How do I convert a GitHub repo to a text file?

Paste the repository URL above and the file tree appears as a checkbox list. Untick what the model does not need, generate, and you get one text file: directory structure at the top, then each selected file under a header with its path. Nothing to clone, nothing to install, free and no sign-up.

How do I convert a private GitHub repository?

Paste a personal access token alongside the URL. Create a fine-grained token scoped to just that one repository with read-only Contents permission and a short expiry — that is the least privilege the API will accept for this. The token is used to call GitHub from your browser and is not stored on our side.

How many tokens is my codebase, and will it fit?

The counter updates as you tick files, so you can see the number before you paste anything. As a rough guide, source code runs around 3 to 4 characters per token, so a 500 KB selection lands near 140,000 tokens. That fits a 200K context window and will not fit into a single chat message on most interfaces.

Which files should I exclude before feeding a repo to an LLM?

node_modules, vendor, and any dependency directory first — they are usually the overwhelming majority of the bytes and contribute nothing about your code. Then lockfiles, build output, minified bundles, test fixtures, and binary assets. On a typical JavaScript project that pass alone often removes over ninety percent of the tokens.

How does this compare to gitingest, repomix, or code2prompt?

Those are strong tools and the honest difference is workflow, not capability. They are command-line first, scriptable, and the right answer inside CI or a batch job. This runs in a browser tab with a checkbox tree, so you can see the repo, deselect what you do not want, and watch the token count move — better for the one-off case where you are still deciding what to include.

Is it safe to put my code through this?

Public repositories are read through the GitHub API from your browser, so the code goes from GitHub to you. For anything confidential the stronger option is the local folder converter, which never uploads at all. Either way, scan your selection for API keys and credentials before pasting it into a model — that is the step people skip.

Understanding AGI vs. LLMs: Key Differences Explained

AGI vs. LLMs: Why Artificial General Intelligence Isn’t Just Another Chatbot

Artificial Intelligence has made extraordinary leaps in recent years. When OpenAI’s ChatGPT burst onto the scene, many people were amazed – here was a chatbot that could answer complex questions, write code, and hold conversations almost like a human. Terms like AGI (Artificial General Intelligence) started popping up in headlines, and some wondered if systems like GPT-4 were early examples of it. But what exactly is AGI, and is it just a fancy word for an advanced chatbot? In reality, artificial general intelligence isn’t just another chatbot – it represents a much more ambitious idea. Understanding artificial general intelligence requires looking at what “general” intelligence really means, and how it fundamentally differs from the narrow AI systems (like large language models) we use today.

What is AGI? Understanding Artificial General Intelligence

Let’s start with the basics. Artificial General Intelligence (AGI) refers to a hypothetical AI system that possesses broad, general-purpose intelligence comparable to a human’s. In other words, an AGI wouldn’t just excel at one task – it could understand and learn to do virtually any intellectual task that a human being can, including tasks it hasn’t encountered before. The goal of AGI is to create a machine that can adapt to new problems and environments the way a person can, rather than being limited to a pre-defined set of functions.

This concept is often contrasted with today’s narrow AI (or “weak AI”), which is designed for specific tasks or domains. For example, your email’s spam filter does one thing well – it identifies spam – but it can’t hold a conversation. A chess program might beat grandmasters at chess, but that same program can’t play a game like Tetris or drive a car. In fact, most AI systems are highly specialized; they can only do a single thing, in a single context . Your spellchecker can’t do mathematics, and the world’s best chess-playing program can’t play Tetris. Human intelligence is different. We can solve a variety of tasks, including those we have not seen before. An AGI, therefore, is envisioned as an AI with human-like versatility – the ability to apply intelligence to any problem, not just a narrow domain.

It’s important to note that experts don’t all agree on one single definition of AGI. In fact, AI researcher Julian Togelius argues that “artificial general intelligence is a confused and confusing idea,” largely because we don’t even have a clear consensus on what “intelligence” or “generality” truly mean. Some define AGI as an AI that can do anything the human brain can do , or an AI that achieves human-level performance across virtually all cognitive tasks. Others set the bar even higher (or lower). Togelius points out that if you interpret AGI as an AI that can literally solve every possible problem, that’s an impossibly high bar – among other issues, the “No Free Lunch” theorem tells us a single solution can’t optimally solve all problems. On the other hand, a more pragmatic definition might be an AI that can perform most economically important tasks about as well as a human, given minimal instruction. In simple terms, think of AGI as trying to build an AI that can do the kinds of thinking and learning that humans do , not just one trick.

Sometimes you’ll hear AGI referred to as “strong AI” or “human-level AI.” All these terms circle around the same idea: a machine with general cognitive abilities. In contrast, today’s AI is narrow because each system is usually tailored to a specific task (hence “weak AI” – not weak in the sense of poor performance, but limited in scope).

LLMs: Large Language Models – Impressive but Limited AI

Large Language Models (LLMs) , like OpenAI’s GPT-3 or GPT-4 (which powers ChatGPT), are a type of AI that has garnered huge public attention. These models are trained on massive amounts of text data and can generate human-like language. They can answer questions, write essays, summarize documents, create poetry, and even produce computer code. Given how versatile LLMs appear, it’s tempting to think they might be a step toward general intelligence. After all, ChatGPT can converse on almost any topic – doesn’t that mean it has broad understanding?

The reality is more nuanced. LLMs are very powerful pattern recognizers and generators . They work by predicting the most likely next word in a sentence based on the patterns they learned from billions of sentences in their training data. By doing so, they simulate understanding. This leads to extremely convincing outputs – the model can chat about history, then switch to helping with programming, then explain a joke, all in one session. It’s no wonder people feel like they’re talking to something intelligent.

However, an LLM like GPT-4 is still fundamentally a product of narrow AI. It has no true understanding of the world beyond the text it was trained on . It doesn’t know facts in the way you or I do; it has no sense of physical reality or personal experience. Essentially, it’s restricted to language . This is why LLMs can sometimes produce answers that sound confident but are factually wrong – they lack a grounded model of the real world. For example, if you ask an LLM a question about an event outside its training data, it might generate a very plausible-sounding answer that is completely false (a phenomenon often called “hallucination” in AI).

LLMs also don’t learn from interaction in real time . Once a model like ChatGPT is trained (a huge offline process), it isn’t automatically learning new facts each time you chat with it. If a new scientific discovery is made tomorrow, your copy of ChatGPT won’t know about it unless its developers retrain or update it with new data. By contrast, a hallmark of general intelligence is the ability to continually learn and adapt. Current LLMs are mostly static – they’re trained once on a broad dataset, and while they can be fine-tuned or updated, they don’t autonomously evolve with each conversation.

None of this is to dismiss the power of LLMs. They are a remarkable achievement in AI – some researchers even consider them a glimpse of “weak emergence” of generality because they can handle a wide range of language tasks. Indeed, models like GPT-4 exhibit surprising breadth: they can solve coding puzzles, translate languages, and pass standardized exams in medicine and law. Togelius acknowledges that some recent AI systems are “generalists to a degree” . But generalist in the realm of text is not the same as general intelligence across all realms . ChatGPT might feel like it talks about everything , but it can’t perform tasks outside of generating text. It can’t directly control a robot, or genuinely understand physical space, or devise a complex plan by itself. It excels within its narrow (albeit broad-seeming) domain. In short, ChatGPT and its kin are incredibly advanced chatbots – products of narrow AI that mimic many behaviors associated with intelligence, yet fall short of the real deal that AGI promises to be.

Key Differences Between AGI and LLMs

To grasp why artificial general intelligence isn’t just another chatbot, it helps to break down some key differences between an imagined AGI and current large language models :

Understanding and Grounding: Pattern-based imitation vs. genuine comprehension. When you talk to an LLM, it gives the impression of understanding, but it doesn’t truly grasp meaning the way humans do. It lacks grounding in the physical world – it knows words statistically, but it doesn’t know why those words refer to real objects or concepts. An AGI, by contrast, would be expected to have true understanding of the concepts it deals with. For example, if an AGI reads about “elephants,” it should ideally have (or be able to acquire) some grounded knowledge of what an elephant is – what it looks like, how it behaves, how it interacts with the environment – beyond just word associations. This relates to the classic “symbol grounding problem” in AI: how to connect abstract symbols (like words) to real-world meaning. Current LLMs don’t solve this; they manipulate symbols without reference to actual experience. A true AGI would need a richer internal model of reality, built from sensory inputs or experiential learning, not just text.
Learning and Adaptability: Fixed knowledge vs. continuous learning. An LLM like GPT-4 comes pre-trained with a vast amount of knowledge, but once that training is finished, it’s relatively fixed. It can be fine-tuned, but it isn’t constantly learning each time it’s used. In contrast, one expectation of AGI is that it would learn actively and continuously . Think of how humans learn: we don’t need to be completely rebuilt to acquire a new skill or update our knowledge. We can learn on the fly, with a few examples, through trial and error, or by exploration. An AGI would similarly be able to autonomously improve and adapt to new information or situations without a complete re-engineering. This is a fundamental difference: current LLMs are like encyclopedias (albeit very interactive ones) – they’re loaded with information but not growing on their own – whereas an AGI would be more like a human student, always learning .
Goal-Directed Behavior (Agency): No self-driven goals vs. autonomous initiative. LLMs do not decide to do things on their own – they only act in response to a user prompt. They have no internal objectives or desires. An AGI, however, is often envisioned as having the ability to set goals and pursue them proactively. In other words, an AGI could have agency . It might notice a problem and figure out how to solve it without being explicitly told to do so. For instance, a household AGI might observe that the plants are wilting and decide on its own to water them. This sort of self-driven behavior is outside the scope of current LLMs (which are more like clever question-answering machines). To be fair, not every conception of AGI requires free will or autonomous initiative – some definitions would be satisfied if the AGI can do anything asked of it without initiating tasks by itself. But generally, the more human-like autonomy an AI has, the closer it approaches the intuitive idea of AGI. ChatGPT, of course, has zero autonomy: if you don’t prompt it, it does nothing. An AGI, in popular imagination, wouldn’t just sit around waiting for instructions – it would actively think and perhaps act on its own goals.

Infographic: Narrow AI (specialized intelligence) vs. General AI. Narrow AI systems excel at specific tasks (e.g. image recognition or language translation), but cannot generalize beyond those domains. AGI would combine these abilities and more into a single, general problem-solving machine, matching or exceeding the flexibility of human intelligence . The infographic above visually contrasts Artificial Narrow Intelligence (ANI) with Artificial General Intelligence (AGI) , highlighting that narrow AI is characterized by a limited domain of work, repetitive or pre-defined tasks, and a lack of broader reasoning beyond its specialization. For example, a narrow AI might handle customer service queries or play chess – each is a single area of expertise. AGI, by contrast, would need to handle any task or decision a human could tackle; its scope of work would span analysis, reasoning, creativity, and learning across different contexts. Right now, all real AI we have is narrow. Even the most advanced AI models in 2025 (like GPT-4 or DeepMind’s AlphaGo) fall under narrow AI because they have clear boundaries on what they can do. AGI remains a theoretical future goal – it isn’t here yet, and chatbots like ChatGPT, while impressive, are not examples of AGI.

Misconceptions About AGI and LLMs

Given all the hype around AI, it’s no surprise that misconceptions abound. Let’s clear up a few common ones:

“ChatGPT (or GPT-4) is basically an AGI already.” This is a misunderstanding. Yes, ChatGPT is extremely advanced compared to earlier software, and it feels general because you can talk to it about almost anything. But as we’ve detailed, it lacks many qualities of a true AGI – it’s not autonomous, not grounded in the physical world, and its “intelligence” is narrow, confined to text. Julian Togelius noted that when people talk about AGI, some may think of a human-like robot or a godlike super-brain, while others might just mean “a souped-up LLM that can write really good code and convincing essays”. The term AGI is often used inconsistently. In reality, GPT-4 and its peers are amazing narrow AIs (with a broad knowledge of language), not general intelligences. Equating today’s chatbots with AGI is like mistaking a very clever parrot for a truly sapient being.
“We’ll get to AGI simply by making our current models bigger and better.” There’s a popular notion that if we just keep scaling up LLMs – more data, more parameters, more computing power – at some point a magical threshold will be crossed and poof! we’ll have AGI. While increased scale has certainly produced surprising capabilities in models (GPT-4 is far more capable than smaller predecessors), many researchers caution that size alone is not a guarantee of general intelligence . As one AI expert put it, “LLMs have stalled in the progress to AGI and increasing scale will not help what is an inherently limited technology”. In other words, simply making neural networks larger might yield diminishing returns. Achieving AGI could require new ideas and architectures, not just brute force. It’s a misconception to think AGI is just “GPT-10” or an extremely souped-up chatbot. True AGI will likely require qualitative innovations – like incorporating reasoning abilities, memory persistence, world knowledge, or other paradigm shifts – not just more layers or data.
“AGI will think just like a human, only faster.” We often anthropomorphize AI, but it’s not guaranteed an AGI would mirror human thought. It’s possible that a general AI could achieve its abilities via very different means than a human brain. For instance, some researchers (like those at DeepMind) suggest we focus on an AI’s measurable capabilities and not on how it achieves them. By that view, an AGI wouldn’t need to replicate the human brain’s structure or consciousness, as long as it can perform generally. Some people assume AGI will automatically be a conscious, self-aware entity with human-like feelings. That’s another misconception – consciousness might not be a required component for problem-solving prowess (and it’s famously hard to even define or detect). In short, AGI might not be a digital human. It could be alien in its thinking. The key is that it would be capable in a general way, not that it would have a human personality or emotions. (Conversely, it’s also not certain that AGI won’t have human-like qualities – the point is, we shouldn’t assume an AGI will be an exact human analog in software.)

Challenges and Hurdles on the Path to AGI

Why don’t we have AGI yet, and why is it so hard to create? The differences we’ve outlined hint at the difficulties. Developing a true general intelligence in machines isn’t just an engineering challenge; it’s also wrapped up in unsolved scientific and philosophical questions. Here are some of the major hurdles:

1. Defining Intelligence and Generality : It’s hard to build something when you’re not entirely sure what it is. As we saw, there’s no universally agreed-upon definition of “intelligence.” Does it mean the ability to achieve goals? To adapt to new situations? To reason abstractly? Human intelligence is complex and multifaceted, making it tricky to decide what exactly an AGI should replicate. Togelius even argues that AGI as a concept might be “undefinable,” because we don’t fully understand what human intelligence entails. If we can’t pin down what intelligence truly is, it’s not easy to program a machine to have it. This lack of clarity is a philosophical challenge – without a precise target, the goal of AGI remains somewhat fuzzy, and discussions about it can become muddled.

2. Common Sense and Background Knowledge: A lot of what we consider “easy” in human cognition is extremely hard for AI. Basic common sense – understanding that gravity makes things fall, that others have minds with their own beliefs, or that if you leave ice in the sun it will melt – is something we take for granted. We accumulate this knowledge through living in the world, but AI systems lack such innate understanding. LLMs, for instance, know of these concepts from text, but they don’t truly grasp them and can be notoriously oblivious to simple physical or social realities that a small child understands. Endowing an AI with common sense is a major challenge. It might require AI to learn from interactive experience in the world, not just from written data. Without common sense, an AGI would be brittle – it might make absurd mistakes outside of the situations it was trained on, because it doesn’t have the everyday baseline of knowledge that humans do.

3. Reasoning and Planning: Today’s AI excels at pattern recognition but still struggles with reasoning through multi-step problems or planning long-term actions. For an AGI, the ability to chain thoughts together logically and to plan ahead is crucial. Current models can follow a logic to an extent (and techniques like “chain-of-thought” prompting help a bit), but truly robust reasoning might require integrating new approaches, such as symbolic logic or more advanced forms of memory and attention. Similarly, planning – devising a sequence of actions to achieve a goal – is something humans do routinely, but AI often needs explicit programming or constrained environments to do effectively. Achieving human-level reasoning and strategic planning in an AI is an open research problem. It likely means developing architectures that can handle abstract concepts, keep track of intermediate results, and dynamically adjust plans – capabilities that current LLMs only weakly emulate.

4. Ethical and Alignment Challenges: Suppose someone does manage to assemble an AGI from all the necessary pieces – intelligence, knowledge, reasoning, learning ability, etc. A huge challenge immediately arises: How do we ensure this AGI’s goals and behavior align with human values? This is known as the AI alignment problem . Even narrow AI systems can behave in unintended ways – for example, a content recommendation algorithm might learn to promote extreme material because it maximizes clicks, even though that wasn’t the human intention. With a powerful, autonomous AGI, the stakes are even higher. We wouldn’t want an AGI to, say, pursue a goal to an extreme that harms people (the classic example: an AGI told to “make humans happy” might decide to inject everyone with a drug, interpreting happiness in a perverse way). Ensuring an AGI understands human ethics and adheres to them is both a technical and moral problem. It might involve building in ethical constraints or the ability to learn human values – an area of active research. Many leading AI thinkers are working on ways to make future general AI safe and beneficial to society, precisely because uncontrolled AGI could pose serious risks.

Conclusion: Why AGI Isn’t Just Another Chatbot

To wrap up, the phrase “ AGI vs. LLMs ” underscores the gap between where AI is today and where it might go in the future. Today’s large language models and other AI systems are narrow but increasingly versatile tools . They are not thinking machines in the holistic, human-like sense. An Artificial General Intelligence , if and when it is created, would be an entirely different beast. It wouldn’t just chat – it would learn, reason, adapt, and possibly even perceive and act in the world with a breadth and depth comparable to a human mind (or beyond).

It’s important to understand that ChatGPT is not an embryonic AGI that’s just a few tweaks away from becoming self-aware. It’s a powerful demonstration of how far narrow AI techniques can go, but it also highlights the remaining limitations. By appreciating those limitations, we can better direct our expectations and research efforts.

For now, whenever you see headlines about “AGI” being right around the corner, it’s worth remembering the nuance behind that term. The differences between our current AI systems (no matter how impressive) and a true AGI are profound. Bridging that gap will be the work of future research – perhaps even of the next generation of scientists. Large language models will certainly get better and may form one part of the blueprint for general intelligence, but they themselves are not the endgame .

In summary, AGI represents the aspiration of AI that can understand, learn, and adapt in a general way – a machine with the versatility of human intelligence . LLMs like ChatGPT are one of the pinnacles of today’s narrow AI ; they are extremely capable in their domain (language), but they lack the generality, self-directed agency, and deeper understanding that would qualify as general intelligence. By recognizing this, we can better appreciate the AI tools we have now without confusing them with science fiction, and keep sight of the real challenges that lie ahead on the journey toward genuine artificial general intelligence.