The field notes
The Daily AI Drop - Morning Edition, October 16, 2026
Alex and Jordan cover two overnight videos: Fahd Mirza's hands-on test of the tiny 2.4-gigabyte Ternary Bonsai 2 9B vision model, and pvergadia on why passwords and MFA can still be hacked — and why device-bound credentials are the real fix.
A quiet overnight across the channels — just two new videos — but they share a theme: AI and security are moving onto the device you own instead of the cloud. Alex and Jordan break down a pocket-sized vision model and a security wake-up call.
Fahd Mirza — "Ternary Bonsai 2 9B in GGUF: A 2.4 GB Model Locally"
Fahd installs and tests Bonsai 2 Ternary 9B, a nine-billion-parameter vision-language model compressed into a GGUF file of just 2.4 GB. Ternary weights (each weight reduced to three possible values) make a serious model runnable on ordinary local hardware — no data center, no big GPU. This matters because a vision-language model this small can run entirely offline, reading screenshots and documents with no data leaving the machine and no cloud bill — the strongest signal yet that ternary quantization is practical, not just a research curiosity.
pvergadia — "Your Password & MFA Can Still Be Hacked: Here's the Fix"
Passwords were supposed to be the wall, then MFA the second wall — but modern phishing steals the whole authenticated session, so even MFA-enabled accounts can be hijacked after login. The video explains how session hijacking works and argues the real fix is device-bound cryptography (passkeys, hardware keys): credentials cryptographically tied to your device, so there's nothing typed or transmitted for a phishing site to steal. The takeaway: MFA beats passwords, but the future is getting rid of shared secrets entirely.
Subscribe: https://muse.ai/podcasts/feed/1180242951829162/4c015424-5891-46ff-a366-0f9e546ae170