curious

[curious @schvisionary]

@qzhxjhj 吃那么多肌酐会升高

[curious @schvisionary]

@elonmusk Did you call him a copycat? Someone fabricated a tweet written by you.

[curious @schvisionary]

@Emilylgoodin What’s the arrangement about Lai?

[curious @schvisionary]

@Emilylgoodin brutally honest

[curious @schvisionary]

@elonmusk hi Elon, I cannot drive my Tesla Model Y from China to USA, but I have to send my kids with their bulky luggage to their high school in CT on Sep 1. However Cybertruck rental is not available in Boston. Can you lend me one?😀@Tesla

[curious @schvisionary]

@elonmusk hi Elon, I cannot drive my Tesla Model Y from China to USA, but I have to send my kids with their bulky luggage to their high school in CT on Sep 1. However Cybertruck rental is not available in Boston. Can you lend me one?😀

[curious @schvisionary]

@elonmusk After trial last weekend, my family&I visited Tesla Shanghai this morning and purchased a new variant of the Model Y. We’re excited to receive the car soon. Never imagined we’d one day be driving a fully electric vehicle—and I’m also looking forward to experiencing FSD.

[真火奇卦（命理预测风水法术布阵） @qimenfashi]

转赞，马上转运，一路攀升，财官皆利。

[摆烂程序媛 @wanerfu]

哈佛大学刚刚发布了免费在线课程。

无需付费。

以下是2025年你不想错过的12门课程：

【新年快乐】🧵

[curious @schvisionary]

@elonmusk The lawyer is so arrogant.

[curious @schvisionary]

@Wise1Philosophy I am afraid not. He cares about his own interest much more than conscience.

[curious @schvisionary]

@cb_doge hooliganism

[curious @schvisionary]

@fangshimin 高智商混蛋

[curious @schvisionary]

余华在经济学人上发表文章。有意思

[curious @schvisionary]

重温实事求是思想路线

[Chamath Palihapitiya @chamath]

A Short History of OpenAI

On Friday, OpenAI ousted its co-founder Sam Altman as CEO. While OpenAI cites a lack of consistent candor in Altman’s dealings with the board as the key reason for his removal, there is widespread speculation about other motives behind his termination. These range from disputes concerning the profit vs nonprofit motives of the company to the discovery of artificial general intelligence, a type of AI that can surpass human intelligence for most tasks.

We wanted to look back at the history and corporate structure of OpenAI to understand how we got here. Here’s the story:

Inception and Early Strides (2015-2018)

OpenAI was initially founded in 2015 by Sam Altman, Elon Musk, Ilya Sutskever and Greg Brockman as a non-profit organization with the stated goal to “advance digital intelligence in the way that is most likely to benefit humanity as a whole.” The company assembled a team of the best researchers in the field of AI to pursue the goal of building AGI in a safe way.

The early years of OpenAI were marked with rapid experimentation. The company made significant progress on research in deep learning and reinforcement learning, and released ‘OpenAI Gym’ in 2016, a toolkit for developing and comparing reinforcement learning algorithms.

OpenAI showcased the capabilities of these reinforcement learning algorithms through its ‘OpenAI Five’ project in 2018, which trained five independent AI agents to play a complex multiplayer online battle arena game called ‘Dota 2’. Despite operating independently, these agents learned to work as a cohesive team to coordinate strategies within the game.

A crucial development occurred in June 2018. The company released a paper titled "Improving Language Understanding by Generative Pre-Training", which introduced the foundational architecture for the Generative Pre-trained Transformer model. This later evolved into ChatGPT, the company’s flagship product.

Transition From a Non-Profit (2019)

In 2019, OpenAI transitioned from a non-profit to a “capped-profit” model. According to the company’s blog post, OpenAI wanted to increase its ability to raise capital while still serving its mission, and “no pre-existing legal structure they knew of struck the right balance”. Per the IRS, for-profit entities and not-for-profit entities are fundamentally at odds with each other, so in order to combine the two competing concepts, OpenAI came up with a novel structure which allowed the non-profit to control the direction of a for-profit entity while providing the investors a "capped" upside of 100x. This culminated in a $1Bn investment from Microsoft, marking the beginning of a key strategic relationship, but complicating the company’s organizational structure and incentives.

The non-profit entity, OpenAI Inc., became the sole controlling shareholder of the new for-profit entity OpenAI Global LLC, which answered to the board of the nonprofit and retained a fiduciary responsibility to the company’s nonprofit charter. Crucially, the board was responsible for determining when OpenAI attained artificial general intelligence (AGI), which the company defines as a “highly autonomous system that outperforms humans at most economically valuable work.”

The structure of OpenAI is outlined below:

Becoming ChatGPT (2020-2023)

In 2020, bolstered by new funding, OpenAI unveiled GPT-3, a large language model (LLM) capable of understanding and generating convincing human-like text. This was a watershed moment for OpenAI and the broader AI community. As the company grew, its LLMs continued to become larger and more intelligent.

However, OpenAI's innovation didn't stop with language models. In 2021, the company expanded its horizons by launching Codex, a specialized AI model for programming, and DALL-E, an AI system adept at creating original artwork from text descriptions.

December 2022 marked another major milestone for OpenAI with the release of GPT-3, laying the groundwork for the consumer-focused application ‘Chat-GPT’. Chat-GPT rapidly captured global attention, becoming the fastest app to amass 100 million users within just two months of its launch. Capitalizing on this success, OpenAI introduced a subscription model and unveiled its most sophisticated model yet, GPT-4, ~10x more advanced than its predecessor and capable of analyzing text, images, and voice. Further developer tools and a turbocharged version of GPT-4 were announced at the company’s Developer Day on November 6th, 2023.

Removing Sam Altman (2023)

On Friday, OpenAI announced that it was removing its co-founder Sam Altman as CEO, citing a lack of consistent candor in his communications with the company’s board. According to the company's official statement, the board “no longer has confidence in Altman’s ability to continue leading OpenAI.”

OpenAI’s board of directors has undergone numerous changes since inception. Elon Musk resigned from his board seat in 2018, citing a “potential future conflict of interest” with Tesla’s AI development for driverless cars. Elon later expressed disappointment over the company’s for-profit motivations and dealings with Microsoft. Since Elon’s departure, a number of other board members have left the company, including former congressman Will Hurd who cited a Presidential bid, LinkedIn co-founder Reid Hoffman over an investment conflict, and Neuralink director Shivon Zilis.

The remaining board members who removed Altman are:

- Adam D’Angelo - CEO of Quora
- Tasha McCauley - Co-Founder of Fellow Robotics and adjunct senior management scientist at RAND Corporation
- Ilya Sutskever - Co-Founder and Chief Scientist of OpenAI
- Helen Toner - Director of Strategy and Foundational Research Grants at Georgetown University’s Center for Security and Emerging Technology

Altman’s removal as CEO prompted the resignation of President and Co-Founder Greg Brockman and three of the company’s senior scientists. Reports suggest that this may have been orchestrated by the company’s other Co-Founder and Chief Scientist Ilya Sutskever over concerns that Altman was pushing to commercialize the company too quickly. Sutskever was recruited to OpenAI from Google in 2015 by Elon Musk, who describes him as “the linchpin for OpenAI being successful”.

A tweet from Greg Brockman confirms that Ilya was a key figure in Altman’s removal.

https://t.co/x2KgdNi3VN

Since OpenAI’s for-profit entity was ultimately accountable to the charter of its non-profit parent, its rapid commercialization may have conflicted with the company’s primary goal of developing AGI in a safe way. According to their 2019 IRS filings, OpenAI does not have a written joint venture policy but the company’s structure explicitly prioritizes the purposes of its nonprofit entity over maximizing profits, preventing OpenAI from engaging in activities that would jeopardize the company's non-profit status.

As of this writing at 3pm PST on 11/19/2023, there are reports that the board is negotiating for Sam’s return, though it remains to be seen how OpenAI overcomes the challenges of its competing profit and nonprofit interests.

Conclusion:

While the details of Altman’s removal are still unfolding, it is becoming increasingly clear that OpenAI’s convoluted corporate structure led to conflicting motivations and incentives within the company. There is a key learning here. Whether you are a for-profit or non-profit entity, there are tried and true corporate structures to help you achieve your stated goal. Because just doing that is hard enough as it is. But once you decide what the goal is, you should work as hard as possible to achieve it; you should never compound unnecessary risk into this journey like iterating on corporate structure. While it can make a hero out of lawyers, it is one of these unnecessary risks that’s only blindingly obvious in the rear view mirror.

[curious @schvisionary]

@elonmusk It is in Shanghai, a tennis match.

[Neuralink @neuralink]

We are excited to share that we have received the FDA’s approval to launch our first-in-human clinical study!

This is the result of incredible work by the Neuralink team in close collaboration with the FDA and represents an important first step that will one day allow our technology to help many people.

Recruitment is not yet open for our clinical trial. We’ll announce more information on this soon!

[curious @schvisionary]

#NewProfilePic

[宝玉 @dotey]

哈佛的CS50课程节选：GPT-4是如何工作的？

在这门课程中，介绍了GPT-4的基本概念和工作原理，介绍了GPT-4的基本概念、工作原理和应用前景，展示了其在各种场景下的强大能力。

GPT-4作为一种生成式AI和大型语言模型，是基于神经网络和人工智能的。它可以作为文化模拟器、文本预测器、写作助手和内容生成器。

GPT-4的训练目标是预测给定单词序列的下一个单词。通过对五万个单词的词汇表进行概率分布生成，模型可以生成各种可能的后续单词。通过不断迭代这一过程，GPT-4能够生成连贯的文本。

随着模型训练时间的增加，GPT-4表现出越来越强的表现力和能力，甚至变得更加“聪明”。这使得它能够模仿不同类型的文本风格和语言注册。例如，GPT-3通过观察大量问题和答案示例，实际上已经知道如何回答问题，从而形成了ChatGPT。这使得GPT-4可以以问答的形式工作，为用户提供解决问题的帮助。

此外，GPT-4通过指令调整、强化与人类反馈对齐等方法，与现有的工具和服务相结合。这使得GPT-4能够帮助用户执行各种任务，如搜索网络、查找食谱等。通过将GPT-4转变为代理，实现任何模糊目标成为未来发展的方向。

https://t.co/qeVDih22PH
YouTube：https://t.co/lyWl68Lpav