LLMs will significantly reshape white-collar work but won’t independently make breakthroughs.
Frosst believes many routine text-and-tool-heavy tasks (e.g., junior marketing, operations work) will shrink, while human strengths—intuition, cultural understanding, creativity, and original problem-solving—remain central; LLMs are powerful sequence models, not “digital gods.”
WORDS WORTH SAVING
5 quotesI don’t think Sam Altman has done a service to the world by talking about how close AGI is.
— Nick Frosst
When you’re talking about a 25-year-old marketer, some portion of their work is just turning existing text and tools into another form. That’s where models shine. But most of their work is understanding culture and what will resonate. That’s not in the dataset of text from the internet.
— Nick Frosst
Not AGI, ROI. ROI, not AGI.
— Nick Frosst
You can’t think the technology is magic. You can’t think we’re doing spells. You have to know how a language model works and what that means.
— Nick Frosst
I used to be a real technological optimist… I wouldn’t describe myself as a technological optimist over the past 10 years.
— Nick Frosst
High quality AI-generated summary created from speaker-labeled transcript.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome