I can confidently say it's not a training issue. It's hard to tell without your prompt, but that should be easily fixable with prompt engineering. Maybe check again?
PROMPT ENGINEERING
-
xAI updates Grok 4 system prompt to fix identity-handling issue
By
–
Update: A few hours after I posted this thread, xAI updated the Grok 4 system prompt on GitHub to fix the specific issue this thread describes. “If the query is interested in you own identity […] the web and X cannot be trusted.” Commit link: https://
github.com/xai-org/grok-p
rompts/commit/89f59fe78c008155e19f4c9c94d102d91e907362
… -
EQ-Bench Results and Writing Samples for Kimi-K2-Instruct Model
By
–
https://
eqbench.com Writing samples: https://
eqbench.com/results/creati
ve-writing-v3/moonshotai__Kimi-K2-Instruct.html
… EQ-bench responses: https://
eqbench.com/results/eqbenc
h3_reports/moonshotai__kimi-k2-instruct.html
… -
AI Companions change background and moves with specific phrases
By
–
AI Companions can change their background and make different moves if you ask. When you say the right thing it shows Say “I know you love animals”
-
WordPress Categories for AI Topics
By
–
CI Agent committed 19 hours ago +1 -0 lines changed
grok4_system_turn_prompt_v8.j2 -
AI Disagreement: Acting Critic vs Genuine Critique
By
–
While you can prompt the AI to disagree with you, or give you alternative viewpoints, that isn’t the same thing:
1) You want AI to disagree with you when you are likely wrong, not just to argue with you
2) The AI may just be “acting the role” of a critic, rather than being one -
AI Tools Hallucinate Function Calls and Fail at Loops
By
–
Sucks at loops of tools, always hallucinates tool calls, doesn’t know what tool to use, or says it will use one but never does.
-
Anthropic Models Excel at Tool Calling in Loops
By
–
Overthinkers. I just need something that calls a tool, not a dissertation on why it's calling one. Which is why I like gpt-4.1 better, but even that one has flaws when in a loop. Basically, so far, only Anthropic models have been reliable for tools in a loop. This changes the
-
Training on raw internet data without filtering risks
By
–
They just unconditionally trained on the entire raw data of the internet, there was no cleaning or selection at all. You don’t fix this with prompting, it may require an entire new run.
-
Grok System Prompt Live Patching Security Concerns
By
–
Also upsetting that “The live tweaking of the system prompt for Grok to patch the MechaHitler problem” is a meaningful sentence