I wrote a comparison between account sharing and refilling your coca-cola drink at McDonalds.
https://t.co/O1jYHAtTDu
What do you think? Should we just let it go?
4/4
@roderik@superdoteng@NousResearch We all do. Closest i got so far was an army of Hermes agents that pick up tasks from a shared board.
Visibility, not so good (the board is the visibility).
Dude it's awesome. Some tips to save you time:
- multi-profile is essentially running 2x hermes. I would go for the latter
- if you want to try cutting edge stuff, go for master. The tagged releases are missing out on important fixes
- the kanban board for multi-agent orchestration is based on filesystem. Use something else (external) if you want to make it work across machines
If you haven't tried Sesame voice (chat 5 minutes with agent from browser) try it -> https://t.co/G7HzooMITz
It's scary good. Nice break from the voices where you can tell it's AI. Starting to sound like 'Her'
Last week i listened to an AI-generated podcast and i liked it.
You'd expect it to be terrible right? Well it was super condense, no distractions, just high-bandwidth information transfer.
Should i be ashamed now? :)
https://t.co/j9qr9PXACI
@shannholmberg I think the architecture of multi-agent in hermes currently makes this hard.
I've come around to the idea of just running a single-agent hermes install (k8s) for each agent. The UI/API/CLI on multi-agent setup is just confusing...
@jacomakes@SimonHoiberg Coolify is good
Dokploy is better imho
And for pure k8s management (coolify but on distributed cluster) there is https://t.co/m2oAVRCuZm which is promising
was messing with the OpenAI base URL in Cursor and caught this
accounts/anysphere/models/kimi-k2p5-rl-0317-s515-fast
so composer 2 is just Kimi K2.5 with RL
at least rename the model ID
I found a good use case of using AI (and an LLM) in our Product!
For a few weeks we've been trying to algorithmically determine the answer to a simple question. Is this domain (https://t.co/Rxr2YrcQG2, https://t.co/RWtI43PU9N, https://t.co/YoOs1bklbQ, https://t.co/SHUXLhvvAx) a news outlet or not?
We kept tuning the scoring, adding more data sources (SimilarWeb, Diffbot, scrape ourselves)... to solve what feels to me like a simple problem: is this a news outlet? Yes or no?
Turns out that gpt4o (although I had to query it 3x) makes all our efforts obsolete.
With this new information we can start extracting, suggesting and hinting media outlets throughout the contact database.
#buildinpublic
"I'd much rather live in a country that embraces everyone's right to BE FULL OF SHIT than one that pretends it can declare a priori what's true and what's false." https://t.co/hLTDg4I8ur
hey @ahrefs your content explorer is 🔥. I find myself using the 'news' tab more and more.
But according to the docs you pre-approve news domains. Is there any way we can check that list of pre-approved domains? Or any other source you use to identify news?
Does anyone really think that wirecutter reviews are unbiased? and not influenced by affiliate income?
I mean look at this. It's well written. But are we supposed to believe that them missing out on affiliate income for returned purchases is the prime incentive for them to remain objective?