N
I know for most people here, LLM is on the nose as is vibe coding.
But a few weekends of vibe coding has gotten me to a very fun place.
I have weekly meal prep and planning and shopping down to under 10 minutes comparing prices from multiple stores nutritionally balanced and varied recipes.
In about two months of testing, we’ve already saved a couple of hundred dollars.
We also have almost 0 food wasted.
All operated through a fun text UI which feels like an oldschool BBS
@mytwobits01 so if no onions, then return home empty handed. Got it!
@gptreb but was it something you can actually use, like $42 of pumpernickel rye?
@grammargirl it’s good enough for Merriam to write a note about it and the reference links from Wikipedia are fun.
I say this as a born again disjuncter. I was one of those insufferable purists but as I’ve grown, I’ve started to enjoy the flourish and clarity that bastardisation of “official” language can bring. Now I bastardise with gleeful abandon.
Ok.. Having filed the provisional patent, I have genuinely tried to REDUCE the 'signal' from my LLM vs. Human provenance discriminator. I even tried rebuilding ALL of the data collection tool-chain and using only the tiniest of lightweight linear classifiers. No matter what I do the signal persists - this is from a set of data using a different scraper, from different sites, with a different set of LLM generators, with NO prompts (just responding to the sample text).. But the signal remains...
#research #patents #ai #llm #human
@newsgroup plus it seems to be generalised across language models. In this test the simple linear classifier had been trained on a TINY set on ChatGPT5.4 responses, the data had never encountered ANYTHING by Gemini yet here’s a separation! SO COOL!!
RE: https://infosec.exchange/@nikolaihampton/116348450784652940
Some very interesting data in further experiments on my text provenance detector!
Check out these PCA plots!
#governance #risk #compliance #grc #authenticity
@newsgroup well this is where it gets cool. Check this experiment result. The quick discriminator mock up, tuned for ChatGPT-5.4 creating REAL separation across multiple Stack Exchange Q&A pairs from coffee, English, engineering and philosophy (with the discriminator being presented only the answer text with no hints).
Plus this is not standard LLM vs. LLM, theres No prompt required!
Wired an alert tone!
#guilfoyle #
For the record. Human text tends to score > 99%; grammarly and AI edited 90-98%
Aaaaargh. You are all killing the server !!
Ok. Limited sneak peak (this will be taken down shortly).. very early PoC of my new LLM vs. Human research.
A novel and efficient technique for auditing LLM content. Very early stage but the signal is present!!
#llm #ai #grc #governance #machinelearning
[edited to remove link - I’m sure it’s in the history but I’ve taken it down for now. Im glad so many of you enjoyed.]
