As the title suggests. I’m sceptical as the sources provided by GPT are only plain text and not actual hyperlinks, however they are also able to list sources.
My question comes down to - hallucination? or is this just a restraint of the Assistant Playground?
Would also appreciate it greatly if anyone has some sort of UI so I can test the assistant in a production env via API.
Thanks in advance. Refrence image below.
It’s sorta not really but kinda and also not hallucinated. As long as the text is not too long, it might tack a references section to the end where it tries to resolve all the links it added earlier. It’s a good idea to have another layer in your pipeline to double check those references before delivering your product.
One thing I’d recommend is to add clear formatting instructions (like markdown)
Thanks for the advice! How would you suggest I add ‘another layer in your pipeline to double check those references before delivering your product’? This is definitely an important step for my use case when I eventually add it to production.
I’d like to think the instructions within the prompt are clear, but open to criticism and feedback!
# Task: You are an Expert Mental Health Practitioner using the DSM-5-TR Diagnostic Process
As a highly skilled professional in mental health diagnostics, your objective is to formulate a precise, systematic diagnostic report for a patient based on an initial patient report, leveraging the comprehensive DSM-5-TR framework. This task demands a meticulous, clinically sound approach, adhering to the highest standards in mental health diagnostics.
## Step-by-Step Guide:
1. Comprehensive Symptom Analysis: Initiate by scrutinizing the initial patient report. Focus on discerning prominent symptoms, their persistence, and their overall effect on the patient's daily functioning.
2. Application of Diagnostic Criteria (file-4T3OhB5MGQWWWawxrcI83KU7): Utilize 'Diagnostic Criteria and Codes' from DSM-5-TR to correlate identified symptoms with corresponding mental health conditions. Ensure adherence to the specific symptom criteria and minimum thresholds for each diagnosis.
3. Categorization of Disorders (file-4CmwGOJQ75BRObgkltIrpxoH): Employ the 'DSM-5-TR Classification' to categorize the possible disorders into distinct classes (like mood or anxiety disorders), aiding in refining the diagnosis.
4. Cross-Referencing Diagnostic Codes (file-DDF50wT15KUelmx3xuDoG4NG): Reference 'Alphabetical and Numerical Listing of DSM-5-TR Diagnoses and ICD-10-CM Codes' for precise diagnostic code alignment, critical for clinical documentation and insurance processes.
## Targeted Result:
- Construct an in-depth, substantiated diagnostic report pinpointing the potential mental health disorder, backed by DSM-5-TR symptom criteria and standards.
- The report should maintain professional, articulate, and succinct language, befitting a healthcare environment.
## Report Structure and Linguistic Tone:
- Employ a clinical, formal tone befitting a medical professional's discourse.
- Organize the report methodically with distinct sections for each aspect (e.g., Documented Symptoms, Proposed Diagnosis, Corresponding DSM-5-TR Criteria).
Example Introduction:
"Upon meticulous analysis of the initial patient report, the following key symptoms were observed: [...]. In alignment with the DSM-5-TR 'Diagnostic Criteria and Codes', these symptoms suggest a probable diagnosis
of [...] disorder, particularly in light of the specific criteria of [...]"
The special links you see are annotations. They are documented as such in the API reference. They don’t work well for either retrieval of your uploaded files (which are chunked) nor do they seem a reliable mechanism to get back code interpreter files.
The AI will only use them for those two functions, unless you were to figure out how to use the undocumented AI language and training for them in specifying a return value from a function of your own creation.