Key Takeaways
- Your current AI models, even public ones like GPT-5.5, hold significant untapped "latent capabilities." OpenAI recently disproved the complex Erdos unit distance conjecture internally, a feat Noam Brown believes GPT-5.5 could mimic with roughly $100,000 in dedicated compute.
- This deep potential often goes unexplored because rapid model release cycles foster a “wait for the next model” mentality. Brown observes this leads to developers asking, “why bother doing any engineering work when I should just wait for the next model?”
- This short-sighted approach means models “exist for a very short context window and then they just like disappear,” without their full potential being discovered or built upon.
- The next major leap for AI isn't just bigger models, but coordinated multi-agent systems that share and build knowledge. Brown sees early signs of this in tools like Multbook and Open Claw, foreshadowing a future mirroring human collective progress.
The Hidden Power in Your Current AI Stack
Consider a problem that stumped mathematicians for decades: the Erdos unit distance conjecture. An OpenAI model, not some future AGI, but one existing today, disproved it. This wasn't a fluke. OpenAI research scientist Noam Brown points to this as clear evidence of "significant latent capabilities" already present in current AI models.
Brown pushes this further for founders: even publicly available models like GPT-5.5 likely hold similar potential. It's not about a new model version, but about commitment. He argues, “nobody had explored sufficiently what happens if I put $100,000 worth of compute into 5.5 what could it do? Um and the answer is like yeah you probably could get stuff like that out of it.” This means the raw intelligence is there. Your current AI stack probably holds more power than you realize, if you're willing to invest in deep exploration and targeted compute.
This isn't just about massive research labs. The core insight applies to anyone building with AI. The capabilities aren't always on the surface; they emerge from dedicated effort to push existing models beyond their standard use cases. You might be ignoring a valuable asset already at your fingertips.
The "Next Model" Trap
Why do these deep capabilities go unexplored? Brown highlights a common industry trap: waiting for the next big release. “you've seen the meme that's like oh I like why why bother doing any engineering work when I should just wait for the next model go on vacation come back two months later and and it's, you know, a thousand times cheaper,” he says. This constant expectation of better, cheaper models around the corner subtly discourages maximizing current versions.
This mentality creates a peculiar cycle for AI models themselves. Brown notes, “We're not seeing that with AI models today. They kind of they're they're born into a world for and they exist for a very short context window and then they just like disappear.” Unlike human knowledge, which compounds across generations, AI models often live short, isolated lives. Their full, deep potential remains undiscovered, replaced by the next iteration before it's truly understood.
For builders, this is a missed opportunity. While new models promise efficiency, they also hide the true, expensive-to-uncover power within existing systems. Betting solely on future releases means you might be leaving significant, immediately accessible value on the table.
Beyond Single Agents: Building Collective AI
The real long-term game, Brown suggests, shifts from individual, powerful models to coordinated AI systems. He envisions a future where AI agents share knowledge, build on each other's work, and progress collectively, much like human civilization. Brown states, “I do think eventually we will and we're starting to see like signs that we're entering a world where they can coordinate on a large scale.” He points to early, if overhyped, examples like Multbook and Open Claw as glimpses of this multi-agent future.
For founders, this means thinking beyond isolated AI tools. Future value may come from architecting systems that enable multiple agents to work together. Instead of a single AI solution, imagine a distributed network of specialized AIs collaborating on complex problems. This reframes the challenge from raw model power to intelligent system design.
What to Do With This
Don't default to "wait for the next model" when facing a hard problem. This week, pick one significant, recurring bottleneck in your current AI pipeline. Instead of looking for a new model, dedicate a small budget—say, $500 to $1,000—to intensely explore what your existing model can do if pushed. This means aggressive prompting, advanced fine-tuning, or temporary, compute-heavy scaffolding to see if you can solve that problem, acting as if the next model release isn't an option.