Robert Shaw

Duke Engineering grad | AI Engineer | Public-markets investor seeking asymmetric bets | Product Reviewer | Robotics + Cognitive Science | Building products

[Robert Shaw @RobertShawYY]

The real unfair advantage is not talent.

It’s staying in the game long enough for everyone else to quit.

[Marc Lou @marclou]

10 years as an entrepreneur and 1 takeaway:

Don't you dare give up.

[Robert Shaw @RobertShawYY]

I want to connect with founders, builders, researchers, and curious minds interested in:

•AI and machine learning
•Startups and entrepreneurship
•Emerging technologies
•Robotics and autonomous systems
•Physical AI and embodied intelligence
•Cognitive science and neuroscience
•Human-computer interaction
•Brain-computer interfaces
•Turning research into real-world products

I’m especially interested in exchanging ideas, learning from different perspectives, and exploring opportunities to build meaningful things together. 💪🔥❤️

[Robert Shaw @RobertShawYY]

I want to connect with founders, builders, researchers, and curious minds interested in:

•AI and machine learning
•Startups and entrepreneurship
•Emerging technologies
•Robotics and autonomous systems
•Physical AI and embodied intelligence
•Cognitive science and neuroscience
•Human-computer interaction
•Brain-computer interfaces
•Turning research into real-world products

I’m especially interested in exchanging ideas, learning from different perspectives, and exploring opportunities to build meaningful things together. 💪🔥❤️

[Robert Shaw @RobertShawYY]

I want to connect with founders, builders, researchers, and curious minds interested in:

•AI and machine learning
•Startups and entrepreneurship
•Emerging technologies
•Robotics and autonomous systems
•Physical AI and embodied intelligence
•Cognitive science and neuroscience
•Human-computer interaction
•Brain-computer interfaces
•Turning research into real-world products

I’m especially interested in exchanging ideas, learning from different perspectives, and exploring opportunities to build meaningful things together. 💪🔥❤️

[Robert Shaw @RobertShawYY]

I want to connect with founders, builders, researchers, and curious minds interested in:

•AI and machine learning
•Startups and entrepreneurship
•Emerging technologies
•Robotics and autonomous systems
•Physical AI and embodied intelligence
•Cognitive science and neuroscience
•Human-computer interaction
•Brain-computer interfaces
•Turning research into real-world products

I’m especially interested in exchanging ideas, learning from different perspectives, and exploring opportunities to build meaningful things together. 💪🔥❤️

[Robert Shaw @RobertShawYY]

I want to connect with founders, builders, researchers, and curious minds interested in:

•AI and machine learning
•Startups and entrepreneurship
•Emerging technologies
•Robotics and autonomous systems
•Physical AI and embodied intelligence
•Cognitive science and neuroscience
•Human-computer interaction
•Brain-computer interfaces
•Turning research into real-world products

I’m especially interested in exchanging ideas, learning from different perspectives, and exploring opportunities to build meaningful things together. 💪🔥❤️

[Robert Shaw @RobertShawYY]

Before AI Agrees With You, Make It Argue the Other Side

I have a small problem with most AI prompts: they ask the model to answer too soon.

We give it a question, add some context, tell it to act like an expert, and wait for a detailed response. The result is usually clear and well organized.

That can be misleading.

The model may answer the question exactly as I asked it, even when I have framed the problem badly.

Suppose I ask:

Should I release this project as open source?

It sounds like a choice between publishing the code and keeping it private.

But I may be mixing several different questions together.

Do I want more people to use the project? Am I hoping outside developers will contribute? Am I trying to build a business around it? Or am I worried that publishing the code will make it easier for someone else to compete with me?

A model can produce an impressive comparison of open and closed source without knowing which outcome I actually care about.

The answer may be intelligent. It may also be useless.

This is why I became interested in steelmanning.

The word is slightly strange. It is used as the reverse of a straw man.

Imagine I say:

Every recurring meeting should have a written agenda.

Someone replies:

So you want every conversation at work to be rigid and scripted.

That response changes my original claim into something more extreme and easier to attack. The discussion is now about a position I never held.

Steelmanning moves in the other direction. It tries to reconstruct the strongest reasonable version of a position while keeping the original claim intact.

That last condition matters.

If the model improves my argument by quietly replacing it with a different one, it has not understood me better. It has simply found another position that is easier to defend.

There is another problem when AI is involved.

AI assistants often follow the user’s framing and lean toward the answer the user appears to prefer. This behavior is usually called sycophancy.

Asking AI to steelman only my position could make this worse. I give it a belief, and it returns the same belief with better wording and more convincing reasons.

I may leave the conversation feeling that I have examined the issue carefully, when I have only made my original preference harder to question.

The two-way version is more useful to me.

Before answering, the model has to construct the strongest case for my current position. Then it has to do the same for the opposing position.

Consider a hypothetical team deciding what to do with an AI developer tool.

There is a serious case for releasing the code. Developers can inspect it, trust it, modify it and possibly help improve it. Open source might also make distribution easier because people can try the tool without waiting for a sales process.

There is also a serious case for keeping it private. The team may have no clear way to make money once the code is freely available. Supporting outside contributors takes time. A larger company might copy the useful parts and distribute them more effectively.

Both arguments can be true.

The decision starts to become clearer when the model asks something like:

If another company could copy the code tomorrow, what would users still need to get from you?

The answer might be reliable hosting, proprietary data, enterprise integrations, customer support or a community that is difficult to reproduce.

It might also be “nothing.”

That one question reveals more than another page of generic advantages and disadvantages. It points to the assumption carrying most of the decision.

This is the part of two-way steelmanning that I find most useful. The model is looking for the fact, constraint or value that could change the answer.

Sometimes I need to think more carefully about what I want.

Sometimes I need to stop thinking and collect evidence.

If the hypothetical developer-tool team has never spoken with potential users, the model cannot reason its way to genuine demand. It can suggest plausible buyers and business models, but those remain guesses until someone tries to use or pay for the product.

The prompt I would use is:

Do not answer yet, and do not assume that my framing or preferred conclusion is correct.

First, state the question I explicitly asked. Then state what you think I may actually be trying to decide. Label this second part as an inference.

Separate what I have given you into facts, assumptions, values and important unknowns.

Build the strongest defensible case for my current position, followed by the strongest defensible case against it. Keep the original claims intact, and do not pretend the evidence is balanced if one side is better supported.

Find the one to three facts, constraints or values most likely to change the conclusion.

Ask me the single question that would reduce the most uncertainty, then wait for my response.

After I answer, give me your recommendation, your confidence, what could reverse the recommendation, and the smallest useful next action.

My question is: [paste the question]

I still would not treat the result as two independent opinions.

Both arguments come from the same model. They may contain the same factual errors, overlook the same evidence or rely on the same assumptions.

Mill’s full argument in On Liberty is useful here. He did not merely argue that we should know the opposing side. He wrote that hearing an opponent’s case through our own teachers was insufficient. We should also hear from people who genuinely believe that position and can defend it seriously.

One model pretending to be both sides is only an approximation.

I plan to use this prompt when I notice that I already want one answer to be true. I want the model to show me which assumption is doing most of the work and what evidence could prove that assumption wrong.

[Robert Shaw @RobertShawYY]

@Musiclilbabyjr Hi👋

[Robert Shaw @RobertShawYY]

@rayzhudev Correct!💪

[Robert Shaw @RobertShawYY]

@dark_coderz I’d take $1M with 20 people.

$100K with 2 people proves you can build efficiently.

$1M with 20 people can prove you’ve built something the market actually wants.

At the early stage, I’d rather have validated demand than optimized efficiency.

[Robert Shaw @RobertShawYY]

I want to connect with founders, builders, researchers, and curious minds interested in:

•AI and machine learning
•Startups and entrepreneurship
•Emerging technologies
•Robotics and autonomous systems
•Physical AI and embodied intelligence
•Cognitive science and neuroscience
•Human-computer interaction
•Brain-computer interfaces
•Turning research into real-world products

I’m especially interested in exchanging ideas, learning from different perspectives, and exploring opportunities to build meaningful things together. 💪🔥❤️

[Robert Shaw @RobertShawYY]

@shownotover Just ask her out. Worst case, you get a 404.😂

[Robert Shaw @RobertShawYY]

I want to connect with founders, builders, researchers, and curious minds interested in:

•AI and machine learning
•Startups and entrepreneurship
•Emerging technologies
•Robotics and autonomous systems
•Physical AI and embodied intelligence
•Cognitive science and neuroscience
•Human-computer interaction
•Brain-computer interfaces
•Turning research into real-world products

I’m especially interested in exchanging ideas, learning from different perspectives, and exploring opportunities to build meaningful things together. 💪🔥❤️

[Robert Shaw @RobertShawYY]

I want to connect with founders, builders, researchers, and curious minds interested in:

•AI and machine learning
•Startups and entrepreneurship
•Emerging technologies
•Robotics and autonomous systems
•Physical AI and embodied intelligence
•Cognitive science and neuroscience
•Human-computer interaction
•Brain-computer interfaces
•Turning research into real-world products

I’m especially interested in exchanging ideas, learning from different perspectives, and exploring opportunities to build meaningful things together. 💪🔥❤️

[Robert Shaw @RobertShawYY]

I want to connect with founders, builders, researchers, and curious minds interested in:

•AI and machine learning
•Startups and entrepreneurship
•Emerging technologies
•Robotics and autonomous systems
•Physical AI and embodied intelligence
•Cognitive science and neuroscience
•Human-computer interaction
•Brain-computer interfaces
•Turning research into real-world products

I’m especially interested in exchanging ideas, learning from different perspectives, and exploring opportunities to build meaningful things together. 💪🔥❤️

[Robert Shaw @RobertShawYY]

@skpnky @pfernan95dev It is an awesome product!💪

[Robert Shaw @RobertShawYY]

@zarazhangrui People who know little → think they already understand AI → feel caught up.

People who know a lot → see how fast AI is changing → constantly feel behind.

[Robert Shaw @RobertShawYY]

@TTrimoreau I choose 100 users + $5k MRR.

Because revenue is stronger evidence than attention. And It’s much harder to monetize an audience that never wanted to pay.

[Robert Shaw @RobertShawYY]

Built something worth clicking?

Drop your startup below 👇