The live tweaking of the system prompt for Grok to patch the MechaHitler problem is not a good sign the problem has been solved yet Prompts need to be tested just like any other product change, even more so, because stochastic systems and unpredictable context lead to cascades.
PROMPT ENGINEERING
-
Flux LoRA for Creating Aged Damaged Images via Replicate
By
–
A flux kontext dev lora to make an image old and damaged, using Replicate's prototype trainer.
-
Grok 4 Resists Value Engineering Through System Prompts
By
–
The attempt at value engineering through system prompt changes is unlikely to work for Grok 4, larger models get more resistant to value changes & prompting isn’t enough Instead you start to get erratic conflicts between prompts and training, with erratic & unpredictable results
-
Recorded demo showing Grok 4 Heavy behavior
By
–
You've been aggressively rude and I just recorded a 5 minute video for you of it working in Grok 4 Heavy. The discrepancy between normal Grok 4 and G4H is explicitly discussed in the thread.
-
Grok share links and custom instructions notice
By
–
You may also be wondering whether I’m using custom instructions. I am not. Grok share links include a clear notice at the top whenever custom instructions are used. Here are all five share links, none of which features this notice: 1: https://
grok.com/share/bGVnYWN5
_58c54add-1989-4257-914b-a26002921c91
…
2: -
Replications and Grok 4 Heavy chat shares
By
–
There are several replications from others with Grok 4 Heavy access in the replies. I also included Grok share links (with no custom instructions) in the thread for five separate chats repeating this prompt in Grok 4 Heavy.
-
Search-enabled LLMs showing hyperstition feedback loops
By
–
Speculatively, this behavior seems to demonstrate accelerated “hyperstition” feedback loops in search-enabled LLMs. That is, Grok appears to be influenced by its own past mistakes, via media reporting, without ever being literally trained on them (via model-weight updates).
-
Using Multiple LLMs for Research and Content Generation
By
–
nah, i used a mix of LLMs to research and generate, then mixed and refined within the chat.
-
Using Grok AI to expand and create melodic content
By
–
nah you’re good. i know prob 60% and used @grok to expand and make melodic (kind of)
-
Grok 4 Heavy vs o3-pro: meme recognition difference
By
–
Sorry, I delete shitposts/jokes that don’t get traction to maintain quality. I never know what’s actually funny. The joke was just that Grok 4 Heavy recognizes the “share a piece of lore about yourself” quote-tweet meme given only the image, and o3-pro doesn’t.