I am human and I often waver between healthy skepticism and optimism about LLMs. I try to test claims, question demos and be honest about what breaks. And, I am also too quick to judge work I do not understand.

My biggest learning from the mobile and social era of technology is that it’s easy to (mis)attribute foresight to people working on cutting edge technology. I remember being close to the technology and really understanding its fortes and foibles. However, that knowledge never translated to additional insight into areas where the tool was deployed: journalism, medicine, law, education, biology, art, or government.

For work, I often thought about how a tool might influence a field, especially as a consuting technologist in my early days. However, one thing that I had to keep relearning was knowing the tool does not automatically translate to understanding the field it enters.

I have made that mistake in many ways. I looked at the final thing someone produces and assumed I understood the work behind it. An article. A memo. A contract. A lesson plan. A diagnosis. A press release. A piece of software. The classic: If you “just” change $small-thing, it’ll be all good.

The hidden work behind the artifact was always hard to see and never obvious. I never appreciated that until actually talking to the people involved and growing older (and hopefully wiser).

A software engineer doesn’t just produce code that runs once. Similarly, a journalist does not just type sentences; a lawyer does not just summarize documents; a doctor does not just match symptoms to conditions; a teacher does not just explain a topic.

Most fields involve decisions, expertise and nuance. And of course there’s waste, internal politics, habit, personal and personnel flaws. The trick is knowing and differentiating why some wastage is justified because experts hit the failure mode that an outsider has not imagined yet.

To me, this is the biggest concern with LLMs. LLMs make the outside view more seductive because they are really good at producing artifacts.

They can write a press release. They can generate a functional app. They can summarize a contract. They can draft an article. They can make a lesson plan. They can make mostly functional software.

That is impressive.

It also woos us into a bad conclusion: if the output is easy to imitate, maybe the field was easier than we thought.

I am noticing this pattern in the AI industry.

Source: Eric Jorgenson on Twitter

A journalist vibe codes an app and wonders how LLMs are going to revolutionize software engineering.

A biologist asks an LLM for a press release and wonders how much communications work really requires.

A software engineer watches an LLM summarize legal text and decides that lawyers are toast.

I am not poking fun. It’s human to fall into this trap and underestimate the taste, maintenance, sourcing, accountability, timing, and the small decisions that separate acceptable work from good work and the judgement that separates good work from the great.

I still want and support skepticism especially when it’s mixed with optimistic gumption. To the note of wastage above, some experts do hide behind complexity. Some fields protect bad norms. Some institutions deserve hard questions. Some work will be changed by LLMs in ways insiders do not want to admit. Blindspots are eveywhere.

Yet, I want my skepticism to start with curiosity, especially when I am outside my field. I strive to start with: “What do they know that I cannot see yet?”

That question slows me down. It makes me ask why the work is shaped the way it is. It makes me more careful about calling something obsolete because I can now imitate one visible part of it.

Permit me to wax nostalgic about the early internet.

The early internet gave me access to people who explained problems and solutions. Forums, blogs, mailing lists, early Twitter, random personal sites. People wrote down what they knew. They (usually)answered questions from strangers sans snark.

That internet was never innocent. There was plenty of ego, cruelty and bad faith. I am not asking you to be polyannish. But, a lot of the culture I valued and luckily got to see was built around being useful.

Over time, too much of online life shifted toward performance: Mock a weak claim; get rewarded. Write take down posts and earn kudos. I can admit that there is a morbid entertainment factor to it. I know I’ve rewarded such “hot takes” with both attention and support.

However, I feel like we’ve taken that to its fanatical end. For whatever reason, today, slow explanations are branded boring; tarning anyone asking a sincere question as naive. We also reward these performances with attention, which is the new currency everyone’s seeking.

Let me repeat, I am not immune to this pull of judging.

A brief aside on judgement: A judgement means that one can stop thinking about it. It means I’ve reached a conclusion. My brain can move on to another thing after a satisfying sharp line. It’s the easy thing to do. And of course it’s natural to want to do easy things.

So, I mustn’t forget cynicism has its own gullibility. It mistakes the least impressed person in the room for being the most right, instead of considering them the least generous.

It might just be me getting old. I do not want to confuse those two again.

The people I trust most are rarely the fastest to dismiss. They are the people who make the room smarter. They ask good questions. They give context. They can be blunt without being cruel. They know how to say, “I do not know enough yet, help me understand…”

I am lucky to have found people like that in past places of work and in life. I found them in the parts of early Twitter that still felt generous. Much of my network still has that quality.

They are not endlessly positive. They are hard to impress. They are direct and their criticism is always pointed toward making the work better. They take care not to humiliate the person doing the work.

That distinction matters to me.

I used to overvalue intelligence. Today, I trust it less on its own. Intelligence without generosity turns into a weapon; without humility overreaches; and without patience misses the hidden work.

I think helpfulness is the better signal.

by helpfulness, I don’t mean politeness. Politeness can often hide a lot. I do not mean niceness either. Niceness can avoid conflict.

I mean the habit of leaving things better than you found them.

That may mean explaining something that was missed but would help improve the situation and the perception of the person.

It may mean asking the question everyone skipped.

It may mean naming a weak argument.

It may mean telling someone they are moving too fast.

It may mean asking questions that help with shared understand of the context.

It often means refusing to reward a cruel joke even when the room enjoys it.

It means standing up for the person who’s feeling the humiliation.

It also doesn’t mean being pleasant. The helpful person is not always pleasant. But they are oriented toward repair, clarity, and better work.

That is who I want around me. This is who I aim to be.

I observe most interactions with this lens now, especially the small moments.

When someone talks about a person who is not in the room, do they become careless? When someone is corrected, do they get curious or defensive? When they deal with a junior person, do they explain or perform? When they have power, do they become more responsible or more entitled? When a group rewards cruelty, do they join in?

No single moment answers the question. People are inconsistent. I am inconsistent.

However, patterns are observable.

What people reward tells me what they want more of. What they excuse tells me what they are willing to live with. What they do when nobody important is watching tells me more than their public values.

I want people who are curious about fields they do not understand. People who build more than they sneer. People who can be skeptical without making skepticism their whole personality. People who are direct, but not addicted to making others feel small.

Most of all, I want people who make it easier to tell the truth.

That is what I’ve best appreciated in my favorite professional cultures. It’s easy to say, “I am wrong,” “I don’t know,” and hence hear “This is good, and this part still needs more work.” It’s what I practice in my own mentoring.

Making admissions costly, causing people to posture, harden, pretend - those are places I want to avoid. That’s a place that will reward contempt as it’s safer than uncertainty.

I am not naive. I do not control any of this at scale. I do not control what the internet rewards. I do not control which behaviors become fashionable.

I only control what I practice.

So, when I look at AI and various fields, I try and remember that the work I can see is probably not the whole work. When I feel certain too quickly, I try to ask one more question. When I criticize (in private), I want the criticism to improve the topic, not the person. When I choose people for my network, I want to choose the ones who make me more honest, more curious, and more useful.

This is the change that I will continue to make.

If this resonates with you, I’d like to get to know you better. Connect with me. On twitter, mastodon, threads, linkedin or email.