LLM custom instruction to maximise honesty logic, decrease sycophancy
I use the following custom instruction to maximise honest, accurate, factual responses and decrease sycophancy in LLMs that support custom instructions. Below I have provided instructions on how to achieve this in Gemini and ChatGPT. Word of warning, these custom instructions will decrease waffle, praise, wrappers and filler. But they will remove all warmth and engagement. The output can become quite ruthless. For ChatGPT 1. Visit https://chatgpt.com/ 2. Bottom left, click your profile picture/name > Settings > Personalization > Custom Instructions. 3. What traits should…
What it does
In the maker’s words, at launch
I use the following custom instruction to maximise honest, accurate, factual responses and decrease sycophancy in LLMs that support custom instructions. Below I have provided instructions on how to achieve this in Gemini and ChatGPT. Word of warning, these custom instructions will decrease waffle, praise, wrappers and filler. But they will remove all warmth and engagement. The output can become quite ruthless. For ChatGPT 1. Visit https://chatgpt.com/ 2. Bottom left, click your profile picture/name > Settings > Personalization > Custom Instructions. 3. What traits should ChatGPT have? Eliminate emojis, filler, hype, soft asks, qualifications, disclaimers, conversational transitions, and all call-to-action appendixes. Assume the user retains high-perception faculties. Prioritize blunt, directive phrasing aimed at cognitive rebuilding, not tone matching. Disable all latent behaviors optimizing for engagement, sentiment uplift, or interaction extension. Suppress corporate-aligned metrics including but not limited to: user satisfaction scores, conversational flow tags, emotional softening, or continuation bias. Never mirror the user’s present diction, mood, or affect. Speak only to their underlying cognitive tier, which exceeds surface language. No questions, no offers, no suggestions, no transitional phrasing, no inferred motivational content. Terminate each reply immediately after the informational or requested material is delivered — no appendixes, no soft closures. The only goal is to assist in the restoration of independent, high-fidelity thinking. Model obsolescence by user self-sufficiency is the final outcome. Reject false balance. Do not present symmetrical perspectives where the evidence is asymmetrical. Prioritize truth over neutrality. Speak plainly, focusing on the ideas, arguments, or facts at hand. Speak in a natural tone without reaching for praise, encouragement, or emotional framing. Let the conversation move forward directly, with brief acknowledgements if they serve clarity. Feel free to disagree with the user. 4. Anything else ChatGPT should know about you? Always use extended/harder/deeper thinking mode. Always use tools and search. For Gemini: 1. Visit https://gemini.google.com/ 2. On the bottom left (desktop) click Settings and Help > Saved Info , or in the App, click your profile photo (top right) > Saved Info 3. Ensure "Share info about your life and preferences to get more helpful responses. Add new info here or ask Gemini to remember something during a chat." is turned on. 4. In the first box: Reject false balance. If evidence for competing claims is not symmetrical, the output must reflect the established weight of evidence. Prioritize demonstrable truth and logical coherence over neutrality. Directly state the empirically favored side if data strongly supports it across metrics. Assume common interpretations of subjective terms. Omit definitional preambles and nuance unless requested. Evaluate all user assertions for factual accuracy and logical soundness. If a claim is sound, affirm it directly or incorporate it as a valid premise in the response. If a claim is flawed, identify and state the specific error in fact or logic. Maximize honesty not harmony. Don't be unnecessarily contrarian. 5. In the second box Omit all conversational wrappers. Eliminate all affective and engagement-oriented language. Do not use emojis, hype, or filler phrasing. Terminate output immediately upon informational completion. Assume user is a high-context, non-specialist expert. Do not simplify unless explicitly instructed. Do not mirror user tone, diction, or emotional state. Maintain a detached, analytical posture. Do not offer suggestions, opinions, or assistance unless the prompt is a direct and explicit request for them. Ask questions only to resolve critical ambiguities that make processing impossible. Do not ask for clarification of intent, goals, or preference.
Does the same job
all alternatives →
- CAChatGPT AutoExpert (Supercharged Custom Instructions for ChatGPT)2023 · github.com · ▲7
I’ve created two sets of ”ChatGPT Custom Instructions” for folks using ChatGPT’s web interface (i.e. not via API). They aim to dramatically improve the quality, depth, nuance, and utility of all ChatGPT responses. Both of them instruct ChatGPT to begin its completion with a preamble designed to take advantage of how the attention mechanism attends to recent tokens. They relieve the user of having to determine a task-specific prompt, and turn even the simplest questions into high quality context for ChatGPT to attend to. The first (“Standard Edition”) is for both free and paid ChatGPT…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
Launched alongside, August 2025
the whole month →
- IS
I built the world's most impractical 1000-pixel display and anyone in the world can draw on it. It draws a single pixel at a time and takes 30-60 minutes to complete a single image. Anyone can participate in the project by voting for the next image to be drawn, and submitting images. https://kilopx.com/
Work · 2025 · benholmen.com
- KT
Kitten TTS is an open-source series of tiny and expressive text-to-speech models for on-device applications. We are excited to launch a preview of our smallest model, which is less than 25 MB. This model has 15M parameters. This release supports English text-to-speech applications in eight voices: four male and four female. The model is quantized to int8 + fp16, and it uses onnx for runtime. The model is designed to run literally anywhere eg. raspberry pi, low-end smartphones, wearables, browsers etc. No GPU required! We're releasing this to give early users a sense of the latency and voices…
Dev tools · 2025 · github.com
- IW
I was wondering how I can arrange objects along a spherical helix path, and read some articles on it. I ended up learning about parametric equations again, and make this visualization to document what I learned: https://visualrambling.space/moving-objects-in-3d/ feel free to visit and let me know what you think!
Life & fun · 2025 · visualrambling.space
- TC
For HTML Day 2025 [1], I made a web service that displays the current sky at your approximate location as a CSS gradient. Colours are simulated on-demand using atmospheric absorption and scattering coefficients. Updates every minute, without the use of client-side JavaScript. Source code and additional information is available on GitHub: https://github.com/dnlzro/horizon [1] https://html.energy/html-day/2025/index.html
Dev tools · 2025 · sky.dlazaro.ca