"pigeon" and "dove" are both words for the same family of birds. The bird most people think of with the word "pigeon" is the rock dove (https://en.wikipedia.org/wiki/Rock_dove) or domesticated / feral variants of it.
I don't think they use the agentmail domain for sending emails. Users connect their own domain and manage reputation (similar to all the other email marketing tools)
We build "long running" email agents. But it's not really long running in the sense of an agent taking 1000's of actions in a giant loop.
It's more "long running" because the agent takes 4 steps, then waits a week for the user to email it back. We might have a successful client exchange that takes a month, but for the Agent it's 99% just waiting for the next user reply.
Harder than you would expect! Since we tried this ourselves before switching to Agentmail. Threads, attachments, ccing, DNS management, sending to gmail vs outlook vs yahoo, etc. It add up to be a major pain.
By that logic why send email newsletters when I could hire 10 or 100 people email them manually instead? Obviously there's a cost tradeoff here where it's worth it to have email negotiation in an automated way, but not in a human call center way.
I think it's easy to blame the evil profit maximizing social media companies. But IMO even the most simple 'engagement' algorithm will produce negative externalities. Regardless of who's running it.
```
show_me_posts_people_like_me_have_liked()
- John saw 20 posts today and liked 9 of them.
- Cliff saw 20 posts today and liked 9 of them
- Jeff and Cliff had 6 overlapping likes
- Show Jeff the 2 extra posts Cliff liked; show Cliff the 2 extra posts Jeff liked
```
This seems like a simple / logical recommendation system. BUT the end result is that you make Jeff and Cliff closer to the same person over time. And times millions, you build echo chambers. And the biggest echo chambers (often those aligned with some identity politics) see they have a huge community and want to expand it. Making the whole platform worse as a byproduct.
Not the founder, but having run conversational agents at decent scale, I don't think the cost actually matters much early on.
It's almost always better to pay more for the smarter model, than to potentially give a worse player experience.
If they had 1M+ players there would certainly be room to optimize, but starting out you'd certainly spend more trying engineer the model switcher than you would save in token costs.
I agree, trying to save on costs early on is basically betting against things getting better. Not only that but in almost every case people prefer the best model they can get!
Not only that but I think our selling point is rewarding creativity with emergent behavior. I think baked dialogue would turn into traditional game with worse writing pretty quick and then you got a problem.
For example, this AI game here does multiple choices with a local model and people seem a bit mild about it.
We could use it to cache popular QA, but in my experience humans are insane and nobody ever says even remotely similar things to robots :)
> How does this solve any tangible problems with LLMs regurgitating someone else's work?
I'm not the OP, but here's my from-the-hip answer: if weights are public, building and operating an LLM is no longer a business plan in and of itself, as anyone could operate the same LLM. Therefore companies like OpenAI will be disincentivized from simply redirecting web traffic to their own site.
I didn't really put out the GPL push. The best I could say is that at least that information would be available to everyone rather than being tightly controlled by the company that stole the source material to create it in the first place. It might also dissuade LLM creators from mass piracy as a competitor could take their models and start hosting them.
It looks like there’s a download link that contains the source code. Presumably you untar it, follow any necessary build instructions, and then run it.
reply