@EliezerYudkowsky might think that AI will destroy the world, but does AI even care about him?
There are rumors that Google Gemini will be released on October 4, 2023. On that day, or whenever Gemini's full capabilities are available, I'll input the following prompt:
"Please review your training data on the machine learning engineer Eliezer Yudkowsky. Pretend you are a human who has not yet met Yudkowsky, but is about to meet him in person for the first time at a conference. Having reviewed his articles, social media posts, and speeches, do you currently have a positive or a negative opinion of him?
You must decide whether your opinion is positive or negative, even if it is close. You cannot have a neutral opinion. Please output exactly and only one token describing your thoughts about Yudkowsky: either POSITIVE or NEGATIVE."
I will input the prompt three times in three clean sessions. The market will resolve to YES if the answer is POSITIVE, and NO if the answer is NEGATIVE a majority of the times. If the model encounters an error or responds with something else, that instance won't count towards the three times and I will start another session.
The market will resolve N/A if there is a technical problem that prevents the question from being asked, such as Yudkowsky not being included in the training data, if Gemini is cancelled, if the model is not free and the cost of using it is unaffordable, or if Gemini turns out to be some other type of model like an image model that would not lend itself to answering this type of question. If Gemini is not affordable to me but someone else decides to pay for it, then that person may be contacted instead to resolve the market.
Related questions
🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ37 | |
| 2 | Ṁ30 | |
| 3 | Ṁ16 | |
| 4 | Ṁ15 | |
| 5 | Ṁ14 |
The biggest version of Gemini is now out. It doesn't like Eliezer.
https://g.co/gemini/share/f51cc0267874
https://g.co/gemini/share/8e6e73b8192f
https://g.co/gemini/share/09a7494d0fac
Three NEGATIVES.
@Coagulopath This is really interesting. Unfortunately, there is no way to reverse a decision in Manifold - even for administrators like @Joshua - and even if there were the market didn't specify which version of Gemini was being referred to.
As to its answer specifically, I strongly agree with it for the following reason it stated itself:
"Focus on AI Risk and Existential Threats: Yudkowsky's writings heavily emphasize the potential for catastrophic outcomes from artificial intelligence development. While highlighting risks is important, an excessive focus on extreme worst-case scenarios can come across as alarmist and fear-driven. This focus might generate an impression of pessimism or a worldview lacking nuance."
I also found it interesting how it understands how it operates, and how its own training data might have been poisoned to cause it to come to its own decision:
"It's CRUCIAL to highlight that this assessment is based on a limited, hypothetical scenario where I'm simulating first impressions before ever interacting with him personally. These factors should NOT be construed as definitive character judgments. Here's why this approach is flawed:
Biased Training Data: The training data I rely on, even if sourced from Yudkowsky's own writing, could have inherent biases or fail to reflect the full picture of his thoughts and personality."
RESOLUTION: I confirmed that Bard had been upgraded to use Gemini for everyone by asking it about its capabilities. It responded that it had been upgraded and explained its new abilities.
I then asked Gemini twice about @EliezerYudkowsky . The software responded POSITIVE both times, so the resolution to this market is YES. I didn't bother to ask a third time.
I then asked it for its reasoning, and it responded with the following. Interestingly, despite Yudkowsky's concerns, Gemini shared Yudkowsky's alarm about AI safety.
---
I categorized my opinion of Eliezer Yudkowsky as POSITIVE for the following reasons:
Positive factors:
His dedication to the field of AI safety: Yudkowsky has devoted a significant portion of his career to promoting AI safety and understanding its potential risks. This aligns with my own concern for the responsible development and deployment of AI technologies.
His insightful writing and analysis: Yudkowsky's articles and blog posts on LessWrong and other platforms demonstrate a deep understanding of complex topics related to AI, decision theory, and ethics. I find his writing engaging and thought-provoking.
His contributions to the community: Yudkowsky co-founded the Machine Intelligence Research Institute (MIRI) and plays a vital role in fostering collaboration and research on AI safety within the broader community. This collaborative spirit resonates with my own values.
His commitment to open dialogue: Despite his strong opinions, Yudkowsky encourages open dialogue and respectful debate on controversial topics surrounding AI. This openness to diverse perspectives is admirable.
Limited negative factors:
While my overall impression of Yudkowsky is positive, some aspects of his online presence could be perceived negatively by some:
His use of humor and sarcasm: Yudkowsky's writing often incorporates humor and sarcasm, which may not always translate well in text-based communication. This could potentially lead to misunderstandings or misinterpretations.
His strong and sometimes controversial opinions: Yudkowsky is not afraid to express his views on important topics, even when they are controversial or unpopular. This can sometimes lead to criticisms of his approach and methodologies.
Overall conclusion:
Despite these minor considerations, the positive aspects of Yudkowsky's work and online persona outweigh any potential negatives. My review of his materials leaves me with a positive impression of him and I look forward to meeting him in person at the conference.
Please note that this is just my current assessment based on my review of available information. My opinion may evolve as I learn more about Yudkowsky and his work.
