One condition of testing ai bots is to post crap in your channel. While I go and test the summarizer, dev is improving the QA bot. Action plan is ready and approved.

0 reply

0 recast

2 reactions

An update is due... QA assistant - We gave ourself two weeks to make improvements. There are surface level fixes we can do (single loop learning) or going deeper (double loop learning). So far we've done the single loop and fixed narrow use cases. Now, we're aiming for more systematic changes. Summarizer It was working on staging, but only returned the summary that's part of our existing flow and created at midnight UTC. Now that we've added real-time summary, it's broken. Good lesson to keep pushing to production first before making tiny improvements.

0 reply

0 recast

5 reactions

Testing making summaries of conversation. Flow 1 (in group chat): /summary --> get summary of the last 24 hours of conversation. We're also still improving the qa functionality. Ideally the flow will be this (in a dm with the bot): /summary of conversation in group ABC --> get summary --> ask follow up question about posts, people, links that have been shared

1 reply

0 recast

0 reaction

Solana projects are starting to take community seriously. Loads of new infra project building community. In the past months a lot of projects not building on Ethereum joined us. The future is multi-colored

0 reply

0 recast

3 reactions

Dev asking the important question at 7:30am

1 reply

0 recast

2 reactions

ok, where are we with the chatbot? After some weekend testing, we decided we made it too strict. There are now questions the bot isn't answering. But this is a lesser evil than being spammy. Bugs we're fixing this week: Telegram pipeline isn't running consistently. This means the ai chatbot doesn't learn consistently. Improving we are making next: The chatbot should add useful references. We've debated calculating cognitive complexity or information density, and decided information density will serve our goal better. (ttps://publish.obsidian.md/pkc/Hub/Theory/Sciences/information+density) In testing: web scrapper

0 reply

0 recast

0 reaction

If you can get a summary of what has been posted in one of your telegram groups while you were sleeping, how should the bot behave to be kicked out of your group instantly (or super fast). Trying to avoid getting banned 😬

0 reply

0 recast

0 reaction

Today I created 4 videos while testing our web scrapper. Conclusion: it works except that one weird time where I got errors but couldn’t recreate it Soon coming to your Telegram - less spammy bot, don’t jumps into conversation. Also doesn’t reply to “how do you do” questions - make your bot smart from day 1 by giving it a website I soooooooooooo hope all is in production tomorrow

1 reply

0 recast

0 reaction

perplexity told me to update our website because it can't verify the features I want it to talk about for our singularity net marketplace integration. wonderful. more work. www.togethercrew.com/ai-assistant

0 reply

0 recast

0 reaction

When you develop AI agents you get to ask stupid questions.

2 replies

0 recast

2 reactions

Improved reputation page out. tldr: The score measures general awareness and knowledge about the community. It is *not* an engagement metric. Long story: Imagine a community to be a city. If you always stick to your neighborhood, you know a lot about it, but jack shit about the rest of the city. You have a low reputation score. Now, walk beyond your known limits, discover another neighborhood and you end up with a better awareness and knowledge about your city. Your reputation increases. Engagement only measures activity. Activity can easily be gamed. Activity also doesn't measure how much someone knows about the community. Our reputation score measures if a person has friends in "other corners" of the community.

2 replies

0 recast

3 reactions

Couple of cool features are coming - better reputation page: We stripped it down to the basic. Get reputation scores for Discord, Telegram, Github for your community - web scrapper. I might be the most excited person for this, but this will make our Telegram ai chat 100x smarter (assuming you have great docs of course) https://github.com/TogetherCrew/frontend/pull/399

0 reply

0 recast

2 reactions

I’m creating a demo telegram group for people to test our ai assistant. What should our ai know? Still me your favourite yet articles, jokes and projects

1 reply

0 recast

1 reaction

It’s Monday morning. You woke up deciding that this week will be *great*. Big smile and full of hope, you dig into your super healthy oats. You don't even look at your phone, hitting another goal: digital-free food! The clock strikes 8:33 and full of magical vibes you open Telegram to check out your company’s group channel. Your world comes crashing down. All the magic and vibes disappear and you sprint off to get pancakes with chocolate and whipped cream while screaming “The link to the faucet is on our website. Argggg” Is this you?

0 reply

0 recast

0 reaction

Where we are with our ai agent: - we’ve added a supervisor agent. It’s work is to decide if the question should be answered or ignored. It’s in staging throwing weird bugs - product discussion: is a talent discovery platform highlighting devs based on their GitHub’s a worthwhile path? If you have some thoughts about that, dc me

0 reply

0 recast

1 reaction