This is how I’m doing it. For critical functions/components/classes, I create test cases to ensure AI doesn’t break code in new generations. Using a TDD approach to manage risks while prompting new features.
Noticing myself adopting a certain rhythm in AI-assisted coding (i.e. code I actually and professionally care about, contrast to vibe code).
1. Stuff everything relevant into context (this can take a while in big projects. If the project is small enough just stuff everything e.g. `files-to-prompt . -e ts -e tsx -e css -e md --cxml --ignore node_modules -o prompt.xml`)
2. Describe the next single, concrete incremental change we're trying to implement. Don't ask for code, ask for a few high-level approaches, pros/cons. There's almost always a few ways to do thing and the LLM's judgement is not always great. Optionally make concrete.
3. Pick one approach, ask for first draft code.
4. Review / learning phase: (Manually...) pull up all the API docs in a side browser of functions I haven't called before or I am less familiar with, ask for explanations, clarifications, changes, wind back and try a different approach.
6. Test.
7. Git commit.
Ask for suggestions on what we could implement next. Repeat.
Something like this feels more along the lines of the inner loop of AI-assisted development. The emphasis is on keeping a very tight leash on this new over-eager junior intern savant with encyclopedic knowledge of software, but who also bullshits you all the time, has an over-abundance of courage and shows little to no taste for good code. And emphasis on being slow, defensive, careful, paranoid, and on always taking the inline learning opportunity, not delegating. Many of these stages are clunky and manual and aren't made explicit or super well supported yet in existing tools. We're still very early and so much can still be done on the UI/UX of AI assisted coding.
Recently I got attracted by KAN, a promising alternative for MLP. I decided to write about my learning with KAN, Spline, and in a single blog post, and share my thoughts about it. TL;DR: MLP is a specific form of KAN
https://t.co/TbdVPqxIos
#KAN#AI#ML#NeuralNetworks
GPT is a new kind of computer architecture that runs on text. Yes it can talk to us, but also to much of our existing software infrastructure. First via apps on top of APIs, now inside ChatGPT via plugins.
What a time right now...
https://t.co/HjeUCv3XE7
I was asked about what AI will look like in 3 decades. Reminder: it has not even been 1 decade yet since the ImageNet moment (though the anniversary is very close, imo October 13, 2022 per https://t.co/NPg2sm2Ojm). Imagining that much change, but 3X, and on an exponential is 🤯
Human tendency to conform, especially when in large groups, is terrifying. Propaganda machines leverage this throughout human history.
The way out is to think freely, detached from the divisive narratives of the day that masquerade as universal truths.
This often feels lonely.
I was interested in the derivation of Neural ODE and spent some time to go through the details and proof. Some friends suggest me to write a blog post about and why not!
#MachineLearning#NeuralODE#NeuralNetworks
https://t.co/pGoeeMwHdw
@fermatslibrary 1. Changing Magnetic field induced Electric field (eq3
2. Changing Electric field induced Magnetic field (eq4
Literally nothing:
Nothing: OSCILLATE!
@marcoleewow@flyyufelix We had tried a lot of new things on @donkey_car including using the new #JetsonNano. By setting the framerate to 60hz, the result is already better than Rpi model! We tried diff stuff including CV, ML model and hardware. Stay tuned for the details!
@Ingmar_Stapel Are you installing keras with apt-get / pip? Your sd card will need to be larger than 32gb and there are small variation across different brand. I suggest installing the package manually.