Is language a sufficient representation for agents to understand the world?
7 recorded positions from 7 people, first said Feb 24, 2025. They do not agree — the readings below are what each one actually argued.
Language remains the core machine interface even as the prompt box disappears
Nick Frosst · Sep 1, 2025 · hedged
Language will become a more important interface for computing, though not the whole of it — graphical interfaces remain better for some tasks — and nobody has yet built the right language-first device
The Rabbit R1 and Humane pin gestured at something interesting but didn't get it right
Scope: 'I don't know if it will' be done right; language is not the best interface for everything
52:30 20VC: Cohere Founder on How Cohere Compete with OpenAI and Anthropic $BNs | Why Counties Should Fund Their Own Models & the Need for Model Sovereignty | How Sam Altman Has Done a Disservice to AI with Nick Frosst
Joelle Pineau · Nov 3, 2025
Language will remain the core paradigm for communicating with machines even as the prompt box disappears
Words are symbols that encode enormous information very efficiently and so much of human communication is language-based
Scope: distinguishes language-as-paradigm from the text prompt box
40:28 20VC: Cohere's Chief AI Officer on Why Scaling Laws Will Continue | Whether You Can Buy Success in AI with Talent Acquisitions | The Future of Synthetic Data & What It Means for Models | Why AI Coding is Akin to Image Generation in 2015 with Joelle Pineau
Language is insufficient agents need a visual world model
Steeve Morin · Feb 24, 2025
He is very bullish on world models / the JEPA approach as the successor to LLMs
Language is only one form of communication and a very narrow, lossy window into the world; JEPA works by minimising the energy to move between states, from which physics and intuitive behaviour emerge, which matches how the brain actually operates; and it comes from Yann LeCun, who is no bullshit
Scope: 'very frontier' — early and unproven
49:58 20VC: Why Google Will Win the AI Arms Race & OpenAI Will Not | NVIDIA vs AMD: Who Wins and Why | The Future of Inference vs Training | The Economics of Compute & Why To Win You Must Have Product, Data & Compute with Steeve Morin @ ZML
Jerry Murdock · Feb 28, 2026
Language is a subpar descriptor of reality, so autonomous agents need a new visual recognition system to get an objective view of the world
Language can only give a feeling about the world, not enough detail — which is why we have poets
37:31 20VC: Why Cursor is Dead | An AI Tsunami is Coming & You Need to Prepare | Systems of Record Become Valueless Databases with Agents | Is This The End of Tech Private Equity with Jerry Murdock, Co-Founder of Insight Partners
Also on the record
Jonathan Ross · Sep 29, 2025
AI arriving as language-based rather than esoteric intelligence is what made it universally usable, which is why AI came later than he expected but is growing far faster
Language makes interaction trivial so anyone can use it, unlike an Alpha Go-style esoteric intelligence
40:05 Language based interface drove universal adoption explaining late but fast growth
Mati Staniszewski · Sep 8, 2025 · speculative
As the different modalities combine, models will gain a much deeper understanding of everything around them
13:18 Multimodal combination deepens world understanding
Alexander Embiricos · Feb 21, 2026
Agents that work with your computer through code, not video and audio multimodal interfaces, are the way AI benefits reach people
Progress in multimodal models (video/audio) was slower than he expected, while code-based agents advanced
54:23 Code based computer use beats multimodal video and audio agents
Your assistant can query this graph directly — 7 positions here, 19,646 across the corpus. Add 996.fm over MCP.