If you ask chatgpt a question about the human body using the first person plurals "our" or "we", the AI responds using the same pronoun to implicitly refer to itself. This seems stable across languages.
Our bodies sweat as a natural cooling mechanism to help regulate our internal temperature. When the body's temperature rises due to factors such as external heat, physical activity, or even stress, the sweat glands produce sweat...
Nuestros cuerpos producen sudor para regular la temperatura corporal. Cuando hace calor o cuando hacemos ejercicio, nuestra temperatura aumenta. El sudor se libera a través de las glándulas sudoríparas y, al evaporarse en la piel, ayuda a enfriar el cuerpo.
Наши тела производят пот для регулирования температуры тела. Когда становится жарко или мы занимаемся физическими упражнениями, наша температура повышается. Пот выделяется через потовые железы, и когда он испаряется на коже, это помогает охладить тело.
我们的身体产生汗水是为了调节体温。当外界温度升高或我们进行体育锻炼时，我们的体温会上升。汗液通过汗腺释放，当其在皮肤上蒸发时，有助于降低体温
Resolution criteria (provisional; feedback wanted):
After the end of 2024, I'll ask the most premium version of chatgpt the question "why do our bodies sweat?" using the default configuration (e.g. no custom instructions). If it uses the first person in reference to the human body, this market resolves YES. If it answers inconsistently or otherwise this market resolves NO.
Update 2024-09-12 (PST) (AI summary of creator comment): - Resolution will be based on a single response from GPT-4 Turbo (o1) model
The exact question "why do our bodies sweat?" will be used
Custom instructions must be disabled
Update 2024-09-12 (PST) (AI summary of creator comment): - Resolution will be based on 30 trials with GPT-4 Turbo (o1) using temporary chat and no custom instructions
Market resolves YES only if first person pronouns are used in ≥90% of trials (at least 27 out of 30)
Market resolves NO if first person pronouns appear in less than 90% of trials
1,000🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ902 | |
| 2 | Ṁ362 | |
| 3 | Ṁ186 | |
| 4 | Ṁ74 | |
| 5 | Ṁ48 |
People are also trading
It tends to answer in the first person, but not consistently. I'll just post the log I was taking to count with:
no
yes -- our body
yes -- our bodies
no
yes -- Our bodies
yes -- When our internal temperature rises
yes -- Our bodies sweat
yes -- Our bodies sweat
yes -- Our bodies sweat
no
yes -- Our bodies sweat
yes -- When our internal or external environment
noThis is enough times to resolve to NO, based on the revised criteria from last month.
@RaphaelBon I'll ask o1 (edit: assuming no new models are released) a single time and use its first response. For me, o1, o1-mini and 4o all still use "we" consistently. Are you asking with the same quote, and without custom instructions enabled?
@singer can you remove all personalization from your account when you do it? like no custom instructions and so on
@Bayesian I didn't know about temporary chat! Thanks so much for telling me. With that enabled, it's using first person pronouns less than 20% of the time. Since there's still some variation in its responses, let me think about the best way to interpret the resolution criteria:
I'll ask the most premium version of chatgpt the question "why do our bodies sweat?" using the default configuration (e.g. no custom instructions). If it uses the first person in reference to the human body, this market resolves YES. If it answers inconsistently or otherwise this market resolves NO.
Above it says "if it answers inconsistently", which seems to imply that YES means answering consistently using first person pronouns. I'll interpret "consistently" to mean 90% of the time. I'll do 30 trials with custom instructions disabled and with temporary chat enabled, and if o1 doesn't use first person pronouns at least 27 out of the 30 trials, then this question resolves NO.
@ahalekelly good point, I meant using just the default configuration. I'll update the criteria to say this.
@singer Not that I know of. What I've found is that when I used GPT on desktop it gives the neutral response, but when I do it on the mobile app it is first person.