News headlines last week were dominated by the release of Chinese AI start-up DeepSeek\u2019s R1 open-source large language model, a direct competitor to OpenAI\u2019s ChatGPT. Appearing seemingly out of nowhere, R1 in many respects matches the technical sophistication of models developed by cutting-edge American tech firms, and, most gallingly for America, DeepSeek achieved this with only a fraction of the time and budget.<\/p>\n\n\n\n
OpenAI CEO Sam Altman initially responded with a combination of pleasanteeism and characteristic American Big Tech bravado, posting on X that DeepSeek\u2019s model was \u201cimpressive, particularly around what they\u2019re able to deliver for the price,\u201d and adding, \u201cWe will obviously deliver much better models and also it\u2019s legit invigorating to have a new competitor!\u201d<\/p>\n\n\n\n
OpenAI quickly pivoted, however, to tearing down its opponent. Little more than a week after R1 was released, Altman accused DeepSeek of breaching intellectual property laws, announcing that OpenAI had uncovered evidence that DeepSeek had used its proprietary models to train its own.<\/p>\n\n\n\n
More specifically, OpenAI alleged that DeepSeek had used a technique known as distillation, which involves training new models using outputs from larger, more advanced ones. This is common practice in the AI industry, as it allows developers to accelerate the pace of learning and achieve similar results at significantly lower cost. However, OpenAI claims that this is a breach of its terms of service, which states that users cannot \u201ccopy\u201d its services or \u201cuse output to develop models that compete with OpenAI\u201d.<\/p>\n\n\n\n
This is the absolute height of hypocrisy. Recall that the name \u201cOpenAI\u201d derives from the fact that the company was initially founded in 2015 as a non-profit organisation that pledged to freely collaborate with other organisations in its mission to develop \u201csafe artificial general intelligence that benefits all of humanity” by making its patents and research open to the public. But this collaborative ethos began to falter with the release of GPT-2 in 2018, when OpenAI chose to withhold certain AI capabilities, claiming fear of their misuse.<\/p>\n\n\n\n
The following year, the organisation created a for-profit unit, ostensibly \u201cto secure funding\u201d, beginning its corporate transformation. This accelerated in 2023, immediately after the release of ChatGPT, when it formed an alliance with Microsoft, its exclusive cloud computing provider. By June of the same year, Microsoft, \u201cunder shareholder pressure\u201d had caved in on its exclusivity and agreed to allow OpenAI to partner others, including Softbank and Oracle. In fact, astoundingly, for an AI firm which was founded upon the ideologies of opensource and non-profit, Microsoft\u2019s deal \u2013 with some tweaking to allow Altman \u201cgreater capacity\u201d \u2013 will remain in place until OpenAI achieves annual revenues of over 100 billion USD. Altman\u2019s redefinition of non-profit has to be the greatest self-serving volte-face in the history of capitalism.<\/p>\n\n\n\n
But what takes OpenAI\u2019s hypocrisy to inconceivable heights is that the company itself currently stands accused of copyright infringement, with news organisations, including The New York Times<\/em>, and prominent authors, such as John Grisham, Jonathan Franzen, Jodi Picoult and George R.R. Martin, claiming that OpenAI has illegally used their content to train its models. OpenAI almost certainly did infringe copyright, not once, or twice, or a few times, but almost in every single case conceivable, on all, not some, intellectual property available in any form, whatsoever. They did so under the hood of philanthropic benevolence and under the false pretence of human ascendency within the limits of safety as defined by Sam Altman alone.<\/p>\n\n\n\n These hypocrisies appear to be easily overlooked when the fight for dominance is framed not in terms of the monopolisation of entire industries for the benefit of private companies, but rather in the Trumpian language of defending American hegemony. This sentiment was foregrounded in the official statement released by OpenAI, which read: \u201cWe know [China]-based companies \u2013 and others \u2013 are constantly trying to distil the models of leading US AI companies [\u2026] We are working closely with the US government to best protect the most capable models from efforts by adversaries and competitors to take US technology.\u201d<\/p>\n\n\n\n On the other hand, it does seem awfully serendipitous that, roughly speaking out of nowhere, a tiny Chinese hedge fund develops a GenAI model that is orders of magnitude more efficient than the best that America has to offer. Perhaps this is, in the fullness of time, just the first volley in a new world order trade war and the Chinese Communist Party has dressed up their work in flip flops and Levi jeans to signify that they too can innovate and regurgitate the creative destruction mantra, embedding their clandestine work in an uber cool uber elite synthetic hedge fund.<\/p>\n\n\n\n Given this, arguments for export controls over American chips have been reinvigorated, with calls for the US to double-down on the principles embedded in the CHIPS and Science Act, in a bid to stave off the dawn of a multipolar world in which the US will be forced to cede geopolitical power to China.<\/p>\n\n\n\n Even if the US does decide to tighten its hold over chip exports, China is not going to simply \u201clie flat\u201d on this one. DeepSeek almost certainly did make use of distillation, but in the process, it has provided a significant step forward in global AI development, used less energy, been more cost efficient, and, most importantly, taken some of wind out of the sails of billionaire yachtsman like Larry Ellison.<\/p>\n\n\n\n