Day 3
InsightForge now checks not just whether a cited URL is reachable, but whether the source appears to support the claim the model made.
much better audit signal than “link exists.”
Still heuristic. Lot's to improve on
#buildinpublic#opensource#AI#LLM#toronto
Daily progress on InsightForge:
Extract cited URLs from AI output
Verify source reachability
Block private/local hosts by default
Fail runs when citations are not verifiable
From “looks sourced” to “actually checked.”
#buildinpublic#AI#LLM#OpenSource#Python
PS: Early days—regex risks, no SDK hooks yet. Next: Real fact-checks, browser ext. Building in public to iterate fast. What's your biggest AI trust pain? DM or reply.
AI models spit out answers like oracles, but we're blind to the "why"—hallucinations, biases, black-box BS. I got fed up debugging opaque outputs, so I built InsightForge: an open-source CLI to trace reasoning, flag risks, and audit any LLM interaction. Model-agnostic
repo : https://t.co/DJgdBWlQYQ
Stars if AI opacity pisses you off too. Let's make AI trustworthy. Open to contibutions 🔍#AI#OpenSource#BuildingInPublic
#BuildingInPublic MVP just shipped: Wrap your AI calls, capture prompts/responses, detect biases/hallucinations with policies, diff traces, export HTML audits. Heuristic for now, but it catches lies like this: [insert split-screen demo GIF of bad vs. good prompt output]
$BTC
As you know 2 weeks ago I suggested 29650 for swing long entries before I went vacation
Price came close to it but I didn’t get most of my bids filled
Today’s long entry was also good but I also closed it
Overall I’m way too less long that’s why it likely pumps
History doesn't predict the future. But this is the history. 4 year cycles so far. #Bitcoin
(Source: Data is public. Table made by @gautamchhugani, I think)