@radekosmulski.bsky.social
4. execnb
• Run nbs as scripts with --verbose flag (PR#75)
• Pass args as env vars
• Infinitely flexible building block for custom behavior
• My hack: import from notebooks up to STOP_WHEN_IMPORTED marker - lets me keep adding/testing below while stable code stays at top
• Run nbs as scripts with --verbose flag (PR#75)
• Pass args as env vars
• Infinitely flexible building block for custom behavior
• My hack: import from notebooks up to STOP_WHEN_IMPORTED marker - lets me keep adding/testing below while stable code stays at top
July 4, 2025 at 1:20 AM
4. execnb
• Run nbs as scripts with --verbose flag (PR#75)
• Pass args as env vars
• Infinitely flexible building block for custom behavior
• My hack: import from notebooks up to STOP_WHEN_IMPORTED marker - lets me keep adding/testing below while stable code stays at top
• Run nbs as scripts with --verbose flag (PR#75)
• Pass args as env vars
• Infinitely flexible building block for custom behavior
• My hack: import from notebooks up to STOP_WHEN_IMPORTED marker - lets me keep adding/testing below while stable code stays at top
3. nbimporter
• Import from notebooks like py files
• Author nailed it, then overthought it 🤷♂️
• Key insight: best practices depend on context - building libs ≠ rapid prototyping/exploration
• Import from notebooks like py files
• Author nailed it, then overthought it 🤷♂️
• Key insight: best practices depend on context - building libs ≠ rapid prototyping/exploration
July 4, 2025 at 1:20 AM
3. nbimporter
• Import from notebooks like py files
• Author nailed it, then overthought it 🤷♂️
• Key insight: best practices depend on context - building libs ≠ rapid prototyping/exploration
• Import from notebooks like py files
• Author nailed it, then overthought it 🤷♂️
• Key insight: best practices depend on context - building libs ≠ rapid prototyping/exploration
2. Claude Code in VS Code
• Great at modifying notebooks (needs File Reloader)
• Reads relevant files - RAG done right, huge value add
• Skip conceptual work - use for mundane tasks & info sourcing
• You drive, CC executes
• Great at modifying notebooks (needs File Reloader)
• Reads relevant files - RAG done right, huge value add
• Skip conceptual work - use for mundane tasks & info sourcing
• You drive, CC executes
July 4, 2025 at 1:20 AM
2. Claude Code in VS Code
• Great at modifying notebooks (needs File Reloader)
• Reads relevant files - RAG done right, huge value add
• Skip conceptual work - use for mundane tasks & info sourcing
• You drive, CC executes
• Great at modifying notebooks (needs File Reloader)
• Reads relevant files - RAG done right, huge value add
• Skip conceptual work - use for mundane tasks & info sourcing
• You drive, CC executes
1. VS Code with Jupyter notebook
• Only way I know to get working LLM completions in notebooks - game changer
• Essential extensions: File Reloader + File Utils
• Custom keybindings for kernel restart/cell execution are a must
• Everything on one screen 🤩 (ssh, tmux, nbs)
• Only way I know to get working LLM completions in notebooks - game changer
• Essential extensions: File Reloader + File Utils
• Custom keybindings for kernel restart/cell execution are a must
• Everything on one screen 🤩 (ssh, tmux, nbs)
July 4, 2025 at 1:20 AM
1. VS Code with Jupyter notebook
• Only way I know to get working LLM completions in notebooks - game changer
• Essential extensions: File Reloader + File Utils
• Custom keybindings for kernel restart/cell execution are a must
• Everything on one screen 🤩 (ssh, tmux, nbs)
• Only way I know to get working LLM completions in notebooks - game changer
• Essential extensions: File Reloader + File Utils
• Custom keybindings for kernel restart/cell execution are a must
• Everything on one screen 🤩 (ssh, tmux, nbs)
Journey before destination, not because the destination doesn't matter.
But because the journey is all we have access to.
Planning gives us only a false sense of security, but robs us of experimenting — our only hope to interact with the terrain and not the map in our minds.
But because the journey is all we have access to.
Planning gives us only a false sense of security, but robs us of experimenting — our only hope to interact with the terrain and not the map in our minds.
June 1, 2025 at 5:47 AM
Journey before destination, not because the destination doesn't matter.
But because the journey is all we have access to.
Planning gives us only a false sense of security, but robs us of experimenting — our only hope to interact with the terrain and not the map in our minds.
But because the journey is all we have access to.
Planning gives us only a false sense of security, but robs us of experimenting — our only hope to interact with the terrain and not the map in our minds.
And the uncertainty in our individual lives is only amplified!
There are so many random factors at play. Plus, our ability to understand our situation and what is possible is minimal!
Our mental map is not the terrain.
Hence, the best we can do is give in to the process.
There are so many random factors at play. Plus, our ability to understand our situation and what is possible is minimal!
Our mental map is not the terrain.
Hence, the best we can do is give in to the process.
June 1, 2025 at 5:46 AM
And the uncertainty in our individual lives is only amplified!
There are so many random factors at play. Plus, our ability to understand our situation and what is possible is minimal!
Our mental map is not the terrain.
Hence, the best we can do is give in to the process.
There are so many random factors at play. Plus, our ability to understand our situation and what is possible is minimal!
Our mental map is not the terrain.
Hence, the best we can do is give in to the process.
We can think in a similar vein of discoveries on a personal level:
I have stumbled into so many things that work:
• kaggling
• sharing my work
• shipping learning project after learning project
One could probably craft a process of self-innovation out of these pieces.
I have stumbled into so many things that work:
• kaggling
• sharing my work
• shipping learning project after learning project
One could probably craft a process of self-innovation out of these pieces.
June 1, 2025 at 5:46 AM
We can think in a similar vein of discoveries on a personal level:
I have stumbled into so many things that work:
• kaggling
• sharing my work
• shipping learning project after learning project
One could probably craft a process of self-innovation out of these pieces.
I have stumbled into so many things that work:
• kaggling
• sharing my work
• shipping learning project after learning project
One could probably craft a process of self-innovation out of these pieces.
So, where do we even start to continue making progress?
We must appreciate that we might know very little.
And that sometimes the best measure of making progress is whether we are working on something interesting or not.
That interestingness, curiosity, is the signal.
We must appreciate that we might know very little.
And that sometimes the best measure of making progress is whether we are working on something interesting or not.
That interestingness, curiosity, is the signal.
June 1, 2025 at 5:46 AM
So, where do we even start to continue making progress?
We must appreciate that we might know very little.
And that sometimes the best measure of making progress is whether we are working on something interesting or not.
That interestingness, curiosity, is the signal.
We must appreciate that we might know very little.
And that sometimes the best measure of making progress is whether we are working on something interesting or not.
That interestingness, curiosity, is the signal.
By extension, we don't know what our current tech can be used for. There are still so many applications to be discovered.
That is why the work of such labs as answer.ai is so crucial.
That is why the work of such labs as answer.ai is so crucial.
Answer.AI - Practical AI R&D – Answer.AI
Practical AI R&D
answer.ai
June 1, 2025 at 5:46 AM
By extension, we don't know what our current tech can be used for. There are still so many applications to be discovered.
That is why the work of such labs as answer.ai is so crucial.
That is why the work of such labs as answer.ai is so crucial.
But even once we discover something, we might not know what it is useful for!
I am 100% certain that the inventors of the Transformer were as surprised as everyone by the ChatGPT moment.
I suspect that to them, the arch was at best intriguing and warranted tinkering more with.
I am 100% certain that the inventors of the Transformer were as surprised as everyone by the ChatGPT moment.
I suspect that to them, the arch was at best intriguing and warranted tinkering more with.
June 1, 2025 at 5:46 AM
But even once we discover something, we might not know what it is useful for!
I am 100% certain that the inventors of the Transformer were as surprised as everyone by the ChatGPT moment.
I suspect that to them, the arch was at best intriguing and warranted tinkering more with.
I am 100% certain that the inventors of the Transformer were as surprised as everyone by the ChatGPT moment.
I suspect that to them, the arch was at best intriguing and warranted tinkering more with.
In fact, the path leading from point A to point B can be so weird that we might never entertain the thought of taking it!
I struggled for years in learning Deep Learning until, out of desperation, I followed advice that seemed insane to me.
x.com/radekosmuls...
I struggled for years in learning Deep Learning until, out of desperation, I followed advice that seemed insane to me.
x.com/radekosmuls...
June 1, 2025 at 5:46 AM
In fact, the path leading from point A to point B can be so weird that we might never entertain the thought of taking it!
I struggled for years in learning Deep Learning until, out of desperation, I followed advice that seemed insane to me.
x.com/radekosmuls...
I struggled for years in learning Deep Learning until, out of desperation, I followed advice that seemed insane to me.
x.com/radekosmuls...
That is why the research off the beaten track is so exciting.
It can bring us closer to ASI, but above all, it explores the search space versus being stuck in a single, narrow corridor.
It can bring us closer to ASI, but above all, it explores the search space versus being stuck in a single, narrow corridor.
June 1, 2025 at 5:46 AM
That is why the research off the beaten track is so exciting.
It can bring us closer to ASI, but above all, it explores the search space versus being stuck in a single, narrow corridor.
It can bring us closer to ASI, but above all, it explores the search space versus being stuck in a single, narrow corridor.
These people weren't stupid!
They just didn't understand the nature of innovation.
In their thinking, they didn't place enough weight on the unknown, the unknowable to them, despite their recent outstanding technological achievements.
They just didn't understand the nature of innovation.
In their thinking, they didn't place enough weight on the unknown, the unknowable to them, despite their recent outstanding technological achievements.
June 1, 2025 at 5:46 AM
These people weren't stupid!
They just didn't understand the nature of innovation.
In their thinking, they didn't place enough weight on the unknown, the unknowable to them, despite their recent outstanding technological achievements.
They just didn't understand the nature of innovation.
In their thinking, they didn't place enough weight on the unknown, the unknowable to them, despite their recent outstanding technological achievements.
If we are to be honest with ourselves, we cannot presume to know the path to ASI until we reach it.
Ever since the 1970s, we have been 3 years from ASI.
However, this idea extends even further into the past!
Ever since the 1970s, we have been 3 years from ASI.
However, this idea extends even further into the past!
June 1, 2025 at 5:46 AM
If we are to be honest with ourselves, we cannot presume to know the path to ASI until we reach it.
Ever since the 1970s, we have been 3 years from ASI.
However, this idea extends even further into the past!
Ever since the 1970s, we have been 3 years from ASI.
However, this idea extends even further into the past!
Precisely because our mental map is so unrepresentative of the terrain!
If the current leading paradigm of autoregressive transformers does not lead to ASI, maximising scores on various benchmarks will not get us there.
If the current leading paradigm of autoregressive transformers does not lead to ASI, maximising scores on various benchmarks will not get us there.
June 1, 2025 at 5:46 AM
Precisely because our mental map is so unrepresentative of the terrain!
If the current leading paradigm of autoregressive transformers does not lead to ASI, maximising scores on various benchmarks will not get us there.
If the current leading paradigm of autoregressive transformers does not lead to ASI, maximising scores on various benchmarks will not get us there.
If we understood the world around us, we wouldn't have to worry at all about getting to point B.
We would get there effortlessly.
But acting on our current understanding is not sufficient. It will just lead to us getting stuck.
Why is that?
We would get there effortlessly.
But acting on our current understanding is not sufficient. It will just lead to us getting stuck.
Why is that?
June 1, 2025 at 5:46 AM
If we understood the world around us, we wouldn't have to worry at all about getting to point B.
We would get there effortlessly.
But acting on our current understanding is not sufficient. It will just lead to us getting stuck.
Why is that?
We would get there effortlessly.
But acting on our current understanding is not sufficient. It will just lead to us getting stuck.
Why is that?
In fact, the path leading from point A to point B can be so weird that we might never entertain the thought of taking it!
I struggled for years in learning Deep Learning until, out of desperation, I followed advice that seemed insane to me.
x.com/radekosmuls...
I struggled for years in learning Deep Learning until, out of desperation, I followed advice that seemed insane to me.
x.com/radekosmuls...
June 1, 2025 at 5:46 AM
In fact, the path leading from point A to point B can be so weird that we might never entertain the thought of taking it!
I struggled for years in learning Deep Learning until, out of desperation, I followed advice that seemed insane to me.
x.com/radekosmuls...
I struggled for years in learning Deep Learning until, out of desperation, I followed advice that seemed insane to me.
x.com/radekosmuls...
That is why the research off the beaten track is so exciting.
It can bring us closer to ASI, but above all, it explores the search space versus being stuck in a single, narrow corridor.
It can bring us closer to ASI, but above all, it explores the search space versus being stuck in a single, narrow corridor.
June 1, 2025 at 5:46 AM
That is why the research off the beaten track is so exciting.
It can bring us closer to ASI, but above all, it explores the search space versus being stuck in a single, narrow corridor.
It can bring us closer to ASI, but above all, it explores the search space versus being stuck in a single, narrow corridor.
These people weren't stupid!
They just didn't understand the nature of innovation.
In their thinking, they didn't place enough weight on the unknown, the unknowable to them, despite their recent outstanding technological achievements.
They just didn't understand the nature of innovation.
In their thinking, they didn't place enough weight on the unknown, the unknowable to them, despite their recent outstanding technological achievements.
June 1, 2025 at 5:46 AM
These people weren't stupid!
They just didn't understand the nature of innovation.
In their thinking, they didn't place enough weight on the unknown, the unknowable to them, despite their recent outstanding technological achievements.
They just didn't understand the nature of innovation.
In their thinking, they didn't place enough weight on the unknown, the unknowable to them, despite their recent outstanding technological achievements.
If we are to be honest with ourselves, we cannot presume to know the path to ASI until we reach it.
Ever since the 1970s, we have been 3 years from ASI.
However, this idea extends even further into the past!
Ever since the 1970s, we have been 3 years from ASI.
However, this idea extends even further into the past!
June 1, 2025 at 5:46 AM
If we are to be honest with ourselves, we cannot presume to know the path to ASI until we reach it.
Ever since the 1970s, we have been 3 years from ASI.
However, this idea extends even further into the past!
Ever since the 1970s, we have been 3 years from ASI.
However, this idea extends even further into the past!
Precisely because our mental map is so unrepresentative of the terrain!
If the current leading paradigm of autoregressive transformers does not lead to ASI, maximising scores on various benchmarks will not get us there.
If the current leading paradigm of autoregressive transformers does not lead to ASI, maximising scores on various benchmarks will not get us there.
June 1, 2025 at 5:46 AM
Precisely because our mental map is so unrepresentative of the terrain!
If the current leading paradigm of autoregressive transformers does not lead to ASI, maximising scores on various benchmarks will not get us there.
If the current leading paradigm of autoregressive transformers does not lead to ASI, maximising scores on various benchmarks will not get us there.
If we understood the world around us, we wouldn't have to worry at all about getting to point B.
We would get there effortlessly.
But acting on our current understanding is not sufficient. It will just lead to us getting stuck.
Why is that?
We would get there effortlessly.
But acting on our current understanding is not sufficient. It will just lead to us getting stuck.
Why is that?
June 1, 2025 at 5:46 AM
If we understood the world around us, we wouldn't have to worry at all about getting to point B.
We would get there effortlessly.
But acting on our current understanding is not sufficient. It will just lead to us getting stuck.
Why is that?
We would get there effortlessly.
But acting on our current understanding is not sufficient. It will just lead to us getting stuck.
Why is that?