AI Development general thread.

semaphore said:
I'm mostly using Luna Max and Sol medium for review.

View attachment 1921133

That was sol high the other night reviewing and putting together documentation and plans (11 hours lol).
Click to expand...
Was mainly on Sol High yesterday since Tibo mentioned he'd be resetting us consistenly. No issues with Luna max for implementation?

Hope he comes through again since I tested Ultra, bug in Codex is real though. Sub-agents use the main agent model and thinking level. Using Claude gets around this though lol for workflows.
 
Couple horror stories coming through of Sol wiping put local stuff on Mac.

Of course this is folks going full bore having allowed Carte Blanche to thier machine.
 
Fable this, fable that.

I’m getting TONs of high quality code out using Sonnet while I go play golf 🤷‍♂️

But I’m obviously spending “a lot” of up front time planning/designing - but this always happened pre AI anyway, it was just done in my head.

Of course I’m not building Netflix/highly distributed systems, but rather mostly “line of business”
 
_kabal_ said:
Fable this, fable that.

I’m getting TONs of high quality code out using Sonnet while I go play golf 🤷‍♂️

But I’m obviously spending “a lot” of up front time planning/designing - but this always happened pre AI anyway, it was just done in my head.

Of course I’m not building Netflix/highly distributed systems, but rather mostly “line of business”
Click to expand...
You will enjoy Sol then. Actually knocked out a few issues I was wracking my head over
 
Been doing a bunch of experimentation now with Codex (now ChatGPT app) with 5.5/5.6 and I can see any number of roles being dominated by design/planning/verification/testing and very little else. If you're a "versatilist" and/or have a deep understanding of any process/business/industry you're going to rule the world.
 
Having a good time with Codex with their infinite resets and Claude running the sub-agents. Just need OpenAI to fix the sub-agent thinking levels and which model they should use.
 
Pho3nix said:
Having a good time with Codex with their infinite resets and Claude running the sub-agents. Just need OpenAI to fix the sub-agent thinking levels and which model they should use.
Click to expand...
Easy to resolve. Add it to your agent.md.
 
quovadis said:
My experimentation has been mostly over July...

View attachment 1922779
Click to expand...

You can tell when you really started enjoying Codex
3efba6c008bc52ab1b6f9a23345a5b52.jpg
 
Anyone looked at K3? Initial thoughts are just like DeepSeek and Grok there are literally no safety guards which I appreciate. Just need it on Foundry to test properly
 
Pho3nix said:
Anyone looked at K3? Initial thoughts are just like DeepSeek and Grok there are literally no safety guards which I appreciate. Just need it on Foundry to test properly
Click to expand...
Yep, played with it, good model. But i see qwen 3.8 max is also coming out so we getting spoilt for choice.
 
Been using Clause for the past few months now. Working really well for me at this point developing an enterprise grade SaaS. Back in the early stages i switched between a few AI tools but that turned into a disaster, so i'm now just sticking to 1 to complete the dev.
 
I've been working on something in this space too primarily private/local LLM workflows with RAG across a mix of structured data, documents and operational records.

Current architecture is API first, with a Python backend, local model serving, vector retrieval, and a web based interface. I'm also exploring agent style workflows where the model can retrieve relevant context, query approved data sources, and return a traceable response rather than simply act as a document chatbot.

My main priority is making it useful in real operations: permission-scoped access, source traceability, audit logging, and keeping sensitive data within my own environment. Still early enough that I 'm keeping the exact stack and implementation details private, but there is definitely practical value beyond the usual “chat over PDFs” use case.
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X