Nice to see we all building similar things, each in our own ways, which makes them somewhat different.
That's my way to say I too built an orchestrator like this and reached into a similar flow. Personal use only but now I'm considering publishing it somehow.
Pretty interesting movement to be honest, but having more soybean internally would probably be better for them, silage is not a miracle, protein still has to come from somewhere, can't ferment whatever and expect it to fill soy's void.
Apart of the operational challenges mentioned in the article, of course, making it happen isn't trivial, it has costs as well.
Yeah, mine did, when chatgpt 3.5 came out I thought it was a fun incoherent toy and dedicated all of 10 minutes to it.
Last year I started using it as a copilot and in the second semester I started building AI/LLM based sales and marketing systems that now are most of my income.
I'm becoming one of those "and this is just the beggining for AI" enthusiasts :P
Ooh this sounds good, I feel the same problem but never imagined a solution, having a group chat sounds good, any chance this can be integrated in something like Antigravity?
"Time to kick off that new priority in your company!"
Web developer, mainly frontend and application logic, with basic infra and backend skills to get my work online.
Examples:
AI systems from scratch (no Langchain and similar), Interactive app screens, PDF generators, payment processing, maps and routes, VoIP, browser extensions, SFTP server client and more.
Stack: SvelteKit, Microsoft Azure functions, Netlify functions, AWS Lambda functions, Node, Docker, Appwrite as a full backend (Supabase alternative), Caddy as a front door (Nginx alternative).
There will always be someone whose job is to program computers to do things.
That's us, developers.
That will never change.
We're the ones dedicated to it.
Execs, managers, HR, salesmen, designers etc won't suddenly want to spend their whole days, not even half of their time, tinkering with a computer so it can do what they want.
Else Basic and Fortran would have made everyone software developers.
It's not a zero sum game. Think, an agronomist visits a farm, instructs to cut a certain plant for the animals to eat at a certain height instead of whenever, the plant then provides more food for the animals to eat exclusively due to that, no other input in the system, now the animals are cheaper to feed, so more profit to the farmer and cheaper food to people.
For JSON I agree, now I can just mention JSON and provide examples and the response always comes in the right format, but for tool calling and information retrieval I have never seen a system actually work, nor in my tests have these worked.
Now, I'm open to the idea that I am just using it wrong, but I have seen several reports around the web that the most that people got in tool calling accuracy is 80%, which is unusable for any production system, also for info retrieval I have seen it lose coherence the more data is available overall.
Is there a model that actually achieved 100% tool calling accuracy?
So far I built systems for that myself, surrounding the LLM, and only like this it worked well in production.
There is an experiment in cows in which they swapped the cow's rumen contents, which is a big gut with way more microorganisms than the regular gut, and their microbiomes reverted back to their original profiles, which is different to these findings in mice here.
Fully agree. I personally had clients backing off from entire projects due to not being able to ship some native like experience on iOS, only Android, for example the prompt to install as a progressive web app.
I'm tired of losing sales due to that. And clients needing to increase their costs massively due to that.
I was waiting for fdroid's voice about this. Google's move is as bad as I initially thought.
This makes me a bit sad honestly, android development is getting worse every year.
I wonder if the same will happen to web as well.
That's my way to say I too built an orchestrator like this and reached into a similar flow. Personal use only but now I'm considering publishing it somehow.