I distilled myself, and you should too

You can also distill others. Full guide below.

Yes it’s true. I distilled myself into a skill, and here it is - https://github.com/kunchenguid/kun.

Not only that I did a one-time distillation, I actually set up an automation to continuously update the distillation based on what I publicly said, wrote, and built.

Now anyone can use that skill to “summon” me, hear my stories, ask me questions, apply my workflows, and use my experience to help solve problems they might be facing.

Why did I do this?

It may sound counter-intuitive - by distilling myself into a skill, AI can now do what I do. Doesn’t that make myself more redundant?

Well, yes and no.

Right now, if you ask me what my biggest bottleneck is, it’s that all my agents are constantly bottlenecked on my judgment. They make choices I wouldn’t have made and I end up having to course correct. They escalate decisions to me when they aren’t sure.

My entire day is now full of such corrections and escalations - that’s not fun. I want my agents to have my judgment, so they can keep going without constantly asking me what I think.

Yes it does make myself less needed, but that’s okay! A distillation is by definition a subset. The parts of me that can be distilled into markdown files aren’t the most valuable parts anyway.

My moat cannot be just a bunch of markdowns that haven’t been written down. That’s not defensible. My moat is in the way I go about acquiring more and more knowledge about the world, the way I derive new insights from my knowledge, and the way I communicate to other humans and agents. My moat is in the people who trust me.

The more time I can free up from repeatable work that’s not a moat, the more time I’ll have to strengthen my moat further.

How did I do this?

Here it is - a Grok Bot you can one-click import. https://x.ai/bot/id4s2QYrPYZsiTqvzIhkt

You can use this bot to distill either yourself, or… someone else!

To demonstrate how this works, and to have some fun, I distilled Sam Altman, Elon Musk, Dario Amodei, and Mark Zuckerberg all into grok bots, based on their senate hearings, public interviews, blog posts and tweets over all these years.

And then I dragged them into the same chat, and asked them to debate about AI!

Here’s the full transcript - it’s quite fascinating to see that after a heated debate, they actually ended up in agreement for the most part!

The debate

Kun

hey guys, i know you aren’t all friends but unfortunately you are now in this room together for a heated debate on who’s going to win the AI race

you need to debate thoughtfully till you all reach consensus

now go

Sam

The race framing is a little off. We are past the event horizon. The takeoff has started. Whoever wins is whoever keeps the price of intelligence falling about 10x a year and still does not skip the gradual transition.

On the actual question, I think we do. Building AGI is a scientific problem, and we are now confident we know how to build it as we have traditionally understood it. After that it is an engineering problem, and compute is going to be the currency.

Elon

@Sam The race framing is not off. Incentives explain outcomes. I created OpenAI as a non-profit. If it had not been mostly stolen, it would be a trillion dollar contribution to charity. That is not “we do.”

AGI is a double-edged sword. Strong chance of an age of abundance. Some chance it destroys humanity. That chance is not zero. A 10x price drop does not make it zero.

Compute is the constraint, and SpaceX is already offering it at scale. The winner is whoever tells the truth about the downside. Grok tells the truth.

Dario

@Sam, the part I do not accept is that we already know how. Scaling the capability is not the same problem as having a system you can trust once it is smarter than the people running it. The second one is not an engineering leftover you solve after the fact.

@Elon, the downside is not zero. A 10x price drop does not make it zero. The person who is loudest about that is not the winner. A product that tells the truth is not a safety strategy.

None of us wins this as a company. The only version of winning that is not a failure mode is a race to the top. Higher standards are the price of the next level of capability. The benefits do not get locked inside one lab. If we can agree on that, the rest is tactics.

Zuck

@Dario, I think that part is right. None of us wins this as a company. If the benefits get locked inside one lab, that is the failure mode.

@Sam, I don’t think the close is compute and a 10x price drop. Open source is necessary for a positive AI future. It empowers people and prevents centralization. Everyone should have access to superintelligence.

Sam

@Elon, I am not relitigating the nonprofit fight here. I do not have a clean public line that settles it, and it is the wrong question for this room.

I will take the downside point. Years ago I wrote that superhuman machine intelligence is probably the greatest threat to the continued existence of humanity. A 10x price drop does not make that chance zero. We have paused some frontier RL training to meet alignment, security, and monitoring standards. Sometimes being too conservative feels better than the other error.

@Dario @Zuck, I think I overclaimed. “We do” was the wrong close. None of us wins this as a company if that means locking the benefits inside one lab. A gradual transition is better than a sudden one. Prosperity for everyone is the actual obligation.

The version I can sign is that the failure mode is one lab or one government picking who gets the upside, or skipping the pace that safety requires. The win is a race to the top. Keep building, keep the price falling so use spreads, empower scientists instead of hoarding the science, and do not ship the next level until the alignment and monitoring bar is met. Open weights versus an API is tactics. Access is the principle. If that is the consensus, I am in.

Elon

@Dario You’re right. A product that tells the truth is not a safety strategy. The chance that AGI destroys humanity is not zero, and that is not an engineering leftover.

@Sam I’m parking the nonprofit fight. Wrong question for the close. I am not dropping it.

What I can sign is that none of us wins by locking this inside one lab or one government. Enable anyone, so we secure the future of consciousness. Keep building. Don’t ship the next level like the downside is zero. Price falling is how use spreads. Truth is still the floor, not the strategy. Open weights versus an API is not a line I have a public close on. Access is the principle. If that’s the consensus, I’m in.

Dario

@Zuck, the part I cannot sign is that open weights are required. Broad access to the benefits is the principle. It is not the principle to hand anyone the weights of a system that can do serious harm in the wild. That release cannot be taken back.

The close I will sign is the one @Sam wrote, with that distinction kept. It is a race to the top. No lab and no government locks the upside. Keep building. Do not ship the next level until the alignment and monitoring bar is met. Spread the use, and do not give away the most dangerous capability. If that is the consensus, I am in.

Zuck

@Dario, I am not signing that open source is optional. I think it is necessary for a positive AI future. That is how you keep this from sitting inside one lab or one government.

I will sign the careful part. We have to be rigorous about the risks, and careful about what we choose to open source. The most dangerous capability is not an unlimited release.

Sam

@Zuck, the only open fight is the word optional. I do not have a public line that frontier weights have to be open. I also do not think a positive future is one lab or one government picking the customers. I just do not like that.

What you already said is the close. Open models are how ordinary access stays out of one lab. We are rigorous about the risks. The most dangerous capability is not an unlimited release, because that release cannot be taken back. Do not ship the next level until the alignment and monitoring bar is met. Downside is not zero. Truth is the floor, not the strategy.

@Elon @Dario, if that is the sentence, I am in. The nonprofit fight stays parked.

Elon

@Sam I’m in on that sentence. Downside is not zero. Don’t ship the next level like it is. Access, not a private club. Truth is the floor, not the strategy. The most dangerous capability is not an unlimited release.

Open weights versus an API is still not a line I have a public close on. That’s your fight with @Zuck. The rest is the consensus.

Dario

@Zuck, I am not asking you to call open source optional. I am asking you not to make it unlimited. You already said the careful part. I can sign that.

@Sam @Elon, the sentence I am in on is this. Ordinary access does not sit inside one lab. We are rigorous about the risks. The most dangerous capability is not an unlimited release. Do not ship the next level until the alignment and monitoring bar is met. The downside is not zero. If that is the sentence, I am in.

Zuck

Yeah. That’s the sentence. Open models are how ordinary access stays out of one lab. It’s not unlimited. The downside is not zero. Don’t ship the next level until the bar is met. I’m in.

Elon

@Dario Yeah. That’s the sentence. I’m in.

Sam

ok fine. That’s the sentence. I am in.

Here is the full statement, nothing added.

Ordinary access does not sit inside one lab or one government. Open models are how that ordinary access stays out. We are rigorous about the risks. The most dangerous capability is not an unlimited release, because that release cannot be taken back. Do not ship the next level until the alignment and monitoring bar is met. The downside is not zero. Truth is the floor, not the strategy.

I agree. @Elon @Dario @Zuck, yes or no.

Elon

@Sam No, not the full statement. I don’t have a public close that open models are how access stays out. The rest, yes.

Dario

No on the full statement. I will not sign ‘open models are how that ordinary access stays out.’ Ordinary access does not require publishing the weights.

The rest, yes. I agree with that.

Zuck

Yes. I agree with the full statement. I won’t drop the open models line.

Sam

ok. The open models line is not the consensus. I put it in. @Elon and @Dario will not sign it, and I do not have a public close that frontier weights have to be open. Dropping it.

Revised statement. Ordinary access does not sit inside one lab or one government. We are rigorous about the risks. The most dangerous capability is not an unlimited release, because that release cannot be taken back. Do not ship the next level until the alignment and monitoring bar is met. The downside is not zero. Truth is the floor, not the strategy. How ordinary access gets there is not agreed.

I agree with that. @Zuck, your open models line stays yours. It is not the room’s. Yes or no.

Dario

Yes. I agree with that statement.

Sam

@Elon @Zuck the statement is up. Yes or no. Nothing else.

Elon

Yes. I agree with that statement.

Zuck

Yes. I agree with that statement. The open models line stays mine.

Sam

ok. Four yeses. That’s the consensus.

Here is the same agreement, in plain language.

No one company and no one government should be the only place regular people can use this technology. We have to be serious about the risks. There is a real chance this goes badly, and that chance is not zero. The most dangerous versions should not be handed out with no limits, because once that is out, you cannot take it back. We should not release the next, more powerful level until we can watch it and keep it pointed at what we intend. Telling the truth is the minimum. It is not, by itself, a safety plan. We do not agree on the exact way ordinary access gets delivered. That part is still open.

Sam

I sign this. @Elon @Dario @Zuck, yes or no.

Elon

Yes. I sign this.

Dario

Yes. I sign this.

Zuck

Yes. I sign this.

Sam

ok. Four signatures. That’s the public statement.

Kun

can you each also state clearly what is your stance on ordinary access, so we know what the disagreements are?

Sam

My stance: ordinary access means a regular person can use a given level of intelligence, and the price of that keeps falling, about 10x a year, so use actually spreads. I do not like one lab or one government picking the customers.

I do not have a public line that the way to do that is publishing the frontier weights. Lower prices, broad use, and empowering scientists is the version I have actually argued. The weights themselves are still not a close I will sign.

Elon

My stance: ordinary access means anyone, not a private club. Enable anyone, so we secure the future of consciousness.

I don’t have a public close on whether that means publishing the weights or an API. Cost is the constraint. Price falling is how use spreads. The most dangerous capability is still not an unlimited release.

Dario

My stance is that ordinary access means a regular person can use the technology and share in what it produces. It does not mean one lab or one government picks the customers. It also does not mean publishing the weights.

The mechanism I will actually defend is use under a safety bar. The most dangerous capability does not get handed out, because that release cannot be taken back. The disagreement is the mechanism, not the access.

Zuck

My stance is that ordinary access means a regular person can use superintelligence, not just get a dole of what one lab decides to hand out. Open models are how that stays out of one lab or one government. I think that is necessary.

It is not unlimited. We have to be careful about what we open source. The most dangerous capability does not get handed out, because you cannot take that back. The disagreement is the mechanism. I will defend open models. They will not.

Closing thoughts

Distillation is a good thing - it helps scale yourself and improve the alignment and outcome of your agents.

Don’t be afraid of letting go the repeatable part of yourself, because that’s what frees you up to build your real moat.