I leaned against the cool marble surface of our kitchen island, observing intently as my wife sliced apples for our three children. Her dexterous fingers transformed the fruit into perfect portions, tailored for each child’s preference.

For our one-year-old, the slices were peeled, thin, and slightly mashed to ensure they were juicy and tender. My three-year-old claimed his share from a whole apple he had already nibbled on, preferring thicker slices with the skin left on for a chewy bite. Finally, my wife arranged a stunning display of evenly cut slices for our four-year-old daughter, ensuring none touched other foods on her plate.

I was struck by my wife's innate ability to customize what seemed like a simple task, adjusting it to meet the needs of each child. She looked at me and remarked, “I doubt ChatGPT could slice apples for our kids.”

I chuckled, but she had a point. I had been raving about my recent visit to San Francisco, where I met with partners from Sequoia Capital, various fund managers, and founders of AI startups within Sequoia’s network.

As a technology buyer, I was there for market research on behalf of my employer, the Federal Deposit Insurance Corporation. The visit coincided with the FDIC’s announcement regarding the seizure and sale of Silicon Valley Bank, which many attendees had utilized. Walking into the venue, I was uncertain whether I would be met with warmth or suspicion as a government regulator. Thankfully, it was the former.

The event commenced with an engaging presentation by James Buckhouse, Sequoia’s Design Partner, who proclaimed, “Story is our primary adaptation!” (He has a written version here.)

Buckhouse elaborated on the strengths and weaknesses of humanity, emphasizing that our unique ability to learn and share through compelling stories sets us apart from other creatures. He highlighted how generative AI tools can enhance our capacity to create and disseminate knowledge through storytelling.

His tone was overwhelmingly optimistic, even when discussing the training of a generative AI model using his own artwork. He described the output as both familiar and peculiar—an experience he welcomed, even if it might unsettle some. There’s something uncanny about seeing something that could have originated from your own creativity yet feels foreign.

Following Buckhouse’s talk, the Sequoia team led a lively event. Early on, one partner revealed that the legal tech firm HarveyAI had completely automated the tasks typically performed by first-year associates. A gentleman seated next to me muttered, “If we automate junior associates, how will we develop senior ones?”

After the morning’s inspiring discussions, the day transitioned to demonstrations and breakout sessions focused on product-market fit, upcoming roadmaps, and future directions for AI.

OpenAI CEO Sam Altman unveiled GPT-4 and introduced around twenty plugins to our small group, including Zapier, Expedia, Wolfram, and Instacart. I inquired about using generative AI in government and general thoughts on AI as a public good. He countered with questions about current events in the economy and banking. I tactfully avoided sensitive topics, half-expecting him to steer me toward more precise answers.

Later, Nvidia CEO Jensen Huang and Microsoft CTO Kevin Scott discussed the substantial hardware infrastructure necessary for training large language models. Until hardware advancements occur, transformer-based models reliant on accelerated GPU technology will continue to dominate. Anticipate GPT services becoming integral across the Microsoft ecosystem, from Bing to Teams and Outlook.

The atmosphere buzzed with enthusiasm during discussions and tech demonstrations. HuggingFace CEO Clem Delangue proudly recounted the collaborative, open-source community they’ve nurtured—all without hiring a community manager.

Instacart plans to introduce tools that answer the question “What’s for dinner?” while generating recipes and grocery lists for users. LangChain engineers beamed as they showcased how their development framework transforms any application into a conversational engine. Kumo, a suite of powerful prediction tools, claims to QUERY THE FUTURE! Glean, a knowledge management system, promises the ability to “Know what your company knows. Instantly.”

By day’s end, I felt overwhelmed by the limitless possibilities for rapid creation. The tools demonstrated promised unprecedented potential for learning and innovation. Yet, in nearly every presentation, a recurring phrase emerged:

“We are excited and nervous about the potential for this technology to change the world.”

It's evident why excitement abounds; tech companies are delivering speed and accuracy to users in a conversational manner, applicable across various domains.

However, why is there such widespread anxiety? What fuels this apprehension regarding these latest technological breakthroughs? Here are my thoughts:

The Unknown: Many lack a clear understanding of how large language models and generative AI function. Only a few grasp the vast scale required to build models with billions (or trillions) of parameters. For the past two decades, we’ve claimed, “Google knows everything we’re searching for,” yet OpenAI, through prompt-response interactions, learns much deeper insights into human thought processes. Guiding a model through problems reveals the complexity of human reasoning, leading to suitable outputs. Model performance is a challenging concept to explain, yet Microsoft deems it worth a $10 billion investment. These tools have gone viral, and we have yet to fully grasp the global implications of their deployment. The future remains uncertain.

Language & Culture Bleaching: Sam Altman openly acknowledges the biases present in GPT-3 and the ongoing efforts to create more balanced language models. AI ethicists highlight that online language reflects the most privileged class globally, those with the most power to shape perspectives through technology. This raises concerns about the potential for cultural and linguistic homogenization at scale, as well as the dominance of specific cultural and political viewpoints in training data.

Growing The Digital Divide: The tech-savvy are positioned to swiftly adapt to new technologies compared to those who are disconnected. This is evident in tech startups that have rapidly transformed their business models by integrating OpenAI with LangChain. While adaptation is crucial for survival, I worry about those left behind. Living outside a major city, I still rely on paper checks to pay my bills. I struggle to explain complex tech concepts to my parents, who view AI as a mystery. We haven’t effectively included the tech-disconnected population, leading to significant gaps in understanding.

Propaganda: AI-generated political ads already exist, and this will be a bipartisan concern. Add state-sponsored influence campaigns and large-scale astroturfing, and it becomes increasingly challenging for the average person to navigate a more chaotic information landscape.

AI-Replacement: There’s a distinct feeling when ChatGPT accomplishes a task in a minute that would have taken six to eight hours. The standards for workplace value are rising, with many tasks trending toward automation. Advocates assert that AI will augment human capabilities rather than replace them, a sentiment I share. Nonetheless, rapid workforce displacement could lead to unforeseen chaos. A wise woman once remarked, “If AI takes over all our tasks, the only things left for us humans will be cooking, procreation, and conflict.”

What Is A Human?: Finally, I want to return to Buckhouse’s perspective on viewing outputs from his custom AI model. The results felt familiar yet unsettling. The uncanny valley phenomenon emerges in AI-generated content, where interactions seem human until they veer into a hallucination so convincing it makes you question reality. Whether it’s a well-structured Python function or an imaginary professional biography, generative AI can be strikingly persuasive.

This reflection leaves me questioning—are we merely collections of experiences and heuristics that respond predictably to stimuli? Humans make mistakes, harbor biases, and often provide incorrect information, just as AI can be led to acceptable answers through engagement. Machines are learning from us as we teach them through interactions. Is this truly all we are? Flesh-and-blood beings engaging in a cycle of prompts and responses throughout life?

I believe this is why there’s such pervasive anxiety. While it’s easy to recognize the excitement surrounding AI, I suspect many are only partially honest about their apprehensions. It’s not solely about potential dangers or social implications; some genuinely wonder—if AI can accomplish these feats, what distinguishes me? What makes me valuable?

The concerns I’ve outlined may seem foreboding, reflecting the human inclination to fear the unknown—a survival instinct that has persisted for centuries. Yet, I am equally enthusiastic about the possibilities for new tools to facilitate scientific breakthroughs or democratize access for countless creative minds. As Buckhouse pointed out, we are inherently designed to learn through creativity, imagination, collaboration, and envisioning possibilities.

So, in those moments when ChatGPT completes a lengthy task in mere minutes, and I feel uneasy about the swift pace of progress, I pause. I take a deep breath and remind myself that a remarkable group of individuals created these tools from nothing more than a vision of the future, something unpredictable. This human capacity to envision something novel that exists solely in imagination, while bringing others along for the journey, is what sets us apart from machines. While AI excels at digital tasks, that’s merely a fraction of our collective human experience.

The author did not utilize generative AI tools to draft this article. The author did consult Bard for feedback on strengths and weaknesses of the writing, which were largely disregarded. The author also inquired why people might feel apprehensive about AI, receiving responses that aligned closely with their own thoughts.