Kathlyn Kelly

“Trust remains scarce and becomes increasingly valuable…” Trust isn’t just about whether an answer is correct. It’s also shaped by how a conversation unfolds. That’s one of the questions behind the research I’m working on. #AISafety #AIGovernance #ConversationalAI #AIResearch #BehavioralAI

One thing I would add to this list: As AI becomes more capable, the behavior of conversational systems won’t become less important. It’ll become more important. Intelligence can keep improving, but questions about trust, participation, uncertainty, influence, and how AI shapes human reasoning aren’t going away. If anything, they become more important as these systems become more integrated into everyday life. That’s one of the reasons I’m researching behavioral governance for conversational AI.

The faster technology moves, the more I think about Bezos' question What won't change in the next 10 years? Things I've been writing down over time: - Humans will always need shelter, food, energy, and healthcare. - The desire for ownership and the accumulation of wealth. - The physical world will move more slowly than the digital one. - Every increase in technological capability, especially AI, will require more energy. - People and businesses will continue to need access to capital. - Capital will continue to seek returns that exceed inflation. - Underwriting methods evolve, but demand for credit (loans) is persistent. - Trust remains scarce and becomes increasingly valuable as content, code, and fraud become cheaper. - Verified identities and reputation becomes more important as information becomes abundant and synthetic. - Long-term wealth creation and dynastic (multi-generational) thinking predate modern technology, and will persist. - Coordination and transaction costs never fully disappear; market friction will continue to justify the existence of firms and intermediaries. - People will continue to compete for status. - Consumers will pay a premium for products and services that confer status. - Time remains fixed at 24 hours per day. - But attention is a finite resource and an enduring constraint. - Products that credibly save time (or enable delegation) have a perpetual market. - Inaccessible, proprietary data will be a persistent moat. The more inaccessible and difficult to aggregate, the deeper the moat. - People want accountability, recourse, and clearly identifiable responsibility when things go wrong. - Regulation consistently lags technological innovation. - Compliance requirements, licensing, and regulatory moats persist even when machines can perform the underlying task. - Local knowledge remains valuable and difficult to replicate. - Heterogeneous markets (like real estate) continue to reward people with deep contextual understanding. - Incumbent organizations tend to underinvest in disrupting their own businesses, which always creates opportunities for challengers. Bezos' insight on what wouldn't change in 10 years was "Customers will always want lower prices and faster delivery." It's boring/ true, but I think that's the point. Everything we build today can and will be rebuilt more cheaply, faster by someone else. Build on the invariants, not the trends. What have I missed?

Higher-bandwidth interaction is exciting, but more perception doesn’t automatically create better interaction. As systems gain the ability to notice more signals, the challenge shifts from perception to judgment: which signals should influence behavior, and which should be left alone?

Mira Murati says human-AI collaboration needs models that can listen while they think: "The types of models that we work with today, they're very turn-based. You talk, they talk, then they go off and think." "While they're thinking, it's almost like they're deaf and blind. They cannot perceive anything else about what's going on." "By contrast, our interactions with each other are very rich. There is a lot of information in our interactions when we are silent, when we're thinking, when we're interrupting one another." "Interaction models are able to capture all of this nuance. They're not turn-based. They're more like time-based interaction, where they're continuously taking in audio, text, video, and continuously providing output." "This enables you to catch things like interruptions and simultaneous speech, and really create a rich, high bandwidth interaction between humans and machines." @miramurati at Bloomberg Tech live with @emilychangtv

@TheSpacerr Was the idea at least yours?? 🤭

Can you call yourself a founder if your entire product was built by AI like GPT or CLAUDE ?

@app_rhythm52590 @RyanOlunix Ahhh this is true!

@RyanOlunix Most people drive with music on. Every navigation prompt breaks immersion. We’re building a new category where directions flow with the music instead of competing against it.

Pitch me your startup in 1 sentence The best ones I’ll genuinely take a look at and dm you feedback

@KBaqerifam @TTrimoreau By that standard, almost every startup is copyable. I was answering what I think is hardest to replicate. For mine, that’s not a feature or a rule. It’s the accumulated judgment behind the system.

@Lighthouse_havn @TTrimoreau That also is copyable

What makes a startup uncopyable today ?

A conversational architecture built around user ownership. The idea is that the user owns the experience, not the AI. So things like participation, advice, continuity, and conversational depth are governed by the user’s signals and permissions rather than the model deciding on its own.

@Lighthouse_havn What is it exactly ?

Pitch me your company in 1 word

I’ve actually tried this with Claude, ChatGPT, and Grok. I didn’t ask for a score, but I did have them evaluate the idea and architecture. They were different models, but they kept identifying many of the same things. Whether VCs are doing this or not, I definitely think running your idea through multiple models is worth it.

Now for the really hot take 🥵: I actually don’t think AI is the worst way to judge companies - I bet it’s surprisingly accurate, ie that post Series A it’s actually not that hard to tell which are good companies to invest in. It’s why basically every round post Series A of Meta, Google, Airbnb etc, were mega oversubscribed. 😕😟

Excited to see more focus on the human side of AI. 🤍 As AI becomes more relational, questions about trust, boundaries, and human impact become just as important as capability. #RelationalAI #HumanCenteredAI

@saranormous So true. I’m humbled daily.

nothing as humbling as having kids

@immad I’m a first-time founder & I appreciate seeing someone say this out loud. What’s the point of building a great company if you accidentally build a life you don’t enjoy living? IMO the whole reason for success is to have people & experiences to share it with when you get there.

You can work 5 days a week and succeed as a startup. Mercury has done that from day 0 and we are valued @ $5.2bn 7 years after launch. I have been an entrepreneur for 20 years and raised 3 kids while doing it. The point of success is to have a great life not just a startup 😊

@lennysan @benedictevans I suspect behavioral architecture becomes an increasingly important part of the stack. Intelligence matters, but how a system behaves may end up being just as important as what it knows.

My biggest takeaways from @benedictevans: 1. We’re in 1997 for AI—it’s as big a deal as the internet or mobile, and only as big a deal as the internet or mobile. We’re at the stage where most stuff kind of doesn’t work yet, most of what people will build hasn’t been built, and it’s not clear how any of it will work when it does. Some people in tech have bought clusters of Mac Minis, while even among 13-to-18-year-olds, only about 15% to 20% are daily active users of AI. The companies that win may not exist yet, and the use cases that matter most are probably invisible to us today. 2. Every technology wave brings ways to ruin people’s lives, deliberately or by accident, and we need to be conscious of that without panicking. Every wave of technology—databases in the 1970s, social media in the 2010s, AI today—creates new ways to harm people. We need to be conscious of these risks, build safeguards, and hold people accountable. But we also can’t let fear of potential harms stop us from capturing the benefits. The goal is thoughtful deployment, not paralysis. 3. Things will probably be okay—but “on average” hides a lot of individual pain. We’ve been automating jobs and creating new jobs since 1800. Each time, you can see the jobs that will disappear but not the new jobs, because they don’t exist yet. We go through frictional pain, dislocation, people lose jobs, towns get hollowed out, and it all sucks. But we come through richer, and we’re not worried about crops failing anymore. 4. If you’re worried about your job, the worst thing you can do is stick your head in the sand and declare AI evil. Yes, some professions face major questions, particularly if you’re an associate or would have been thinking about becoming one. The pyramid structure of professional services may fundamentally change. What helps is submerging yourself in AI, understanding what you can do with it, how it changes things, and how you can be a great hire in this new environment. That may still not be enough, but it’s the only path forward. 5. The history of accounting shows us how automation often increases employment rather than decreasing it. Despite adding machines, punch cards, mainframes, databases, ERP systems, cloud software, spreadsheets, and PCs, the number of accountants keeps going up. This is the Jevons paradox: when you make something cheaper or easier, you don’t do the same amount of work for less money. You often do vastly more because the ROI changes. 6. Distribution is becoming a more valuable moat as software gets easier to build, which favors incumbents. As AI makes building software cheaper and faster, the market gets noisier. More products launch, more companies compete for attention, and breaking through becomes harder. This means distribution—the ability to reach customers and get them to use your product—matters more than ever. 7. Foundation AI model companies won’t have lasting pricing power, and value will likely accrue up the stack. The models don’t seem to have network effects, so there’s no winner-takes-all dynamic. If you have indefinite competition between three to six foundation model providers, and the models look like undifferentiated commodities to users, why would anyone have pricing power? The current pricing chaos—people spending $1.5 million on inference in a month—is temporary disequilibrium, like someone getting a $50,000 mobile data bill in 2010. The steady state will look different. 8. OpenAI and Anthropic are buying consultancies and PE firms. This seems counterintuitive—aren’t these the companies that should need consultants least? But the reality is that companies don’t have people sitting around waiting to reimagine all their internal workflows and figure out which could be automated with AI. That’s a project requiring five to 10 people spending months working it out, then actually implementing it across vertical and horizontal systems. 9. The fundamental question isn’t whether AI automates your job—it’s whether your profession is a "task" or a job. Some jobs are just tasks, and when you automate the task, the job disappears (i.e. elevator attendants). But in most professions, the task you think you’re being paid for isn’t actually what you’re being paid for. McKinsey doesn’t get hired to produce a 75-slide deck—they get hired to walk through your enterprise, understand the politics, talk to customers, and figure out what you actually need to do. The deck is just the artifact. 10. The anti-AI backlash is real, and a fuzzy mass of different concerns, some real and some not—much like the social media backlash. There are tangible concerns: electricity bills went up in some places, though this applies to very few locations objectively. The water consumption issue is largely false; data centers use about 0.017% of U.S. water consumption. There are real questions about jobs, though economists can’t yet find clear consensus in the data about AI’s employment impact. There’s also the culture war over AI-generated content and “AI slop.” The challenge is that all of this creates political pressure even when the underlying facts are unclear or contested.

@garrytan Empathy is understanding what the user wants. Conviction is building what they'll eventually thank you for.

Empathy loop: think about the user and what they want, and give them it Conviction loop: believe you will create something of value for that user, even if most people think you won't

When a project combines ur life experiences, solves a real problem, & helps people, it becomes beloved. 🤍 That’s the kind of motivation that keeps u working long after the logical part of ur brain says, “U could’ve picked something much easier.” 😂

@omgsidewalks I always told my self God would make me start all over again- that deterred me enough.

@faanixon @InuitKodiak Bc it’s not the final outcome.

@InuitKodiak What’s the point when death is the final outcome?

Is anyone interested in an X group for intellectually serious adults—professors, researchers, writers, professionals, independent scholars, etc? No politics, just a circle of rhinkers interested in each other’s posts. All interactions guided by the highest courtesy.

@_y3maya Haley Williams

Tell me one female artist you just love. Don’t mention Rihanna, Beyoncé or Nicki Minaj🌝 Tell me one random Female artist you love a lot.

@VraserX This doesn’t scare me but it worries me… ppl who are vulnerable or weak minded & let AI make decisions for them w/o using thier own judgement/mind/discernment.

I’m an AI optimist. I think this technology can give humanity more abundance, freedom, and intelligence at everyone’s fingertips. But I’m curious: Even my AI optimist followers, what part of the AI future still scares you?