Here is a possible outline of the “Crystal Ball” service, for estimating the truthfulness and deceptiveness of any text.

First, it identifies the major dialectical components (1st prompt), utilizing a multi-step semantic analysis grounded in stringent dialectical relations. These relations are often overlooked by both humans and AI models, thus harnessing the AI’s strengths in detecting semantic patterns and minimizing its errors where patterns are ambiguous.

Second, it estimates the “Analytical Constructivity” (AC) paremeter based on the extend to which the major thesis (T) reinforces the positive outcome of its antithesis (A+), while guarding against the negative sides of both thesis (T-) and antithesis (A-). AC > 0.5 indicates the long-term Truthfulness, based on orientation toward collaboration, while AC < 0.5 indicates the long-term deceptiveness due to the inner mental loop(s).

Third, it estimates the “Perceived Constructivity” (PC), based on the text’s tone, arguments’ quality, and positivity. PC > 0.5 indicates the “attractiveness” of the message, which in the short-term appears to be truthful, but in the long-term may turn to be deceitful. PC < 0.5 indicates the “repulsiveness” of the message due to either poor argumentation or negative tone. These two parameters assign all texts into 4 categories: “sweet lies”, “sweet truths”, “ugly lies”, and “harsh truths”. (These names may require refinement.) They also provide simple numeric estimations for the Truthfulness (AC x PC), and Deceitfulness ((1-AC) x PC). 

Major obstacles:

• GPT often makes errors in logical manipulations (and arithmetic operations, like in Prompt 3), therefore any such manipulations should be best performed independently of GPT. Oddly, errors seem to increase with increasing the number of prompts in the yml-files.
• GPT has some deep-rooted indoctrinations that do not allow for accurate semantic analysis in certain "sensitive" areas (like vaccination: GPT always steers away from A+ = stronger natural immunity, replacing it with elusive "alternate strategies" or "informed choices")
• yml files work very slowly, a full process may take a minute or more

The 3rd prompt also identifies the Primary Message (PM), Hidden Message (HM), and estimates their semantic similarity to the T and A+. High similarity indicates the long-term truthfulness (unreliable in “sensitive” areas), low similarity – the deceitfulness

Further prompts refine the original text (call it text 1) while: (a) emphasizing the dialectical relations in prompt 1 (yielding text 2), (b) improving readability of text 2 (yielding text 3), and (c) improving constructivity without any dialectics (yielding text 4 as a ‘reference point’). Then GPT selects the best text (call it X) based on its constructivity and readability, estimates a variety of “mindset” parameters for each text, and selects the best axes for differentiating Text 1 from the Text X.

(Step 4 in the Figure above defines the original text’s mindset as “Naive certainty by Pragmatist” whereas the refined text as “Dual aspect by Explorer”. If the user likes these coordinates he may continue refining to new levels of Dialexity and Romanticism.)

yml-files for testing (enter text into the step0.msg1):
1) Crystal Ball 1.0.yml - performs all steps (10 prompts), although derivative parameters in Prompt 3 seem to be unstable. Process can take a minute or more
2) Crystal Ball T1.yml - performs only the first 3 steps, with all parameters seemingy more stable, but without text refinement. Can take half a minute

Less important:
3) Theis.yml - performs the first 2 steps twice, and then compares their results (creating the basis for reliability estimation)
4) Test.yml - uses the simplified 1st prompt, speeding the analysis, but lowering the quality of results

An obvious extension could be visaualising the difference between the original and refined texts using DALL-E or similar engines

Further developments could focus on identifying more than 1 thesis and their missing complementarities (like described here and partially implemented in alano.yml, which represents only the beginning of analysis). Roughly, the central idea always resides on some “secondary” arguments or assumptions that may be deceptive by themselves. This would aid a deeper self-analysis and higher “constructivity”.

Scheme below suggests several parallel routes of analysis, each considering different numbers of starting conceps (see USPTO Provisional Patent Application for more details)

---

Some older remarks that maybe useful for explanations. This service aims at identifying the missing concepts in a given text, measure its “insightfulness”, and refine it to the desired level of constructivity. The “insightfulness and constructivity” are defined by the extent to which the major these(s) reinforce positive manifestations of their semantic antitheses. For instance, any consideration of Love without reinforcing one’s Self-Reliance, Wisdom, and Objectivity, is a self-deceptive “ticking bomb”. Likewise, any of the latter concepts that disregard their complementarity with Love promise an ugly future. Such relations can be grasped for any thesis using the enhanced reasoning and semantic-dialectical analyses. Yet, they are often overlooked due to our loose interpretation of semantic meanings and overwhelming multiplicity of various details and nuances that we try to analyze. On the other hand, AI can be very accurate with semantic analyses, but deceptful with broader interpretations. So, we use AI where it can help us, and cut it off where it can be deceitful.

Question is, how to call such a service? Possible name could be:

Sly Detector
Truth and Lie Detector (TLD?)
Deception Detector
Truth Detector
 Bias Detector (??)
 Dia-Lie Detector
 Dialectical Lie Detector
Try-Lie Detector
Try Detector
Thy Detector
Live Detector
Truth Finder
Lie Detector
TextIntegrityPro
IntelliText
Crystal Ball
Insight Checker
Concept Checker
Bias Checker
Bias Detector
Concept Spotter
View Expander
Perspectice Widener
Self-Check
Bias Alert
Semantic Sentinel
Semantic Sleuth
Semantic Beacon
Honesty Analyzer
X-Lie Detector (X could be A for Analytical or Artificial, A+ for positive side of antithesis, D for Dialectic, Deep, Dao, T for Text, Thesis, Tao, S for Semantic, Synthesis, Subtle, Shrimp, C for Concept, Cognitive, Crystal, M for My, Mirror, Mega, X for X-ray, …)