Thank you for your comment and resources.
I really appreciate your help.
In the new patch I did my best to serve the sentence English expected sentence very clearly - and have a very low metric for making up new ideas, keeping within the context provided very strictly, and continually asking for clarification / tests.
I dont suggest that this is a replacement for reasoning, but it serves as a handy tool integrated with hooman reasoning to make it work - starting from a ‘I think this part suggests x, what do you think’ viewpoint and expanding from there, and having a preconception to be critical about the response, is what makes it useful, me thinks.
May this thread serve as a warning for people wanting to try it out, of the known issues with the tech. Because I am now aware, I can be careful (me thinks). I like where it is at now, and I take upon any burden this causes me in the future.
//////////////
JP: せっかく二人のためにオーダーしたケーキなのに、なんで一人でたべちゃったの？
EN: Why did you eat it alone when I ordered a cake for the two of us?
AI: Let’s discuss this sentence: 「せっかく二人のためにオーダーしたケーキなのに、なんで一人で たべちゃったの？」. What would you like to explore first?
Commands: /anki | /skip | /image
You: Dang this is a hard sentence - sekkaku ? na no empahsising and ni indicating purpose
AI: The Japanese sentence is in a casual register.
Here is the mapping of the Japanese sentence to the provided English sentence:
- 「せっかく」 → No direct English substring; relates to overall meaning. — Emphasizes the effort or special occasion, implied in the context of ordering a cake.
- 「二人のために」 → “for the two of us” — Indicates the purpose of ordering the cake.
- 「オーダーした」 → “ordered” — Describes the action of ordering the cake.
- 「ケーキ」 → “a cake” — Refers to the cake that was ordered.
- 「なのに」 → “when” — Indicates a contrast or unexpected situation, connecting the two parts of the sentence.
- 「なんで」 → “Why” — Asks for the reason or explanation.
- 「一人で」 → “alone” — Describes the action of eating by oneself.
- 「たべちゃったの」 → “did you eat it” — Refers to the action of eating the cake.
Does this mapping match your intention?
You: other example wih sekkaku ?
AI: Certainly! Here’s another example using 「せっかく」:
JAPANESE SENTENCE: せっかく作った料理を誰も食べなかった。
ENGLISH SENTENCE: No one ate the meal I went to the trouble of making.
Mapping:
- 「せっかく」 → “went to the trouble of” — Emphasizes the effort put into making the meal.
- 「作った」 → “making” — Describes the action of preparing the meal.
- 「料理」 → “the meal” — Refers to the food that was prepared.
- 「を」 → No direct English substring; relates to overall meaning. — Marks the direct object in Japanese.
- 「誰も」 → “No one” — Indicates that nobody ate the meal.
- 「食べなかった」 → “ate” — Describes the action of eating, negated in Japanese to indicate it didn’t happen.
Does this mapping match your intention?
You: yes. What about this tabechatta form ?
AI: The form 「たべちゃった」 is a casual contraction of 「たべてしまった」, which indicates the completion of an action, often with a sense of regret or unintended consequence.
In the provided English sentence, “did you eat it” corresponds to this form, capturing the action of eating the cake.
Does this explanation match your intention?
You: woah. cool
//////
Here it can be seen that because it works within the context provided, very strictly - breaking down the component parts first to the best of its ability without bias helps to avoid illusions because I prompted the agent to very specifically use the english sentence paired up with the jp sentence, creating a type of bond which keeps the answer relevant, and contextually meaningful.
Example #2 - using it as a kind of internal monalogue to encode knowledge rather than get automatic breakdowns and working with the AI ‘i.e. I think x’ to help make compariosons and build knowledge
AI: Certainly! Here’s a bit more detail on the mapping for 「二ゲームとも」:
- 「二ゲームとも」:
- 「二」 means “two.”
- 「ゲーム」 means “games.”
- 「とも」 is a particle that means “both” or “together,” indicating that the action applies to all items mentioned (in this case, both games).
So, 「二ゲームとも」 collectively translates to “in both games,” as it refers to the action happening in both of the games mentioned.
Does this mapping match your intention?
You: I feel like isshou is used moreso for togetherness and tomo is used more for ‘both’
AI: Yes, you’re correct in noting the nuances. In this context, 「とも」 is used to indicate “both” in the sense of including all items mentioned, which are the two games. It emphasizes that the action applies to both games collectively.
Does this mapping match your intention?
You: yes
AI: Great! If you have any more questions or need further clarification, feel free to ask.
You: /anki
///////////////////
Example 3 - AI strictly follows context window for responses, even when I suggest something else - it speaks less with the user, and more with the sentence… kind of like it is speaking at itself in some internal monologue, and we are the observer. I think this helps to avoid hallucination → it is confident in a good way, because the context has been provided.
You: ayo
AI: REGISTER/FORMALITY: Neutral
- 「来週」 → “next week” — This directly corresponds to the time frame mentioned in the English sentence.
- 「初級」 → “beginner level class” — “初級” means “beginner level,” which matches the English phrase.
- 「から」 → “from” — Indicates the starting point or origin, matching “from” in the English sentence.
- 「中級」 → “intermediate level class” — “中級” means “intermediate level,” which matches the English phrase.
- 「に」 → No direct English substring; relates to overall meaning. — Indicates direction or target, implied in the transition.
- 「レベルが上がります」 → “I’m moving to the intermediate level class” — This phrase means “level goes up,” which corresponds to the action of moving to a higher level in the English sentence.
Does this mapping match your intention?
You: I swear kara meant ‘until’ doe ?
AI: In this context, 「から」 means “from,” indicating the starting point of a transition. It is used to show the origin or starting point, as in moving from one level to another.
Does this mapping match your intention?
You:
////
I 1000000% agree that the AI is flawed, but my argument is that if we can carefully navigate this situation by helping it out, and enforcing strict rules that aknowledge the flaws, then we can create a useful tool.
