#bigpickle — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #bigpickle, aggregated by home.social.
-
While everyone is getting excited about #broligarchy models...
... lets spare a thought for the #China #AiI am using Zhipu AI's GLM-4.6 #GLM model in the harness (as "bigpickle") which is as good as #Anthropic #Sonnet for coding.
Its so good, that the router is actually feeding 27% (Top) of work units to #Bigpickle
-
🤖 Big Pickle: el modelo "sealth" que OpenCode ofrece gratis
Big Pickle es un modelo de lenguaje misterioso disponible en OpenCode Zen sin costo temporalmente. Nadie conoce su origen exacto, pero promete ser una opción interesante para desarrolladores.
Gratuito por tiempo limitado
Accesible via OpenCode Zen
Datos pueden usarse para mejoras
Período promocional indefinidoIdeal para probar sin compromiso antes de que empiece a cobrar.
-
Before I shut down for the night...
I now have an effective WOOFER, My unit or arbitrary LLM performance based on a Probe_prompt. The probe_prompt can be generic or run vs the mission.md for a realistic assessment of a engine efficacy.
(There is a column called WOOFER with values).Some surprising results;
#Opencode #bigpickle is only scoring 25 (out of a hundred)#Claude #Opus has assessed its own response and it thinks its a fucking genius.
But also nematron (The New - 3 week old) #Nvidia model of which people talk well, if very good by WOOFER.
The consensus panel, that runs local LLMs until they all agree is very low scoring only 18 WOOFERS.
The 4 models that are currently used in consensus conclave are LFM2, Gemma3 (2B) Llama32 and Quen25.
All under 3 Billion parameters. I might throw another 2GB of Ram onto this VPS just to get the (slightly) larger locals going. -
My harness orchestrator V3 has some surprising and encouraging results;
#Claude model (Most expensive, pay for) has significantly dropped;
Over 80% of mission compute is now carried by free #bigpickle and #llama !!!
This is the #Ai future I want to live in, picking up the free compute off the street. #vibecode
-
Used OpenCode to tidy up my tags in my #Obsidian vault. A painful manual, tedious job that used to take me an hour executed in seconds. 🥲
My vault is pretty old, so it evolved with my systems, so as a result I have to do a lot of tidying up and organizing and I can take days to do this. 😭
PS: #BigPickle is amazing!
-
Good morning!
I have been happily using headscale (self hosted implementation tailscale's management service) for some time. An update was over-due but I was struggling to get things working in a satisfactory way. And I didn't like that my set-up wasn't portable (a lot of manual set-up is required).
This led me to plan out what my ideal headscale project would look like:
* Everything runs in docker
* Containers for headscale, headscale-ui, and caddy
* Re-implement user names in the MagicDNS (eg "mobile" is a user, resulting in "iain-t480s.mobile.example.com") that was removed in 0.23.0+ versions of headscale
* Backup and restore procedures
* Single file configurationOver a few cocktails while on holiday, I instructed OpenCode, with the Big Pickle model, to develop this project.
The result is https://github.com/bigcalm/headscale-compose
A simple project I can clone to any public facing server and hit the ground running with minimal config set-up.
A few notes:
* I have 20 years experience of linux server management, programming, devops, project management, and quality assurance.
* While OpenCode created all of the file contents, I provided the requirements, code review, testing, feedback, and more testing.
* This has been an iterative process to create a fully working project that I am happy with. Not a single prompt and ship whatever gets created.
* I am not a tech bro.
* Agentic development is a useful tool. But only a tool that I have been able to use due to my prior knowledge.The project isn't perfect, but it suits my needs. Maybe it can help somebody else too :)
-
#opencode
Looks like #BigPickle is severely buggy...
(it is probably #GLM46 ) -
OpenCode Zen's "big pickle" free model is quite decent for simple and small coding tasks and bug fixes to be honest. I believe it's GLM 4.x on the backend, but I may be wrong though.
-
