"Write python code for a genetic algorithm that does neural network architecture and hyperparameter search for a ViT based computer vision classifier."
👇
♊️ GEMINI:
Link to chat:
My analysis:
- Not a long preamble.
- It used a genetic algorithm library for python "deap".
- It parameterized a good set of hyperparameters for the vision transformer, particularly patch size which is crucial. g.co/bard/share/980…
♊️ GEMINI cont'd:
- It used the deap's built-in functions for defining the genes
- It used deap's crossover and mutation
- It did not define the fitness function, just used a random generator, even though I said to use a classifier in the prompt
- It did not use any sample data
More and more companies will start posting jobs for “prompt engineers”. Fortunately you don’t have to be smart to be a good prompt engineer, the smarts are what’s generated by AI! You just need to ask if the right questions in the right way. Giant new industry with skill levels.
Prompt engineering is inherently a creative endeavor. Those clever humans always adapt and find new roles as tech improves…
All of Earth will be involved in terraforming at some point. All political discourse will shift towards terraforming as their base topic for their silly social issues, you just wait, I know I’m right.
Right now the political topics focus on trannies and AI and censorship. These topics aren’t going away but will be drowned out by the next topics future generations will start caring about. Think further ahead than next decade, anon.
“As people get old, they get out of touch!” not if you define the future early.
At this point I’m just asking GPT-4 to just teach about more about the inner workings of neural networks and how to modify them. Nice recursive feedback loop.
This seems, to me, to be the most intelligent thing I can do with this new technological marvel.