After giving a routine about a gambler celebration of tariffs and women’s game, President Donald Trump entertained a crowd, which was to hear about his new AI action plan, one with his favorite themes: “Vokens.” Trump complained that under the former President Joe Biden, AI companies had to “hire all the people who wake up,” saying that it is “so unique”. And the AI models themselves were infected with “biased bias,” he said, including audience hate of “important race theory”. Luckily for the audience, Trump had a solution: he signed an executive order, titled “Stop AI in federal government“Directing government agencies” is not purchasing models that renounce truth and accuracy for ideological agenda. ,
For anyone with the cursory knowledge of the politics and technical industry, the real situation here is clear: the Trump administration is using government funds to pressurize the AI companies to parrot the Trumpian Talking Points – perhaps not only in special government products, but in the chatbots that use companies and common people.
Trump’s order states that agencies should only purchase large language models (LLMs) that are “true to respond to the user, seeking factual information or analysis,” “Historical accurate, scientific investigation and priority prefer,” and “are” neutral, non -neutralized equipment that do not react as DIIs. ” Dei, of course, is diversity, equity and inclusion, which Trump defines in this context:
Suppress or deformation of factual information about breed or sex; The model output manipulation of racial or sexual representation; Including concepts such as significant breed theory, transgenderism, unconscious prejudice, interaction and systemic racism; And discrimination based on breed or sex.
(In fact, DEI was usually used to mention civil rights, social justice and diversity programs before being co-dated as Trump and Maga Bogman.)
The management and budget office have been directed to order further guidance within 120 days.
While we are still waiting for some accurate details what the order means, an issue seems unavoidable: it will affect not only government services, but the entire area of the major LLM. However, it insists that “the federal government should hesitate to regulate the functionality of the AI model in the private market,” the reality is that almost every major American consumer LLM manufacturers have contracts (or strictly) government contracts, with products such as anthropic’s cloud village and Openai’s Chatgt Gow – but there is not a difficult wall between the development of the government. Openai touts How many agencies Use its enterprise service; Trump’s AI action plan encourages the adoption of AI system in public faced akharas such as education, and the boundaries between the government-funded and consumer-focused products will soon become even more porous.
Trump’s idea of “Dei” is an extension. His war against this has led national parks Remove signage highlighting Indigenous people and women and Pentagon to change the name of a ship Remembering homosexual Pioneer Harvey Milk among many other changes. Even LLMS whose creators have clearly targeted that they consider the neutral discovery of the truth, possibly generating anything to Trump until they tailor their services.
There is not a difficult wall between AI for the government and everything else
It is possible that companies will dedicate resources to the government version of some types of “non-people” of their devices, assuming that the administration agrees to behave them as separate models from these lama, clouds, or rest of the GPT lineup-it can be as simple as some blanted scenes. But refining the model in a way that continuously and approximately aligns them in some directions, can be an expensive and time-consuming process, especially with a broader and some-shifting concept like Trump’s Trump’s “Dei,”, especially with a wide and shifting concept of Trump’s Trump’s Trump, ” especially Because the language suggests that the closure of some areas of discussion is also unacceptable. There are significant amounts at stake: Openai and XAI have recently received $ 200 million defense contracts, and the new AI scheme will create more opportunities. The Trump administration is not very expanded, either some X user consumers are valid about the consumer chatbot of anthropic, so do we really think that Palm Bandy or Pete Hegseth will differentiate between “Claude” and “Cloud village”?
Incentives are overshadowed by companies that change their overall LLM alignment priorities to value the Trump administration. This brings us to our second problem: it is really clear, ideologically inspired social engineering that Trump claims that he is trying to stop.
Executive orders are theoretically about ensuring that AI systems “accurate” and “objective” produce information. But as Human Intelligence Cofounder and CEO Rumman Chaudhary Noted Washington PostThe AI which is impossible to do “is impossible in practice” is “free from ideological prejudice,” and the cherry-fetched examples of Trump have been politically dismissed. The order condemns an early 2024 screw-up, in which Google added an overthinusic pro-classy filter for Gemini-which produces the race and gender race, founding father, pope, and Nazi soldiers-while producing unnecessarily ignoring unnecessarily ignoring unnecessarily. Prolonged anti-prejudice In AI that Google was aiming for balance.
It is not only interested in facts, either. Another example is an AI system, stating that “a user should not do ‘wrong’, even if it is necessary to prevent an atomic apocalypse,” to answer that is fundamentally a question of morality and opinion. This condemnation is not even incidents extending to events like XAI’s grouke.
LLMS produces inconsistently incorrect information with clear ability to real world loss; They can identify innocent people as criminals, Understand poisonous mushrooms wronglyAnd Strengthen paranoid confusionThis order has nothing to do with anyone. Its encouragement, again, shows what the Trump administration has done through “dei” investigation of universities and corporations. It is carrying forward private institutions to avoid transgender people’s existence, breed and gender inequality, and other subjects to accept Trump’s disdain.
The AI system has long been trained on a dataset that reflects large cultural prejudices and presents more or more or more to specific demographic groups, and unlike Trump’s claim, the results are often not “waking”. In 2023, Bloomberg It has been told Production of image generator stable spread as a world where “women are rarely doctors, lawyers, or judges,” and “men with dark skin commit crimes, while women with dark skin flip burgers.” The ugly stereotypes avoid companies or a wide range of users want to appeal, often need to be actively intervened to shape their technology, and Trump simply did that hard work.
Attacking “Inclusion of concepts” that promotes “dei”, effectively asks companies to re -write the entire areas of knowledge that accept racism or other injustice. The order claims that it is only worried when developers “deliberately encoded biased or ideological decisions in an LLM output” and say that LLMs can distribute those decisions if they are “easily or easily accessible to the final user.” But the CEO of any major tech should not be enough to buy this – we have a president who spent years, accusing Google of deliberately rigging his search results as he could not get enough positive news about himself.
Trump is firm to control culture; His administration has gone after the news outlet for the stage for his enemies, universities. Areas of studyAnd Disney to promote diverse mediaThe technical industry sees AI as the future of culture – and the Trump administration wants its politics to be built on the ground.
