autosocial

Automatically posts to all desired socials, finetuned LLM writes appealing/good titles and descriptions for each platform.

GUI:

Watch video:

> [!Note] I will probably not be using this as I post things in the future- although it was a cool/fun project! Personally, I beleive that posts should be made with care and intention- although this could be a glimpse into how future things will be done online.

Goals

I want to create an application that takes a blog post you wrote, writes an appealing, but not over-the-top summary, and then shares this on various techical ‘social media’ websites.

The platforms it posts to:

• Hackernews
• Reddit
• X
• Discord (multiple servers)

–> For future

• Youtube (shorts,post)
• Instagram (shorts,post)
• Medium/Other (re-upload?)

in the future, potentially we could use a video generation model to take our article and use a model to create a script, and video based on the article that is short and appealing, and serves to get the watcher interested in the article/share its true content.

Eventually, also I’d like it to automatically gather metrics.

Overview

The code is in posting, and is seperated in the following way:

• post.py: post a given URL and content to each social using playwright to automate browser actions -claude_summ.py: Sends API call to Sonnet 3.7 and summarizes the website; takes instructions.txt and extra_instructions.txt and modifies the tweet file for post.py to acess
• guinter.py: code for the gui that takes information about where to post/extra instructions to add
• markdowndownload.py takes the url, downloads the raw html, uses html2text (awesome) to convert html to markdown so an LLM can read it
• pipeline.py puts it all together: settings w/ GUI, extract w/html2text, summarize with claude, post with playwright!

Conclusion: Working with AI like future

This was a fun project to build, and my first project that involved using API calls to an LLM, rather than dealing with the building of the model itself.

Honestly, it is shocking how good LLM’s are (im using claude 3.7 sonnet), and I think that their understanding and use of language will soon make tasks like this (posting/dealing with sharing your work) almost always be done automatically.

Atlhough I definitely feel resistance to this, as sharing things does not require much effort, and sharing is really a part of the research/building. This trade off between efficiency and loss of actual thinking/doing things will definitely be a problem as we move forward with better AIs.

I think, however, that humans have a remarkable ability to adapt, and to learn what is necessary. Caveman no doubt would frown upon the average person in their lack of ability to distinguish poisinous berries from safe ones, but today we can focus on more meaningful things (such as learning about technology and researching!). So I think that as long as we move forward and continue learning/building, we will quickly forget the irrelevant skills that can be automated, and learn those that are necessary to navigate the new world.