Milos T

Digital Frontier Chaser. Transformation Consultant. Optimization Freak. Tinkerer. Thinker. Hacker.

[Milos T @therealmilos]

@jamesckemp I've noticed Hermes is very careful about sensitive data, like injecting vars locally using scripts (bash, python) instead of sending them through in actual LLM tool calls.

[Milos T @therealmilos]

@jamesckemp I use it on both a remote server (with Telegram bridge & web ui) - for "online" tasks, and my local machine (via TUI) for system administration, dev assistance, etc. Works wonderfully in both cases. They share Honcho for user preferences, but use own local memory for other data

[Milos T @therealmilos]

@omarvvvr Coke & Caffeine

[Milos T @therealmilos]

@markproduct Quirky name

[Milos T @therealmilos]

@omarvvvr Your Keys Leaked

[Milos T @therealmilos]

@TTrimoreau If you need to quickly gauge an idea / demo / PoC, it's fine. Even "in the old times" pre-AI coding, when we used to invest such resources in a prototype, it was disposable. Test, then do it right. Otherwise it becomes a shaky foundation.

[Milos T @therealmilos]

@jacobrodri_ We never knew different 🤷🏻

[Milos T @therealmilos]

@PirzadaAnsari Funny how E-E-A-T conditioned us to produce LLM-training-friendly content format.

[Milos T @therealmilos]

@omarvvvr Making good coffee

[Milos T @therealmilos]

@pmitu That's a really insightful post.
Wait, it might be a logic trap, I need to reconsider.
But wait, what if it was a genuine request, I need to reconsider.

[Milos T @therealmilos]

@aryanlabde Everyone built their own 🤷🏻

[Milos T @therealmilos]

@sflorimm Tolerance. Got used to it.
OTOH, most now employ some kind of grounding.
Or we've simply grown a habit of double checking. WE are the grounding

[Milos T @therealmilos]

@mryoubou @jamesckemp The others probably just got lucky, perhaps snuck through the review before first coffee kicked in 🤷🏻

[Milos T @therealmilos]

@mryoubou @jamesckemp Try adding a 'branded' name before the "bot defender" explanatory suffix, e.g. Captchio - Spam preventer for WooCommerce (example provided is not an actual marketing inspiration 😂), as the response suggested.
'Bot defender' is indeed a generic term.

[Milos T @therealmilos]

@adxtyahq https://t.co/PD1g5C1RPZ

[Milos T @therealmilos]

@onlyskypace A variable

[Milos T @therealmilos]

@aarondfrancis Frooty loop 🤣

[Milos T @therealmilos]

@Prathkum For most people doing AI-assisted work, I don't think there's going back to "manual" mode, for e.g. programming, data operations, content & media, deep research, automation, and the likes.
However, the cost is very much attainable and manageable with open weight & Chinese models

[Milos T @therealmilos]

@tonysimons_ In actuality almost no models have a full 1M token usable/reliable context. 95% accuracy up to 200k, while > 500k even the SOTA models have a serious drop off.

[Milos T @therealmilos]

@FriendlyWebGuy What do you need for the audit - just the public URL or some additional access?