i feel like python is very limiting for prompt engg/agents. anyone else feel this way? maybe we need more expressive languages/dsls/languages built with editor in mind (like smalltalk).
10/15/2023, 3:39 AM
Like how even Python experts prefer shell languages when operating their computer interactively, line-by-line?
10/16/2023, 1:22 AM
Yup, sth like that
10/19/2023, 3:05 PM
Just curious, what are the alternatives? Which language would you WANT to use for prompt engineering?
By all means dunk on Python, but I'd love to hear what your ideal looks like 🙂
10/19/2023, 3:25 PM
ofc, dunking w/o alternative is mid
rn i am developing a forth-variant dsl for prompt engg
been demoing to dozens of ppl with good feedback!
also open to demoing
10/19/2023, 4:57 PM
devlog it up
10/26/2023, 6:46 AM
Elixir has a good ecosystem for building ML + Web apps including agents all in a single language.
We are currently a team of Python and Typescript developers, but were discussing Elixir this week and liked what we saw so far.
• Has concurrency and fault-tolerance built-in (WhatsApp backend uses the same VM)
• Built-in support for async processes simplifies the stack. Don't need separate process (like Celery) for queues or background jobs or recurring jobs.
• Already has an Agent construct which is a wrapper around state.
• Numerical Elixir lets it invoke models directly (using GPUs while supporting batching, partitioning and streaming)
• Phoenix LiveView (which uses Websockets to stream HTML down to the client) simplifies creation of dynamic webapps while keeping all state management on the server. Phoenix is the web application framework.
• Ecto provides data validation and persistence
• Livebook is an interactive notebook
• Bumblebee - provides easy access to pre-trained models
A couple of demos to showcase the capabilities of this stack: