There’s no tinker: there’s only picking a specific category - and destroying it.
Plus, don’t get “special consideration” from OpenAI, or be DQ’d.
At least this time they went through a contest administrator. Who got those Christmas radios for tweeting from the last one, with 24 hours to make something?
Probably the wrong model for my prompt -> SPICE -> autorouted Gerber wishes…You aren’t going to smash any benchmarks.
I’d give advice “think locally” in general. The local category is a subset of that, the VC buzzword “agent” without using internet (aka no API services). And the model is knowledge-bound, so it’s not going to be filling in the lyrics fields of your MP3s.
Certainly. It sounds like the new oss models are operating at o4-mini-ish levels for most tasks, with fine tuning and training making it more and more appropriate for a specialized task over time.
Though I don’t envision an internet-agnostic scenario, personally; not that I don’t think it’s useful. Guess I’ll have to look into the definitions… because wouldn’t you want to fine tune a local agent?
Just my thread here from the last forum hackathon thread, a simple local Agentic system a bit like ‘Operator’ with the ability to share self-describing agents packed in GIFs.
It’s not very professional like you guys would do though, it’s a hobby project.
Feel free to use it / repurpose it for your project if it’s in any way useful though.
I don’t use things like Github, more interested in hacks around the edges that a smart AI might take. The value for me is in sharing these concepts though rather than any prize.
Hi everyone!
I’m not a developer, but I enjoy working out ideas and system structures. I’d like to share a concept — maybe someone here will find it worth building.
Idea #1 – “Health AI”
An application that:
Accepts lab results in PDF/JPEG, converts them to text, and translates into any needed language.
Stores them in a database and analyzes via an LLM (ChatGPT, Ollama, vLLM).
Compares results with both current and historical reference ranges.
Suggests additional tests or procedures to get a complete health picture.
Uses a node-based visual interface for easy navigation and linking.
Idea #2 – “Real-time ECG Analysis”
API stream from ECG → AI analyzes waveform patterns → provides recommendations or alerts in real time.
Idea #3 – “Profile Memory”
A feature for ChatGPT (or any OSS LLM) to store a small “profile window” with user info, which can be enabled/disabled for new sessions.
Idea #4 – “Node Share”
A visual node system where each block can be code, a function, or a prompt. Users can share ready-made node chains with each other.
If anyone here would like to expand on these ideas — I’d be happy if they help someone’s project!
I am finding team members. I am a Data Scientist, interested in LLM, RL and Causal Inference. Let me know if you have similar interests. We can brainstorm ideas together.