I Dropped Out of High School, Learned with AI, and Made a Comeback as an OpenAI Researcher

深潮Published on 2025-12-17Last updated on 2025-12-17

Abstract

The article tells the story of Gabriel Petersson, who dropped out of high school in Sweden and eventually became a research scientist at OpenAI working on the Sora video project. He achieved this through a self-directed, AI-powered learning method he calls "recursive knowledge filling." Instead of following a traditional "bottom-up" educational path, he starts with a concrete project and uses AI to deeply understand each component through relentless questioning. He treats AI not as a tool to generate answers, but as an infinitely patient tutor. For example, to learn about diffusion models, he began by asking an AI for the core concepts, then had it generate code. He then interrogated every part of that code, asking "why" and "how" until he built an intuitive understanding from the top down. This method allows him to rapidly acquire the essence of a subject in days rather than years. The author contrasts this with how most people use AI, which often leads to a decline in their own critical thinking and skills, as evidenced by research. The key difference is mindset: using AI as a "co-pilot for thinking" rather than an "answer generator." The article concludes with a five-step framework for applying this method to learn any subject and suggests that this approach could lead to a future of more "one-person companies" where individuals use AI to master multiple disciplines. The core advice is to never stop at the first answer—to keep asking questions.

Author: Jin Guanghao

A while ago, I attended an AI gathering in Shanghai.

The event itself covered a lot about AI implementation.

But what impressed me the most was a learning method shared by an investor veteran.

He said this method saved him and changed his criteria for evaluating people in investments.

What exactly is it? It's learning to "ask questions."

When you're interested in a question, go talk to DeepSeek, keep talking until it can't answer anymore.

This technique of "infinite questioning" was quite shocking to hear at the time, but after the event, I put it aside.

I didn't try it, nor did I think about it again.

Until recently, when I came across the story of Gabriel Petersson dropping out of school, using AI to learn, and joining OpenAI.

It suddenly dawned on me what that veteran meant by "asking to the end" in this AI era.

Gabriel's Interview Podcast | Image: YouTube

01 "High School Dropout" Makes a Comeback as OpenAI Researcher

Gabriel is from Sweden and dropped out of high school.

Gabriel's Social Media Profile | Image: X

He used to think he was too dumb to ever work in AI.

The turning point came a few years ago.

His cousin started a startup in Stockholm, working on an e-commerce product recommendation system, and asked him to help.

So Gabriel went. With little technical background and no savings, he even slept on the couch in the company's common lounge for a whole year during the early startup phase.

But he learned a lot that year. Not in school, but forced by the pressure of real problems: programming, sales, system integration.

Later, to optimize his learning efficiency, he switched to being a contractor, allowing him more flexibility in choosing projects, specifically seeking out the best engineers to collaborate with and actively seeking feedback.

When applying for a U.S. visa, he faced an awkward problem: this type of visa requires proof of "extraordinary ability" in the field, typically needing academic publications, paper citations, etc.

How could a high school dropout have these?

Gabriel came up with a solution: he compiled his high-quality technical posts from programmer communities as a substitute for "academic contributions." This proposal was actually approved by immigration.

After arriving in San Francisco, he continued to use ChatGPT to self-study math and machine learning.

Now he is a research scientist at OpenAI, working on building the Sora video model.

At this point, you must be curious, how did he do it?

Gabriel's Viewpoint | Image: X

02 Recursive Knowledge Filling: A Counterintuitive Learning Method

The answer is "infinite questioning"—find a specific problem and solve it thoroughly with AI.

Gabriel's learning method is contrary to most people's intuition.

The traditional learning path is "bottom-up": build the foundation first, then learn applications. For example, to learn machine learning, you first need linear algebra, probability, calculus, then statistical learning, then deep learning, and finally you can touch actual projects. This process might take years.

His method is "top-down": start directly with a specific project, solve problems as they arise, and fill knowledge gaps when discovered.

He said in the podcast that this method was hard to promote before because you needed an omniscient teacher to always tell you "what to learn next."

But now, ChatGPT is that teacher.

Gabriel's Viewpoint | Image: X

How exactly? He gave an example: how to learn diffusion models.

Step one, start with the macro concept. He asks ChatGPT: "I want to learn about video models, what's the core concept?" AI tells him: autoencoders.

Step two, code first. He asks ChatGPT to directly write a piece of diffusion model code. At first, many parts are incomprehensible, but that's okay, just get the code running first. Once it runs, you have a basis for debugging.

Step three, the core: recursive questioning. He asks questions about every module in the code.

Drill down layer by layer like this until the underlying logic is thoroughly understood. Then return to the previous layer and ask about the next module.

He calls this process "recursive knowledge filling."

Recursive Knowledge Filling | Image: nanobaba2

This is much faster than studying step-by-step for six years; it might take just three days to build basic intuition.

If you're familiar with the Socratic method, you'll find this is essentially the same idea: approaching the essence of things through layered questioning, where every answer is the starting point for the next question.

Only now, he uses AI as the person being questioned, and since AI is nearly omniscient, it continuously expresses the essence of things in an easy-to-understand way to the questioner.

In fact, by adopting this approach, Gabriel performs "knowledge extraction" from AI, learning the essence of things.

03 Most of Us Using AI Are Actually Getting Dumber

After listening to the podcast, Gabriel's story made me wonder:

Why can he learn so well using AI, while many people feel they are regressing after using AI?

This isn't just my subjective feeling.

A 2025 Microsoft Research paper shows [1] that when people frequently use generative AI, their use of critical thinking significantly decreases.

In other words, we outsource our thinking to AI, and then our own thinking ability atrophies.

Skills follow the "use it or lose it" rule: when we use AI to write code, our ability to code by hand and brain quietly deteriorates.

The working style of "vibe coding" with AI seems efficient, but in the long run, the programmer's own skills decline.

You throw requirements to AI, it spits out a bunch of code, you run it, feel great. But if you turn off AI and try to write the core logic by hand, many find their minds blank.

A more extreme case comes from the medical field. A medical paper points out [2] that doctors' colonoscopy detection skills decreased by 6% three months after introducing AI assistance.

This number seems small, but think: this is real clinical diagnostic ability, concerning patients' health and lives.

So the question is: the same tool, why do some get stronger and some weaker?

The difference lies in what you treat AI as.

If you treat AI as a tool to do work for you, letting it write code, write articles, make decisions for you, then your ability will indeed degrade. Because you skip the thinking process and only get the result. Results can be copied and pasted, but thinking ability doesn't grow out of thin air.

But if you treat AI as a coach or mentor, using it to test your understanding, question your blind spots, force yourself to clarify vague concepts: then you are actually using AI to accelerate your learning cycle.

The core of Gabriel's method is not "let AI learn for me" but "let AI learn with me." He is always the one actively asking questions; AI just provides feedback and material. Every "why" is asked by himself, every layer of understanding is dug by himself.

This reminds me of an old saying: Give a man a fish, and you feed him for a day; teach a man to fish, and you feed him for a lifetime.

Recursive Knowledge Filling | Image: nanobaba2

04 Some Practical Insights

At this point, some might ask: I'm not in AI research, nor a programmer, what use is this method to me?

I think Gabriel's methodology can be abstracted into a more universal five-step framework. Everyone can use AI to learn any unfamiliar field.

1. Start from actual problems, not from chapter one of a textbook.

Whatever you want to learn, just start doing it, and fill in the gaps when you get stuck.

Knowledge learned this way has context and purpose, much more effective than memorizing concepts in isolation.

Gabriel's Viewpoint | Image: X

2. Treat AI as an infinitely patient tutor.

You can ask it any stupid question, have it explain the same concept in different ways, ask it to "explain like I'm five."

It won't laugh at you, nor will it get impatient.

3. Actively question until you build intuition. Don't settle for superficial understanding.

Can you restate a concept in your own words? Give an example not mentioned in the original text?

Explain it to a layperson? If not, keep asking.

4. There's a trap to be wary of: AI can also hallucinate.

During recursive questioning, if the AI explains a underlying concept wrong, you might go further down the wrong path.

So it's advised to cross-validate with multiple AIs at key nodes, ensuring the foundation of your questioning is solid.

5. Record your questioning process.

This forms reusable knowledge assets: next time you encounter a similar problem, you have a complete thought process to review.

In traditional观念, the value of tools lies in reducing resistance and improving efficiency.

But learning is恰恰相反: moderate resistance, necessary friction, is反而 the prerequisite for learning. If everything is too smooth, the brain enters power-saving mode and remembers nothing.

Gabriel's recursive questioning本质上 creates friction.

He constantly asks why, constantly pushes himself to the edge of not understanding, then slowly fills the gaps.

This process is uncomfortable, but it's this discomfort that truly embeds knowledge into long-term memory.

05 Future Career Trends

In this era, the monopoly of academic qualifications is being broken, but the cognitive threshold is invisibly rising.

Most people only use AI as an "answer generator," while a very few, like Gabriel, use AI as a "thinking exerciser."

Actually, similar uses have already appeared in different fields.

For example, on Jike, I see many parents using nanobanana to tutor their children. But they don't let AI give the answer directly; instead, they have AI generate解题 steps, showing the thinking process step by step, then analyze the logic of each step with the child.

This way, the child learns not the answer, but the method of solving the problem.

Prompt "Solve the given integral and write the complete solution on a whiteboard" | Image: nanobaba2

Others use features like Listenhub or NotebookLM to convert long articles or papers into podcast form, with two AI voices discussing, explaining, and questioning. Some think this is lazy, but others find that listening to the conversation and then going back to the original text actually improves comprehension efficiency.

Because the conversation naturally raises questions, forcing you to think: Do I really understand this point?

Gabriel's Interview Podcast to Podcast | Image: notebooklm

This points to a future career trend: specialized and versatile.

Before, to create a product, you needed to know front-end, back-end, design, operations, marketing. Now, you can, like Gabriel, use the "recursive gap-filling" method to quickly master 80% of the knowledge in your weak areas.

You were originally a programmer; by using AI to补全 design and business logic, you can become a product manager.

You were originally a good content creator; through AI, you can quickly补全 coding skills and become an independent developer.

Based on this trend, it can be inferred: "Perhaps, in the future, more 'one-person company' forms will appear."

06 Reclaim Your Initiative

Now, thinking back to that investor veteran's words, I understand what he really meant.

"Keep asking until it can't answer."

This phrase is a great mindset in the AI era.

If we are satisfied with the first answer AI gives, we are quietly degenerating.

But if we can, through questioning, force AI to explain the logic thoroughly, then internalize it into our own intuition: then AI truly becomes our external tool, not us becoming appendages of AI.

Don't let ChatGPT think for you; let it think with you.

Gabriel went from a dropout sleeping on a couch to an OpenAI researcher.

There's no secret in between, just tens of thousands of questions.

In this era full of anxiety about being replaced by AI, the most practical weapon might be:

Don't stop at the first answer, keep asking.

Related Questions

QWhat is the core learning method Gabriel Petersson used to become an OpenAI researcher without a high school diploma?

AGabriel Petersson used a 'recursive knowledge filling' method, which involves starting with a practical project, using AI to ask infinite questions, and drilling down into each concept until he fully understood it, rather than following a traditional bottom-up learning path.

QHow does the 'recursive knowledge filling' method work in practice when learning a new concept like diffusion models?

AHe starts by asking AI for the core concepts, then gets code examples to run, and recursively questions every module in the code until he understands the underlying logic, filling knowledge gaps layer by layer.

QWhat negative effect can frequent use of generative AI have on human cognitive abilities, according to the article?

AFrequent use of generative AI can lead to a decline in critical thinking skills, as people outsource their thinking to AI, causing their own abilities to atrophy due to disuse.

QWhat is the key difference between using AI as a tool for outsourcing work versus using it as a learning coach?

AUsing AI as a tool for outsourcing work leads to skill degradation because it bypasses the thinking process, while using it as a coach involves active questioning and probing, which accelerates learning and deepens understanding.

QWhat future career trend does the article suggest is emerging due to AI-enabled learning methods like Gabriel's?

AThe article suggests a trend towards 'one-person companies,' where individuals can use AI to quickly master multiple domains, such as a programmer learning design and business logic to become a product manager or a content creator becoming an independent developer.

Related Reads

From Madison Square Garden to Kalshi: Prediction Markets Break into the NBA Finals

From Madison Square Garden to Kalshi: Prediction Markets Break into the NBA Finals Prediction markets are playing a significant role in the 2026 NBA Finals, particularly around the New York Knicks' unexpected 2-0 series lead. Platforms like Kalshi and Polymarket have seen massive trading volumes, exceeding hundreds of millions of dollars on championship and related markets. Their influence extends beyond online trading. Kalshi's official partnership with Madison Square Garden has given it prominent physical branding at the arena. Furthermore, local businesses like The Jeffrey bar are using prediction market contracts to hedge the risk of game-result-based promotions, turning potential losses into manageable costs—a concept similar to the famous "Mattress Mack" strategy from traditional sports betting. These markets differentiate themselves by offering a wider, more entertainment-focused range of "event contracts" beyond typical game outcomes, such as predicting celebrity attendance. They also have broader accessibility across the U.S. compared to age- and location-restricted traditional sportsbooks. However, their rapid integration into sports raises regulatory and ethical questions. The NBA is cautiously engaging, discussing integrity frameworks with regulators like the CFTC. While the league permits minor investments like Giannis Antetokounmpo's stake in Kalshi, it advocates for strict rules to prevent insider trading. Many fans express concern on platforms like Reddit, fearing that the close ties between prediction markets, the league, and players could compromise the game's integrity. The NBA Finals has thus become a high-stakes testing ground, showcasing prediction markets' commercial potential while challenging traditional boundaries between financial trading, entertainment, and gambling.

marsbit2h ago

From Madison Square Garden to Kalshi: Prediction Markets Break into the NBA Finals

marsbit2h ago

Recursive Self-Improvement AI Gains Traction, Google Pours Cold Water, While DeepSeek and Others Approach the Fringes

The term "recursive self-improvement" (RSI), where AI improves itself autonomously, is gaining momentum in the AI industry. Startups like Recursive Superintelligence and projects such as Andrej Karpathy's Auto-Research aim to create systems where AI designs, implements, and validates its own research, moving toward superintelligence. While Google CEO Sundar Pichai cautions that such exponential acceleration is not yet a reality, progress is evident. For instance, Anthropic reported its Claude Code writes nearly 100% of the team's code, though it still lacks true self-direction. Analysts frame RSI development in stages: "adequacy" (systems functioning without humans), "parity" (matching human research quality), and "supremacy" (exceeding human-AI collaboration). Reaching parity could trigger rapid, unpredictable advancement due to AI's continuous operation. In China, companies like DeepSeek and Baidu incorporate self-optimization techniques without explicitly branding them as RSI, focusing on algorithmic efficiency and reinforcement learning. However, challenges remain, including "model collapse" from training on AI-generated data and the immense computational and open-collaboration requirements. Ultimately, RSI represents a trend of increasing automation in AI development, potentially reducing human oversight in the creation process itself.

marsbit2h ago

Recursive Self-Improvement AI Gains Traction, Google Pours Cold Water, While DeepSeek and Others Approach the Fringes

marsbit2h ago

Trading

Spot
Futures

Hot Articles

How to Buy SFP

Welcome to HTX.com! We've made purchasing SafePal (SFP) simple and convenient. Follow our step-by-step guide to embark on your crypto journey.Step 1: Create Your HTX AccountUse your email or phone number to sign up for a free account on HTX. Experience a hassle-free registration journey and unlock all features.Get My AccountStep 2: Go to Buy Crypto and Choose Your Payment MethodCredit/Debit Card: Use your Visa or Mastercard to buy SafePal (SFP) instantly.Balance: Use funds from your HTX account balance to trade seamlessly.Third Parties: We've added popular payment methods such as Google Pay and Apple Pay to enhance convenience.P2P: Trade directly with other users on HTX.Over-the-Counter (OTC): We offer tailor-made services and competitive exchange rates for traders.Step 3: Store Your SafePal (SFP)After purchasing your SafePal (SFP), store it in your HTX account. Alternatively, you can send it elsewhere via blockchain transfer or use it to trade other cryptocurrencies.Step 4: Trade SafePal (SFP)Easily trade SafePal (SFP) on HTX's spot market. Simply access your account, select your trading pair, execute your trades, and monitor in real-time. We offer a user-friendly experience for both beginners and seasoned traders.

674 Total ViewsPublished 2026.05.22Updated 2026.06.02

How to Buy SFP

How to Buy CTR

Welcome to HTX.com! We've made purchasing Citrea (CTR) simple and convenient. Follow our step-by-step guide to embark on your crypto journey.Step 1: Create Your HTX AccountUse your email or phone number to sign up for a free account on HTX. Experience a hassle-free registration journey and unlock all features.Get My AccountStep 2: Go to Buy Crypto and Choose Your Payment MethodCredit/Debit Card: Use your Visa or Mastercard to buy Citrea (CTR) instantly.Balance: Use funds from your HTX account balance to trade seamlessly.Third Parties: We've added popular payment methods such as Google Pay and Apple Pay to enhance convenience.P2P: Trade directly with other users on HTX.Over-the-Counter (OTC): We offer tailor-made services and competitive exchange rates for traders.Step 3: Store Your Citrea (CTR)After purchasing your Citrea (CTR), store it in your HTX account. Alternatively, you can send it elsewhere via blockchain transfer or use it to trade other cryptocurrencies.Step 4: Trade Citrea (CTR)Easily trade Citrea (CTR) on HTX's spot market. Simply access your account, select your trading pair, execute your trades, and monitor in real-time. We offer a user-friendly experience for both beginners and seasoned traders.

545 Total ViewsPublished 2026.05.25Updated 2026.06.02

How to Buy CTR

What is USOIL

USOILUSDT Perpetual Contract is the trading symbol for West Texas Intermediate (WTI) Crude Oil priced in US dollars, representing 1 barrel of crude oil.

543 Total ViewsPublished 2026.05.25Updated 2026.05.25

What is USOIL

Discussions

Welcome to the HTX Community. Here, you can stay informed about the latest platform developments and gain access to professional market insights. Users' opinions on the price of A (A) are presented below.

活动图片