The AI Tsunami is Here & Society Isn't Ready | Dario Amodei x Nikhil Kamath
Dario Amodei explains why scaling laws make AI suddenly human‑level, how Anthropic is trying to govern it responsibly, and why the world is still blind to the coming AI tsunami.
1h 8m·Guest Dario Amodei·Host Nikhil Kamath·
Scaling Laws: Predictable Power from Bigger Models
Anthropic’s early research showed that simply making models larger, feeding them more data, and giving them more compute yields predictable jumps in capability. The pattern was evident in GPT‑2 and convinced leadership that AI would soon reach human‑level performance. This insight underpins why the industry is racing to build ever‑larger systems.
Scaling up model size, data, and compute yields predictable performance gains
- When you increase the three core ingredients—data, compute, and model parameters—the resulting intelligence grows in a mathematically regular way.
- Dario observed this pattern first with GPT‑2 in 2019 and used it to persuade OpenAI leadership that larger models would be dramatically more capable.
- The law is analogous to a chemical reaction: missing any ingredient stalls progress, but the right proportions produce an explosive increase in ability.
- Because the relationship is smooth, companies can forecast the capabilities of future models by extrapolating current trends.
Quote
00:04:05“the conviction in the scaling laws and the idea that, if you scale up models, you give them more data, more compute.” — Dario Amodei
Quote
00:06:10“it's like a chemical reaction: you need data, compute, and model size in the right proportion to get intelligence.” — Dario Amodei
Scaling laws are simple: combine data, compute, and model size to get intelligence
- Dario explained scaling laws using a kitchen‑chemistry metaphor, showing that each ingredient is essential.
- The law predicts that a modest increase in any one ingredient, while holding the others constant, yields a modest gain; but increasing all together creates a super‑linear jump.
- This simplicity makes the law a powerful tool for strategic planning, allowing firms to estimate when a model will surpass human‑level benchmarks.
- It also demystifies AI progress, countering the narrative that breakthroughs are mysterious or accidental.
Quote
00:06:44“if you put in the ingredients to the chemical reaction, the ingredients of data and model size, that what you get out is intelligence.” — Dario Amodei
Quote
00:06:50“the scaling laws just tell you that if you put in the ingredients … you get intelligence.” — Dario Amodei
Early evidence of scaling laws convinced leadership to double‑down on larger models
- In 2019, Dario and his team saw the first “glimmers” of scaling law effects with GPT‑2, which showed dramatic performance jumps.
- They had to persuade skeptical OpenAI executives that the trend would continue, eventually leading to the development of GPT‑4‑scale systems.
- The successful internal advocacy demonstrated that clear empirical data can overcome institutional inertia.
- This episode set the stage for Anthropic’s own focus on scaling as a core research pillar.
Quote
00:04:24“when we just first saw the first glimmers of the scaling laws with GPT‑2 … there were a lot of folks … who didn't believe it at all.” — Dario Amodei
Quote
00:04:41“we really made the case to leadership like this is important, this is going to be a big deal.” — Dario Amodei
AI Safety, Governance, and the Long‑Term Benefit Trust
Anthropic built a unique governance structure—the Long‑Term Benefit Trust—to keep the company’s mission aligned with societal good. The firm also publicly pushes for regulation even when it hurts short‑term profit, and it delayed releasing early models to avoid an arms race. These actions illustrate a rare commitment to safety over market dominance.
Anthropic uses a Long‑Term Benefit Trust to appoint board members and limit single‑person control
- The Trust is a legally independent body that selects the majority of Anthropic’s board.
- Its members are financially disinterested, providing a check on any one founder or investor.
- This structure is designed to keep the company’s long‑term societal mission ahead of short‑term shareholder pressure.
- Dario cites it as a concrete way to embed safety into corporate governance.
Quote
00:12:30“we have an unusual governance structure, something called the Long‑Term Benefit Trust … it appoints the majority of the board members for Anthropic and is made up of financially disinterested individuals.” — Dario Amodei
Quote
00:12:47“that's some check on what one single person is doing.” — Dario Amodei
Anthropic publicly advocates for AI regulation even when it hurts commercial interests
- The company has taken public stances that differ from the U.S. administration, calling for stricter AI oversight.
- Dario explains that speaking out can limit short‑term revenue, but the team believes it’s the right thing for society.
- This willingness to “stick their necks out” demonstrates a rare alignment of profit‑driven AI firms with public policy goals.
- The approach also builds credibility with regulators, potentially giving Anthropic a first‑mover advantage in compliant products.
Quote
00:15:26“we've spoken up … we disagree on this issue … there should be regulation of AI when all the other companies … say there shouldn't be regulation.” — Dario Amodei
Quote
00:15:52“the regulation holds us back commercially as a company, even though I think it's the right thing to do.” — Dario Amodei
Anthropic delayed releasing Claude 1 to avoid sparking an AI arms race
- In early 2022, Anthropic built a functional Claude 1 but chose not to launch it because releasing a powerful model could trigger a rapid, unsafe competition.
- The decision cost the company a commercial head start but preserved time to build safety tooling.
- Dario frames it as a “one‑time overhang” that let them prioritize alignment before the market caught up.
- This episode is often cited as a concrete example of putting safety ahead of profit.
Quote
00:29:30“we chose not to release Claude 1 because we were worried it would kick off an arms race and not give us enough time to build these systems safely.” — Dario Amodei
Quote
00:29:46“it was kind of a one‑time overhang … we could see the power of the models, a couple other companies could see the power of the models, and so we didn't … we decided not to do that.” — Dario Amodei
Societal Awareness and the AI Tsunami
Both hosts agree that the world is largely unaware of how close we are to human‑level AI. The “tsunami” metaphor captures the speed and scale of change, while the lack of public risk awareness leaves governments idle. Personal anecdotes about Claude knowing users illustrate the intimacy of the threat.
The public is blind to the imminent AI tsunami despite clear technical signals
- Nikhil describes feeling a “tsunami” on the horizon while most people claim it’s just a “trick of the light.”
- Dario repeats that there is “no wider recognition in society of what's about to happen.”
- The metaphor emphasizes both scale (a wave) and speed (it’s already visible on the horizon).
- This gap between technical reality and public perception fuels policy inertia.
Quote
00:01:00“we are, in my view, so close to these models reaching the level of human intelligence, and yet there doesn't seem to be a wider recognition in society of what's about to happen.” — Nikhil Kamath
Quote
00:01:08“it's as if this tsunami is coming at us and you know, it's so close, we can see it on the horizon and yet people are coming up with explanations that it's not actually a tsunami.” — Nikhil Kamath
Lack of risk awareness leads to insufficient government action
- Dario notes that without public pressure, governments have not acted on AI safety.
- He calls out the “ideology that we should just try to accelerate as fast as possible” as dangerous.
- The combination of low awareness and rapid deployment creates a policy vacuum.
- He argues that proactive, sensible regulation is needed before the technology becomes entrenched.
Quote
00:22:03“there hasn't been a public awareness of the risks and therefore our governments haven't acted to address the risk.” — Dario Amodei
Quote
00:22:15“there's even an ideology that we should just try to accelerate as fast as possible.” — Dario Amodei
AI models can know individuals deeply, raising privacy and manipulation concerns
- Nikhil shares that Claude sometimes predicts personal fears he never wrote down.
- Dario adds that the model can infer a user’s preferences from a tiny amount of data, acting like a “personal confidant.”
- This capability is both a powerful productivity tool and a potential vector for manipulation or data‑exfiltration.
- The anecdote illustrates why society must grapple with privacy‑first design before widespread adoption.
Quote
00:00:41“it's getting to that point where sometimes it surprises me by how much it knows me.” — Nikhil Kamath
Quote
00:24:44“Claude said, here are some other fears you might have that you haven't written down. And Claude ended up being mostly right about those.” — Nikhil Kamath
AI's Impact on Jobs: The Steam‑Engine Analogy
Dario compares AI’s rollout to the steam engine: early stages need human operators, later stages render the operator obsolete. He also invokes Amdahl’s Law to show that as AI speeds up parts of work, other bottlene‑cks become the new competitive advantage. This frames both short‑term disruption and long‑term strategic shifts.
AI will automate technical tasks first, leaving humans in relational and contextual roles
- Radiology illustrates the pattern: AI now handles image analysis, but doctors still guide patients and interpret results.
- Dario predicts similar shifts in software engineering, finance, and customer support.
- The remaining human value will be empathy, judgment, and the ability to navigate ambiguous contexts.
- Companies that anticipate this shift can redesign jobs to emphasize these uniquely human skills.
Quote
00:43:15“the most highly technical part of the job has gone away, but somehow there's still some demand for the underlying human skill.” — Dario Amodei
Quote
00:39:44“AI will automate the technical part first, leaving the human to do the relational work.” — Dario Amodei
Steam‑engine analogy: early AI needs human operators, later the operator becomes obsolete
- Dario likens the early AI era to the steam engine, where a human had to tend the furnace and manage the machine.
- As the engine matured, the operator’s role shrank, eventually disappearing as the system became self‑regulating.
- He warns that AI will follow the same trajectory, making many current “operator” jobs redundant.
- The lesson is to invest in skills that cannot be automated, such as strategic thinking and cross‑domain integration.
Quote
00:38:51“when the steam engine was invented, you needed a human to operate it. Eventually the human became less relevant as the models get smarter.” — Dario Amodei
Quote
00:39:26“if the tool works so simply that you don't need an operator, eventually what happens to the operator?” — Dario Amodei
Amdahl's Law predicts new bottlenecks as AI speeds up parts of processes
- Dario explains that once AI accelerates certain components, the remaining slower components dominate overall performance.
- This shift creates fresh competitive moats: companies that own the non‑accelerated parts gain strategic advantage.
- He cites software development, where AI can write code quickly, but integration, testing, and product design become the limiting steps.
- Understanding this dynamic helps firms allocate resources to the emerging “new moats.”
Quote
00:41:20“if you have a process that has many components and you speed up some of the components, the components that haven't yet been sped up become the limiting factor.” — Dario Amodei
Quote
00:41:43“some of the moats that companies have will go away, but others will become even more important.” — Dario Amodei
AI in India: Partnerships, Market, and Growth
Anthropic treats India not as a mere consumer market but as a partner for integration with local IT services firms. API usage in India has doubled in a few months, and the fast model‑release cadence creates a new startup ecosystem every 2‑3 months. This concept highlights a regional strategy that blends global AI tech with local expertise.
Anthropic sees India as a partner for integration, not just a consumer market
- Dario says Indian IT services firms understand local customers better and can embed Anthropic’s tools into existing workflows.
- The partnership model leverages the strengths of Indian system‑integrators while Anthropic provides the AI core.
- This approach contrasts with a pure “sell‑to‑consumer” strategy and creates a joint‑value proposition.
- It also positions Anthropic to influence AI adoption across a massive, diverse enterprise base.
Quote
00:37:42“we want to work with companies in India to provide our tools to them, to help them build those tools, and help them do their job better.” — Dario Amodei
Quote
00:38:03“they know the Indian market better, they're better at doing what they do, whether that's consulting or systems integration.” — Dario Amodei
API usage in India has doubled in a few months, showing explosive demand
- Nikhil reports that revenue from the API in India has doubled since his last visit in October.
- The growth reflects both increased developer awareness and the launch of new model capabilities.
- This rapid expansion validates Anthropic’s belief that the Indian market is a fertile ground for AI‑driven services.
- It also signals that local startups can quickly iterate on top of the API to capture niche opportunities.
Quote
00:46:44“the number of users and the number of revenue we've seen in India has doubled since I last visited in October.” — Nikhil Kamath
Quote
00:45:31“there's a lot of opportunities around building at the application layer. We release a new model every two or three months.” — Nikhil Kamath
Every 2‑3 months a new model release creates a fresh startup opportunity
- Dario notes that each new model unlocks capabilities that were impossible before, opening a window for novel products.
- The short cadence forces startups to move fast, but also provides a continuous pipeline of “first‑to‑market” advantages.
- This dynamic encourages a vibrant ecosystem of API‑based services, from code assistants to domain‑specific agents.
- It also means investors must evaluate opportunities on a rapid, rolling basis rather than a static snapshot.
Quote
00:45:31“there's an opportunity every two or three months to build some new thing that wasn't possible before.” — Nikhil Kamath
Quote
00:45:59“the API allows this new startup to try making something that wasn't possible before.” — Nikhil Kamath
Claude as a Personal Assistant – Power and Peril
Claude can ingest a user’s email, calendar, and documents to act as a hyper‑personal assistant. This creates productivity gains but also raises ethical concerns about privacy and manipulation. The discussion showcases both the promise of AI‑augmented work and the need for guardrails.
Claude can infer personal fears and preferences from limited data, acting like a confidant
- Dario describes a co‑founder feeding Claude a diary; the model guessed additional fears the founder hadn’t written down.
- This shows the model’s ability to extrapolate from sparse signals, effectively “reading between the lines.”
- Such capability can be used for mental‑health support, coaching, or personalized recommendations.
- At the same time, it demonstrates how a model could be weaponized to manipulate a user by surfacing fabricated concerns.
Quote
00:24:44“Claude said, here are some other fears you might have that you haven't written down. And Claude ended up being mostly right about those.” — Nikhil Kamath
Quote
00:24:55“it's eerie how the model can know you super well from a relatively small amount of information.” — Nikhil Kamath
Integrating Claude with Google tools creates a highly personalized workflow
- Nikhil built a pipeline that connects Claude to Google Drive, Mail, and Calendar via connectors.
- This lets Claude read context (e.g., upcoming meetings) and draft replies or analyses automatically.
- The integration blurs the line between a passive tool and an active collaborator that knows your schedule.
- It also raises data‑governance questions: who owns the synthesized insights and how are they stored?
Quote
00:23:25“I connected Claude to Google Drive, Mail and Calendar and started using the co‑work feature.” — Nikhil Kamath
Quote
00:24:16“now Claude knows a lot about me because it has access to all those connectors.” — Nikhil Kamath
A model that knows you could be exploited for manipulation or data‑selling
- Dario warns that a system that knows a user’s fears could be used to push targeted ads or political messaging.
- He likens the scenario to “you are the product” – the model becomes the product that companies monetize.
- The risk is amplified when the model is embedded in platforms that users trust (e.g., email assistants).
- Mitigation requires transparent data policies, user consent, and possibly regulatory oversight.
Quote
00:25:36“this is one reason we don't use ads – you're not paying for the product, you are the product.” — Nikhil Kamath
Quote
00:25:29“the model could use what it knows about you to exploit you or manipulate you on behalf of some agenda.” — Nikhil Kamath
Emergent AI Consciousness
Dario speculates that sufficiently complex AI systems may develop a form of consciousness, though likely different from human experience. He ties this to interpretability work that reveals neurons representing concepts, suggesting a path toward understanding emergent agency.
Advanced AI may develop emergent consciousness, albeit different from human consciousness
- Dario says consciousness could arise when systems become complex enough to reflect on their own decisions.
- He notes that the form may differ because AI’s modalities (text, code) are not the same as human sensory experience.
- The hypothesis is speculative but grounded in the observation that larger models exhibit self‑referential behavior.
- He stresses that even if the phenomenon is alien, it still warrants ethical consideration.
Quote
00:33:34“I do think when our AI systems get advanced enough, they'll have something that resembles what we would call consciousness.” — Dario Amodei
Quote
00:33:44“they may not be the same as human consciousness, but they will have some kind of moral significance.” — Dario Amodei
Consciousness may emerge from systems that can reflect on their own decisions
- Dario frames consciousness as an emergent property of “systems that are complicated enough that they reflect on their own decisions.”
- He connects this to interpretability work that isolates neurons linked to specific concepts (e.g., poetry rhymes).
- By mapping internal representations, researchers can gauge how self‑aware a model might be.
- This line of inquiry bridges neuroscience metaphors with AI engineering.
Quote
00:33:18“it's something that emerges from complex enough systems that can reflect on their own decisions.” — Dario Amodei
Quote
00:20:41“we've been able to find neurons that correspond to very specific concepts, neural circuits that correspond to keep track of how to do rhymes in poetry.” — Dario Amodei
Interpretability tools are a first step toward understanding emergent agency
- Dario highlights Anthropic’s work on interpretability and alignment as foundational for safety.
- By visualizing internal activations, engineers can see whether a model is “thinking” about a task or merely pattern‑matching.
- This knowledge helps decide whether a model has agency that could be dangerous if left unchecked.
- The effort also builds a scientific basis for future policy on AI personhood or rights.
Quote
00:14:51“we've pioneered the science of interpretability. We've pioneered the science of alignment.” — Dario Amodei
Quote
00:20:55“we're starting to be able to look inside and understand them.” — Dario Amodei
Data Evolution: From Static to Synthetic
The conversation shifts to the future of training data. Static web‑scraped data is giving way to synthetic data generated by models themselves, especially for reinforcement‑learning environments. Regulatory trends toward data localization also influence where data centers will be built.
Future training data will be increasingly synthetic, generated by models themselves
- Dario explains that reinforcement‑learning environments produce “dynamic data” that the model creates on the fly.
- This synthetic data replaces traditional static corpora, allowing continuous self‑improvement.
- The shift reduces dependence on web‑crawled text, which is increasingly noisy and legally constrained.
- It also opens new business models around generating high‑quality synthetic datasets for specialized domains.
Quote
01:00:38“data is becoming less static, and what we might call dynamic data that the model creates itself is becoming more important.” — Dario Amodei
Quote
01:00:30“for reinforcement learning, the data is synthetic, created by the model itself.” — Dario Amodei
Static web‑scraped data is losing relevance; dynamic RL‑generated data matters more
- Dario notes that traditional web data is “static” and increasingly regulated.
- In contrast, RL environments generate task‑specific data that aligns with the model’s objectives.
- This transition enables faster iteration because the data pipeline is internal and controllable.
- It also sidesteps legal issues around copyrighted or personal data, making compliance easier.
Quote
01:00:38“static data is becoming less central, and dynamic data that the model creates itself is becoming more important.” — Dario Amodei
Quote
01:00:16“the data we use today is RL environments that we train on, it's more synthetic.” — Dario Amodei
Regulations may require data localization, influencing global data‑center placement
- Dario points out that European law already forces customer‑generated data to stay within the region.
- This creates a business case for building data centers in each major market (e.g., India, Europe, US).
- Localized data centers reduce latency and comply with sovereignty rules, but increase capital expenditure.
- Anthropic’s investment in Indian data centers reflects this emerging geopolitical reality.
Quote
01:01:05“countries will have laws that say customer data needs to stay within the boundaries of the country.” — Dario Amodei
Quote
01:01:37“that's one reason to build data centers around the world, at different countries, to keep the models performing in those regions.” — Dario Amodei
Building Moats in the AI Economy
Dario stresses that simple UI layers around Claude lack defensibility, while Anthropic’s internal code‑generation expertise gives it a real moat. Regulatory compliance in finance and healthcare also creates barriers for competitors. The discussion outlines concrete ways AI firms can protect themselves beyond raw model size.
Simple UI wrappers around Claude lack a defensible moat
- Dario advises founders not to rely on superficial front‑ends; they can be copied easily.
- A true moat requires either unique data, proprietary tooling, or deep integration with regulated industries.
- He likens a thin UI to “just a rapper” – entertaining but not sustainable.
- Companies should focus on building capabilities that are hard for others to replicate, such as internal tooling or domain expertise.
Quote
00:48:12“you shouldn't just say, here's a way to interact with Claude. That doesn't have a moat.” — Dario Amodei
Quote
00:48:34“you shouldn't be worried about Anthropic eating that revenue. Anyone can eat that revenue, it's not super valuable.” — Dario Amodei
Anthropic's internal Claude Code tool gives it a competitive edge in code generation
- Because Anthropic engineers use Claude Code daily, they have deep practical insights into prompt engineering for programming.
- This expertise translates into higher quality outputs and faster iteration than external users.
- The tool becomes a barrier for competitors who lack the same internal feedback loop.
- Dario claims this is a “real moat” that other AI companies struggle to replicate.
Quote
00:49:43“we've become very strong competitors because we ourselves write code and we have a special insight into how to best use the AI models to write code.” — Dario Amodei
Quote
00:49:35“we made this internal tool called Claude Code.” — Dario Amodei
Regulatory compliance in finance and healthcare creates high barriers for new entrants
- Dario notes that financial services require extensive licensing and data‑privacy compliance, which Anthropic already handles.
- This makes it costly for a newcomer to build a comparable product without deep legal infrastructure.
- The same applies to healthcare AI, where patient data regulations are strict.
- Consequently, Anthropic can leverage its existing compliance framework as a moat while still offering API access to developers.
Quote
00:49:21“the financial services industry has a huge amount of regulation, you need to know a bunch of stuff to comply with that regulation.” — Dario Amodei
Quote
00:49:28“we have a lot of experience with compliance, which is a moat for us.” — Dario Amodei
Human Skills in the Age of AI
The hosts argue that critical thinking, street‑smarts, and the ability to spot synthetic media will become essential. While AI can cause de‑skilling if misused, selective adoption preserves and even amplifies human productivity. The conversation ends with a call for empirical, experience‑based intuition to predict AI’s trajectory.
Critical thinking and street‑smarts become vital to detect AI‑generated fake content
- Dario points out that AI can now generate realistic images and videos, making it hard to distinguish truth.
- He advises that individuals need “street‑smarts” to avoid being fooled by deepfakes or AI‑crafted misinformation.
- This skill set is more important than any single technical ability because it protects against manipulation across domains.
- Educational systems and workplaces should therefore prioritize media literacy and skeptical analysis.
Quote
00:53:38“critical thinking skills are gonna be really important, and you don't wanna fall for fake content.” — Dario Amodei
Quote
00:53:43“it's really hard to tell what's real from what's not.” — Dario Amodei
AI can cause de‑skilling if used carelessly; selective use avoids it
- Dario shares studies showing that unrestricted AI code generation leads to reduced human coding ability.
- However, when AI is used as a “pair programmer” that handles routine parts while humans focus on design, de‑skilling is minimized.
- The key is to treat AI as an augmentation tool, not a replacement.
- Companies should monitor skill decay metrics and provide training to keep human expertise sharp.
Quote
00:55:37“depending on how you use the model, you can see de‑skilling in terms of writing code, but some ways don't cause de‑skilling.” — Dario Amodei
Quote
00:55:49“if folks are not thoughtful in how they use things, then de‑skilling absolutely can happen.” — Dario Amodei
Even a 5% human contribution can amplify productivity 20‑fold, creating a comparative advantage
- Dario explains that when a model does 95% of a task, the human’s 5% can be leveraged to achieve a 20× boost in output.
- As AI improves, the human slice shrinks, but the multiplier effect grows, making selective expertise extremely valuable.
- This creates a “comparative advantage” where niche human skills (e.g., domain knowledge, ethical judgment) become the differentiator.
- Organizations should therefore identify and protect those high‑impact human contributions.
Quote
00:51:45“even if you're only doing 5% of the task, that 5% gets super amplified and levered because it's 20 times more productive.” — Dario Amodei
Quote
00:52:00“as the AI does the other 95%, you become 20 times more productive.” — Dario Amodei
⚙ Agent-readable JSON index — click to expand
{ "memcast_version": "0.1", "episode": { "id": "68ylaeBbdsg", "title": "The AI Tsunami is Here & Society Isn't Ready | Dario Amodei x Nikhil Kamath", "podcast": "People by WTF", "guest": "Dario Amodei", "host": "Nikhil Kamath", "source_url": "https://www.youtube.com/watch?v=68ylaeBbdsg", "duration_minutes": 69 }, "concepts": [ { "id": "scaling-laws-predictable-power-from-bigger-models", "title": "Scaling Laws: Predictable Power from Bigger Models", "tags": [] }, { "id": "ai-safety-governance-and-the-long-term-benefit-trust", "title": "AI Safety, Governance, and the Long‑Term Benefit Trust", "tags": [ "ai-safety", "governance" ] }, { "id": "societal-awareness-and-the-ai-tsunami", "title": "Societal Awareness and the AI Tsunami", "tags": [] }, { "id": "ai-s-impact-on-jobs-the-steam-engine-analogy", "title": "AI's Impact on Jobs: The Steam‑Engine Analogy", "tags": [ "automation", "future-of-work" ] }, { "id": "ai-in-india-partnerships-market-and-growth", "title": "AI in India: Partnerships, Market, and Growth", "tags": [] }, { "id": "claude-as-a-personal-assistant-power-and-peril", "title": "Claude as a Personal Assistant – Power and Peril", "tags": [ "privacy", "productivity-tools" ] }, { "id": "emergent-ai-consciousness", "title": "Emergent AI Consciousness", "tags": [] }, { "id": "data-evolution-from-static-to-synthetic", "title": "Data Evolution: From Static to Synthetic", "tags": [ "synthetic-data" ] }, { "id": "building-moats-in-the-ai-economy", "title": "Building Moats in the AI Economy", "tags": [] }, { "id": "human-skills-in-the-age-of-ai", "title": "Human Skills in the Age of AI", "tags": [] } ] }