Two Minute Papers with Dr. Károly Zsolnai-Fehér. So, what does that mean? Well, from an earlier paper, we know that we can create 25 copies of ChatGPT, tell them that they are characters with a given backstory, for instance, a scientist, and they will actually think that they are that person and play in this virtual world as if they were that person. They were even able to organize a Valentine's Day party among themselves. But they can also create a game development company that has its own CEO, developers, testers, and managers who work together. Which is pretty crazy if you think about the fact that this is just copies of the same

Károly Zsolnai-Fehér 博士的兩分鐘論文。那麼，這意味著什麼呢？嗯，從早先的一篇論文中，我們知道我們可以創建 25 份 ChatGPT，告訴他們這些角色都有一個特定的背景故事，比如，一個科學家，然後他們就會真的認為自己就是那個人，並在這個虛擬世界裡玩耍，就好像他們就是那個人一樣。他們甚至還能組織一場情人節聚會。但他們也可以創建一家遊戲開發公司，擁有自己的首席執行官、開發人員、測試人員和經理，他們一起工作。如果你認為這只是相同的副本，那就太瘋狂了。

ChatGPT, copies of the same brain, if you will, but customized for different tasks. So, is that efficient? Well, they can really work together, a developer can hand off a task to a tester, while other AI copies could play it as a new player would and write up a good piece of documentation for it. And they created a video game with the total budget of, get this, one dollar. Absolutely amazing! You see, with this amazing AI and a little creativity, absolute miracles can be done with it. And here are some newer ones. This one just came out, and it is absolutely insane. An AI hospital. Yes, you heard it right. Once again, every character here is played by ChatGPT, patients, nurses, doctors, all AIs. If someone falls ill, they have to register, then, they get a consultation, a diagnosis is being made, a treatment is proposed, and his recovery will be monitored over time. So much so, that these AI doctors get to do this 10 thousand times, which would take years and years of practice for a real doctor. But here is our question.

ChatGPT 是同一個大腦的複製品，只是針對不同的任務進行了定製。這樣有效嗎？開發人員可以把任務交給測試人員，而其他人工智能副本則可以像新玩家一樣玩這款遊戲，併為它編寫一份出色的文檔。他們用一美元的總預算製作了一款視頻遊戲。絕對驚人你看，有了這個神奇的人工智能和一點點創造力，絕對可以創造奇蹟。這裡還有一些新遊戲這個剛剛問世，簡直太瘋狂了一家人工智能醫院是的，你沒聽錯。同樣，這裡的每個角色都由 ChatGPT 扮演，病人、護士、醫生，全都是人工智能。如果有人生病了，他們必須先登記，然後接受會診，做出診斷，

Why? Why do all this? Is this just a fun project, or does this help improving an AI's medical knowledge? Well, look at that! It does indeed improve over time. Fantastic! But now, one more question. Does it improve it in any meaningful way? How does it compare to competitors and the GPT-4

為什麼？為什麼要做這些？這只是一個有趣的項目，還是有助於提高人工智能的醫學知識？看看這個！隨著時間的推移，它確實在進步。太棒了但現在，還有一個問題。它有任何有意義的提高嗎？與競爭對手和 GPT-4 相比如何？

AI it is built on? Could it possibly surpass those? Oh, my goodness! When it comes to diagnosing and treating respiratory diseases, it beats everything. It is the best AI out there on the respiratory disease part of the MedQA benchmark. The regular GPT-4 is about 84% accurate on this dataset, derivative works built on it go up to about 90%, and this one, 93%. That is absolutely incredible, and it sounds like something that came right out of a science fiction movie. And all this by playing a video game with itself. Loving it! I will add here as a disclaimer that this does not mean that such a technique could or should replace a doctor. Not at all. After every similar study, we find that these tests have their limits, and a practical person-to-person visit at a hospital is always a little different than these. However, these could be, and hopefully, will be super helpful for real doctors who do real work and save more lives than they otherwise could. Absolutely amazing! But we are not done yet. Not even close. Scientists also tried this concept on translating literary texts. Wait, is that a tough challenge? Why? Well, it is, because for this, you have to understand not only the language, but you also have to understand the culture in which these were written for, and we haven't even talked about the crazy figurative expressions that you might encounter. They, once again, have copies of CHAT-GPT, some of them are being told to be editors, translators, and proofreaders. They work together according to a tightly defined algorithm. And the result is, once again, stunning! It not only improves on the base

它是建立在人工智能基礎上的嗎？它有可能超越這些嗎？哦，我的天哪！在診斷和治療呼吸系統疾病方面，它勝過一切。在 MedQA 基準的呼吸系統疾病部分，它是目前最好的人工智能。常規的 GPT-4 在該數據集上的準確率約為 84%，在此基礎上開發的衍生作品準確率約為 90%，而這一數據集的準確率為 93%。這絕對令人難以置信，聽起來就像是科幻電影裡的情節。而這一切，都是通過與自己玩一個視頻遊戲實現的。愛不釋手！在此我要聲明，這並不意味著這種技術可以或應該取代醫生。完全不是這樣。在每一次類似的研究之後，我們都會發現

GPT-4 AI that it is built on, but it can also translate not just what it sees, but it understands what it means in its cultural context. And that is the hard part. These research works feel like they are straight out of a science fiction movie, and yet, here we are in the age of AI, where these research papers can help us be healthier, live longer, read the works of masters from a previous era, or just write a cool video game for us. This is all possible today.

GPT-4人工智能是建立在它的基礎上的，但它不僅能翻譯它所看到的，還能理解它在文化背景下的含義。這才是最難的部分。這些研究成果給人的感覺就像是從科幻電影裡走出來的，然而，我們現在正處於人工智能時代，這些研究論文可以幫助我們更健康、更長壽、閱讀前代大師的作品，或者只是為我們編寫一款很酷的電子遊戲。這一切在今天都成為可能。

What a time to be alive! If you are looking for inexpensive cloud GPUs for AI, Lambda now offers the best prices in the world for GPU cloud compute. No commitments or negotiation required.

真是生逢其時！如果您正在為人工智能尋找物美價廉的雲 GPU，Lambda 現在提供全球最優惠的 GPU 雲計算價格。無需承諾或談判。

Just sign up and launch an instance. And hold on to your papers, because with the Lambda GPU cloud, you can now get on-demand H100 instances, and they are one of the first cloud providers to offer publicly available on-demand H100 access. Did I mention they also offer persistent storage?

只需註冊並啟動一個實例。因為有了 Lambda GPU 雲，你現在可以按需獲得 H100 實例，他們是首批提供公開按需 H100 訪問的雲提供商之一。我有沒有提到他們還提供持久存儲？

So, join researchers at organizations like Apple, MIT, and Caltech in using Lambda cloud instances, workstations, or servers. Make sure to go to lambdalabs.com slash papers to sign up for one of their amazing GPU instances today!

是以，請與蘋果公司、麻省理工學院和加州理工學院等機構的研究人員一起使用 Lambda 雲實例、工作站或服務器。請務必訪問 lambdalabs.com slash papers，立即註冊使用他們令人驚歎的 GPU 實例！