More Than Just Pattern Matching
Neural nets and pattern recognition have made finding the precise solutions to problems exponentially easier. The more well-thought-out your question, the more accurate the response. It’s more than just the sum of its parts.
Beyond Basic Pattern Recognition:
- LLMs can now reason through problems, not just recall
- Can adapt to your particular style and tone
- Retain context across sessions
- Generate original content, synthesize, and help with design and coding patterns that aren’t copied from training data.
What else will be possible a year from now?
A Better Way of Learning
Because of these recent innovations, my exploration into learning through these tools feels much more intuitive and therefore easier.
As an example, I started learning TypeScript early this year by integrating it into a React App learning course called Full Stack Open. In the beginning, I sifted through basic online tutorials, which did give me a good base to start from. This approach is time-intensive and works but there is a better way.
With a basic understanding of how to integrate TypeScript into a React App, I supplemented my tutorials by feeding questions to LLMs as they arose. I quickly learned that supplying code and asking how it would implement TypeScript into provided examples was the best approach because:
- You see how to implement a particular use-case vs spending hours digging through TypeScript documentation. Both will get you where you want but one is instantaneous and the other, well, not really.
- Because your brain is an actual neural-network, learning by prompt reveals “patterns of implementation” for the given topic that facilitate understanding in a way that traditional methods cannot. This edge allows information to integrate much quicker and before you know it, you’re asking more advanced questions.
- Since LLMs retain context across sessions, you no longer need to expend extra energy note taking unless you want to. You can reference a particular prompt anytime and this frees up more time and focus.
There are plenty of other advantages but in all, supplementing an AI assistant for your topic of interest creates positive feedback loops that is a more comprehensive approach to learning. So it’s not about abandoning traditional methods (yet) but supplementing them.
Github: Lost Files
I keyed in a command line prompt I thought would push my local files to a project’s remote repo, and instead, it deleted important files. I was left trying different restoration methods but to no avail. I was only able to restore some files.
It was then that I wondered if an LLM would be able to rebuild it for me based on the partial restore, so that I wouldn’t have to start from scratch again. After only a few prompts and supplying the available files, it was able to rebuild the entire project in less than one minute. Below is the full directory tree, of which 1/3 was initially supplied by me.
This incident demonstrates how much better it is to have an AI assistant than work without one. Legacy learning and research still have there place but how much sweeter the process becomes depends on the precision of your question.