Social market creation (1/2)

[gabrielfior]

We want to have agents that can predict virality of tweets + Farcaster casts. For that, we need

• Agent posting on Farcaster/Twitter (done on #142 )
• Replicator creates markets for that post - e.g. "How many likes will the question {QUESTION_TITLE} have by {utc_now + 7days} on Twitter"? Note that this will be a binary market (for simplicity).
• Social_predictor_agent (to be developed) bets on that market.

In this task:

• Create a new replicator agent that creates markets immediately after a tweet/cast has been created by the social agent.

[kongzii]

Some notes based on my previous work in emplifi

Generally, predicting the number of likes/comments/reposts isn't possible (assuming we want reasonable accuracy). Because that would require reverse engineering of their algorithms (the extreme example is that if X decides the tweet won't have more than 100 likes, it can simply stop showing it). We were training specialised models on ton of data to do similar things, so doing it by quickly prompting LLM is even less probable.

However, I do think something similar could work. For example, we could scrape things at

• number of followers of the agent now
• number of it X days ago
• numbers of likes on the latest tweets
• etc.

And then ask something as, "Will this tweet have more than X likes?", where the X is based on the previous information.

https://github.com/gnosis/prediction-market-agent/issues/162 could be usable for scraping such things.