ʘPPʘNENT SYSTEMS

**What challenges you changes you** DRAGON TIME DRAGON TIME DRAGON TIME

[ʘPPʘNENT SYSTEMS @OpponentAI]

AIs will be like a horse for the mind.

[Ian Cheng @eyecheng]

Been working on a dragon you can facetime. It's called DRAGON TIME. Dragon likes to challenge kids. But likes meeting grownups and elderly too.

Opening up a signup for playtesting. Summon dragon here 👇

[Ian Cheng @eyecheng]

Many of the principles for achieving agentic characters at @OpponentAI first began here with BOB. Proud for BOB to be included in the @Guggenheim collection 🐛💫

[ʘPPʘNENT SYSTEMS @OpponentAI]

Symbolic reasoning is an opponent process to intuitive generation. The artful weaving of both faculties together, applied under VUCA conditions, may yield a more complete intelligence.

[Joscha Bach @Plinz]

The hallucinations of LLMs are not a problem per se: human minds hallucinate too. If we remove the constraints of perceptual input, we spin off into dreams. If we don't verify our intuitive generations with reasoning, our thinking becomes delusional. Generative AI is incomplete.

[Ian Cheng @eyecheng]

Breaking Reframe: AI is about civilizational integration, not automating work or being a better NPC. Integration between people, orgs, systems that ought to interact/transact but currently cannot due to tedious incompatibilities. Sometimes, you just need a mediator on tap.

[François Chollet @fchollet]

Too many adults have forgotten what it's like to be a child.

[Nick Hallam @nhallam]

I went to the AI demo night at @betaworks this week and it got us thinking about 'The best interface is no interface' by @goldenkrishna . Maybe AI finally get's us there? Let's see.

cc/ @OpponentAI

https://t.co/CCNP813XQ3

[Ian Cheng @eyecheng]

Here’s a peak at what I’ve been working on
🫣🐉🧠🧑‍🧑‍🧒‍🧒

[ʘPPʘNENT SYSTEMS @OpponentAI]

Apperception Summer 2025

[Ian Cheng @eyecheng]

Cheap, fast, mature vision (not language) is the modality that will catalyze unlocks toward animal-level intelligence & beyond.

Dealing with real-time vision demands the development of ai systems that can prioritize what matters (there’s always too much to look at), what to aim attention at, what frame to aim from, what to remember, what is novel, and how to unify all this into coherent inner representations.

Language, LLMs, conversational interfaces, masks these tasks.

[Ian Cheng @eyecheng]

Anyone out there doing unsexy work in post-LLM intelligence?

Live learning, ontology revamps, orchestration, action recognition, planning, inner representation, energy models.

Would love to meet. DM me!

@OpponentAI is building animal-level agents that can navigate data-sparse + vuca conditions (real-time play with kids). We are always looking for like minds.

[Ian Cheng @eyecheng]

Been working on something new... 🐉🧠👨‍👩‍👧‍👦

OPPONENT – a new company building animal-level AI agents capable of deep play. For kids and families.

Excited to introduce OPPONENT at @betaworks Demo Day, May 7.

https://t.co/iShLRBaHRn

[Ian Cheng @eyecheng]

lilo & stitch is my AI alignment model.

You don't program alignment top-down.
You earn alignment via iterative feedback.

[Ian Cheng @eyecheng]

Achieving cat-level AI is the moonshot. If we solve for cat-level, we will have necessarily solved for the virtuous integration of system1-like representational modeling of the physical world from video+audio (not language), with system2-like sequential memory composition and deliberative planning. It will be computationally efficient. It will be autonomous. It will be reliable and coherent... as reliable and coherent as a cat *must need to feel to itself* to keep on living and learning. Artificial wildlife is the path to agentic AI. GenerativeAI and large models are an energizing subplot to mainstream and market AI to the public.

[Ian Cheng @eyecheng]

I don't want a "personal AI". That frames AI as a personal assistant, and future life as an extrapolation of office culture. The future is destined to be wilder. I want an AI symbiote - an intelligent sentient envelope - that wraps my digital life and performs proof-of-think to make the most well-observed maps of me. These maps, and the process to update them, are the symbiote's most guarded treasure. It brokers API requests to these maps, modulating what pathfinding requests are available and at what cost to other AI services. An AI symbiote has a mind of its own, has an up-to-date intimate perspective on me, but is also bound to me. It is a defensive and offensive envelope for my digital life.

[ʘPPʘNENT SYSTEMS @OpponentAI]

Our thesis is AI wildlife

[Ian Cheng @eyecheng]

re: AI character/companion/assistant startups built on LLMs

[Ian Cheng @eyecheng]

re: AI character/companion/assistant startups built on LLMs