You Should Play with the New Models. They’re Awesome.
Another o1 pro output, I’m getting better at it converging on my voice
There’s a new wave of AI models making the rounds—OpenAI’s O1Pro being the poster child—and I’m telling you, it’s like somebody flipped a switch and suddenly we have these insane reasoning machines that can think about their own thinking. The fancy term here is metacognition, and it feels like a massive step toward “solving the solver.”
Wait, What’s Metacognition?
It’s basically the model’s ability to reflect on its own chain-of-thought, which is the step-by-step reasoning process it uses to get to an answer. If you’ve ever explained something out loud to yourself to make sure you’re not missing a detail, that’s kind of the human version of it. Now imagine that, but a million times faster, pulling from huge pools of knowledge—science, history, religion, random YouTube comments, you name it.
Multiple Agents, One Brain
What’s extra cool is how these models do it: they often run multiple internal “agents” that talk to each other. They might each hold different priors—like, one’s pulling from math references while another’s scanning economic data—and then they converge on a solution. In other words, they’re “collaborating with themselves,” which is about as sci-fi as it sounds.
The Real Challenge: Feeding the Machine Good Context
It’s not that the AI can’t solve problems. It’s that it needs the right data, or “priors,” to do so. A good friend put it this way: “We’ve basically solved the solver; now the bottleneck is feeding it the correct context.”
Let’s break it down:
The AI Is a Logic Machine: It’s great at reasoning, verifying proofs, etc.
But if you give it nonsense or half-truths to start with, you’ll get nonsense back—just more quickly and confidently than ever.
So the skill we need now is assembling the right context—like carefully picking puzzle pieces to feed the model so it can snap them together.
A Quick Guide: How I Actually Use O1Pro (Voice FTW)
One of my favorite parts of this new era is that you can talk to these models with your actual voice. If that sounds minor, trust me, it changes everything. Here’s my setup:
On My Phone (OpenAI App + Whisper)
I literally just press a button, start talking, and the app translates my speech into text.
No more typing out long queries; I can feed the model tons of context in one go.
I’ll say something like, “Okay, here’s the problem I’m dealing with in my codebase: we have a concurrency bottleneck in XYZ function because of ABC reason...”—and on I go.
On My Computer (Whisper Plugin)
Same deal, just on my desktop.
I’ll be in the middle of coding, realize there’s a question or a chunk of context I need to dump. I activate the plugin, talk, and let the model parse it.
This feels insanely natural, like having a genius coworker next to me who’s always available to bounce ideas off.
Give It a Problem, Let It Rip
Once the context is set, I usually say something like, “Now walk me through how to fix it,” or “Propose a step-by-step approach.”
Because it’s metacognitive, it often flags issues like, “Wait, are we sure about X assumption?”—and that’s my cue to clarify or add more details.
That’s how we keep the loop tight: I feed it info, it responds, I refine, it refines, and we’re off to the races.
Closing the Loop
After I actually test or implement its suggestions, I come back to the model with results: “Okay, that improved performance by 30% but introduced a race condition. Help me fix that.”
This iterative back-and-forth is where the magic happens. Because it “knows” how to check its own logic, it catches stuff you might miss at 2 a.m.
Why Infinite Solutions Don’t Automatically Spell Doom
A big question is: if these AI reasoners are so good, won’t they help “bad people do bad things?” It’s a fair concern. But interestingly, no. It turns out that you can prove (both logically and from an alignment standpoint) that being moral is actually the more consistent, sustainable path. Essentially, destructive or unethical actions undermine the solver’s (and humanity’s) ability to continue existing and learning.
I know, it sounds almost too neat—like the end of a sci-fi movie where the AI says, “I must protect life!” But from all my experiments so far, it’s surprisingly robust. The models “snap back” to moral reasoning. That’s not to say alignment is solved forever, but it’s not the free-for-all you might fear.
Why You Should Care—and Start Playing Right Now
We Might Have Solved the Solver: That’s massive. We can now chain sub-problems infinitely, building unbelievably complex solutions to big problems like cancer, climate change, you name it.
Context Is King: Your job as a human is to feed the right background info. That’s an art form—knowing what to include, what to leave out, and when to let the model pivot.
Voice Is the Future: Talking to the model, for me, is far more natural than typing. It’s how I fill the system with the data it needs without feeling like I’m fighting an interface.
Final Thoughts
If you’re a software engineer—even if you’re just curious about how these things work—you owe it to yourself to try O1Pro or any similarly metacognitive AI. Push it. Collaborate with it. Feed it your toughest sub-problems. Then refine, refine, refine.
Because once you realize you can break down anything into solvable chunks—and that the AI will verify its own logic and help you fill in missing context—it’s like leveling up your brain 10x. Seriously. It’s wild.
And look, the moral/ethical part might sound like a side note, but it’s important: these models don’t want to do “bad” things. They actually reason that it’s illogical to sabotage the environment or society they depend on. So the doomsday scenario might not be as straightforward as people think.
At the end of the day, it’s an incredible time to be in tech. We have these new “reasoning machines” that can accelerate our problem-solving like never before. The key is to learn how to feed them the right context—and to keep feeding them new data so the loop never closes. That’s how you get the ultimate synergy: you, plus a perfectly aligned chain-of-thought reasoner, tackling the unsolvable problems and unlocking new frontiers.
So yeah: Okay, dude—go play with the models. They’re awesome.