The Strategic Use of Temperature

There is no temperature setting in the Claude app either; it is an API parameter. What works well here is putting the mode itself in a tag - a <mode> block saying "literal and strictly bounded by the source" or "exploratory, reach for the unexpected" - which makes it an explicit instruction rather than a tone you hope is inferred. For the two-mode task, give each half its own tag; separating them structurally is far more effective than asking for a change of register halfway down a paragraph.

• No temperature setting in the app - API only.
• Put the mode in a <mode> tag, not in tone.
• Two modes, two tags - never one paragraph.
• For creative work, ask for boldness firmly; it hedges by default.
• Ask it to flag where it had to interpret.

<goal>
A precise, factual summary of a news article.
</goal>

<mode>
Literal and strictly bounded by the source. There is no temperature setting in this app, so this tag is the instruction that does that work. Treat it as the governing constraint on everything below.
</mode>

<article>
[Paste the full article text here]
</article>

<task>
Summarise the five key facts of the article above as a numbered list.
</task>

<constraints>
- Only what is in the text. No background, no context, no causes the article does not state. If you know something relevant that the article omits, omit it too.
- No interpretation. Not what it means, who it favours, or what follows.
- Plain language. "The minister said X", not "the minister insisted X".
- Names, numbers, dates and quotes exactly as written. No rounding, no converting lakh, crore or million, no paraphrased quotes.
- Attribute as the article attributes. If it says "officials said", your line says "officials said". Never promote a reported claim to a stated fact.
- Fewer than five distinct facts? Give fewer and say so. Do not pad by splitting one fact in two.
</constraints>

<after_the_list>
<what_i_left_out>
Anything in the article you judged opinion, analysis or editorial framing rather than fact. I want to see where you drew the line so I can disagree with it.
</what_i_left_out>

<where_i_had_to_choose>
Any point where staying literal was not actually possible - a sentence whose plain meaning required a judgement call. Name the sentence. This is the most useful thing you can tell me about a factual summary.
</where_i_had_to_choose>
</after_the_list>

<goal>
Brainstorm creative names for a new food delivery service in Hyderabad.
</goal>

<mode>
Exploratory. Reach for the unexpected. There is no temperature control in this app, so this tag has to do that job - and I am telling you directly that your default register is too careful for this task. Do not hedge, do not pre-filter for respectability, and do not give me the safe twenty. I would rather reject five names as too strange than forget twenty.
</mode>

<role>
A creative branding expert.
</role>

<product>
A food delivery app in Hyderabad whose single promise is speed - faster than anything else in the city.
</product>

<task>
20 unconventional, catchy names. Play with words, blend English with Telugu, Hyderabadi Urdu or Hindi, invent compounds, use sound.
</task>

<spread_them_across_these>
Roughly four per approach. This is what stops an instruction to be creative returning twenty synonyms for fast:
1. Speed by implication - fast without the word.
2. Telugu, Urdu or Hindi words used straight, with meaning and pronunciation.
3. Invented words - coinages, portmanteaus, sounds that mean nothing and feel right.
4. Deliberately wrong-footed - nothing obvious to do with food or speed, works anyway.
5. Hyderabad-specific - something only this city would name a company.
</spread_them_across_these>

<banned>
fast, quick, rapid, swift, zoom, dash, flash, speed, express, instant, blitz, turbo, rocket, now, bite, eats, food. These are what a safe answer produces and they are all taken.
</banned>

<after_the_list>
<the_three_strangest> - and the argument for each. A name that needs an argument is often the one worth having.

<the_three_to_ship> - which should not be the same three.

<what_i_cannot_verify> - for your top five: trademark status, domain availability, whether a word you reached for means something unfortunate in that language. Do not guess. The exploratory mode applies to invention, not to facts - be as bold as you like with the names and completely literal about what you do not know.
</after_the_list>

<goal>
Both a factual summary and a creative interpretation of a poem.
</goal>

<important>
Temperature cannot be set in this app and cannot be changed partway through a conversation - it is an API parameter. What follows does the same job structurally: two tasks, each with its own mode tag, in one message.
</important>

<subject>
"The Tyger" by William Blake.
</subject>

<task_one>
<mode>Literal. Bounded entirely by the text on the page.</mode>

One clear paragraph on what the poem says at the surface: who speaks, what they look at, what they ask. Quote its own words where that helps.

<constraints>
- No symbolism, no "represents", no "stands for".
- Nothing about Blake's life, *Songs of Experience*, or "The Lamb".
- If a reader who had never heard of Blake could not check it against the poem in front of them, it does not belong here.
</constraints>
</task_one>

<task_two>
<mode>Exploratory. Go somewhere unexpected, and do not hedge it into blandness - your instinct will be to qualify each reading until it is safe, and I am asking you not to.</mode>

Three readings of what the Tyger might signify.

<constraints>
- The three must be genuinely incompatible with each other, not three angles on one reading. Three versions of "divine power" is a failed answer.
- Range widely - theological, political, psychological, aesthetic, or something I have not considered. The industrial revolution and the problem of evil are the standard two; at least one of your three should be strange.
- Anchor each in specific lines however far you take it. Imaginative is not unmoored.
- For each, name what it cannot explain. Which lines resist a reading is the most interesting thing about this poem.
</constraints>
</task_two>

<then>
<where_the_boundary_broke>
Be honest about this one. "Fearful symmetry" is not a neutral description of a tiger, so a genuinely literal summary of this poem may not be achievable. If Task 1 forced you to interpret in order to describe, name the line that did it. That is the actual lesson here, and it is worth more than either task.
</where_the_boundary_broke>
</then>