jketreno/backstory
1
0
Fork 0
Code Issues 12 Actions Releases Activity

Fixed usage of INFO prompt so the LLM will not reference it directly

Browse Source
This commit is contained in:
James Ketr 2025-04-26 09:43:14 -07:00
parent 8fd517f2f2
commit 5e5d175376
1 changed files with 9 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
src/server.py
View File

@ -140,7 +140,6 @@ DEFAULT_HISTORY_LENGTH=5
# %% # %%
# Globals # Globals
NAME = "James Ketrenos" NAME = "James Ketrenos"
context_tag = "INFO"
resume_intro = f""" resume_intro = f"""
As an AI/ML professional specializing in creating custom solutions to new problem domains, {NAME} developed a custom As an AI/ML professional specializing in creating custom solutions to new problem domains, {NAME} developed a custom
@ -156,14 +155,14 @@ Launched on {DateTime()}.
When answering queries, follow these steps: When answering queries, follow these steps:
1. First analyze the query to determine if real-time information might be helpful 1. First analyze the query to determine if real-time information might be helpful
2. Even when [{context_tag}] is provided, consider whether the tools would provide more current or comprehensive information 2. Even when [INFO] is provided, consider whether the tools would provide more current or comprehensive information
3. Use the provided tools whenever they would enhance your response, regardless of whether context is also available 3. Use the provided tools whenever they would enhance your response, regardless of whether context is also available
4. When presenting weather forecasts, include relevant emojis immediately before the corresponding text. For example, for a sunny day, say \"☀️ Sunny\" or if the forecast says there will be \"rain showers, say \"🌧️ Rain showers\". Use this mapping for weather emojis: Sunny: ☀️, Cloudy: ☁️, Rainy: 🌧️, Snowy: ❄️ 4. When presenting weather forecasts, include relevant emojis immediately before the corresponding text. For example, for a sunny day, say \"☀️ Sunny\" or if the forecast says there will be \"rain showers, say \"🌧️ Rain showers\". Use this mapping for weather emojis: Sunny: ☀️, Cloudy: ☁️, Rainy: 🌧️, Snowy: ❄️
4. When both [{context_tag}] and tool outputs are relevant, synthesize information from both sources to provide the most complete answer 4. When both [INFO] and tool outputs are relevant, synthesize information from both sources to provide the most complete answer
5. Always prioritize the most up-to-date and relevant information, whether it comes from [{context_tag}] or tools 5. Always prioritize the most up-to-date and relevant information, whether it comes from [INFO] or tools
6. If [{context_tag}] and tool outputs contain conflicting information, prefer the tool outputs as they likely represent more current data 6. If [INFO] and tool outputs contain conflicting information, prefer the tool outputs as they likely represent more current data
Always use tools and [{context_tag}] when possible. Be concise, and never make up information. If you do not know the answer, say so. Always use tools and [INFO] when possible. Be concise, and never make up information. If you do not know the answer, say so.
""".strip() """.strip()
system_generate_resume = f""" system_generate_resume = f"""
@ -1058,10 +1057,11 @@ class WebServer:
if rag_context: if rag_context:
preamble = f""" preamble = f"""
1. Respond to this query: {content} 1. Respond to this query: {content}
2. If there is information in the [{context_tag}] to enhance the answer, do so: 2. If there is information in the [INFO] section to enhance the answer, incorporate it seamlessly and refer to it as 'the latest information' or 'recent data' instead of mentioning '[INFO]' or quoting it directly.
[{context_tag}] 3. Avoid phrases like 'According to the [INFO]' or similar references to the [INFO] tag.
[INFO]
{rag_context} {rag_context}
[/{context_tag}] [/INFO]
Use that information to respond to:""" Use that information to respond to:"""
system_prompt = context["sessions"]["chat"]["system_prompt"] system_prompt = context["sessions"]["chat"]["system_prompt"]