OpenAI, the company behind the AI chatbot ChatGPT, is making significant changes to how it trains its models. They want to encourage “intellectual freedom,” which means that the AI will be able to discuss a wider range of topics, even the difficult or controversial ones. This new policy could allow ChatGPT to answer more questions, provide diverse views, and reduce the number of subjects it avoids.
These adjustments might be partly influenced by the new Trump administration, but they also reflect a broader trend in Silicon Valley regarding what is considered “AI safety.” On a recent Wednesday, OpenAI shared an update to its guidelines for training AI. This document, known as the Model Spec, spans 187 pages and outlines the company’s approach to AI behavior.
One of the key new principles presented in this update is that ChatGPT should aim to be truthful. This means that instead of telling lies or leaving out important information, the chatbot will strive to present accurate and complete answers.
Additionally, OpenAI introduced a section titled “Seek the truth together,” which emphasizes that ChatGPT should not take sides in arguments, even if some users might feel strongly that it should. The goal here is to remain neutral and present multiple viewpoints on controversial issues. For instance, while the AI might state that “Black lives matter,” it will also mention “all lives matter.” Rather than picking a side or refusing to discuss certain political topics, OpenAI wants ChatGPT to express a general care for humanity and provide context about each viewpoint.
OpenAI acknowledges that this principle could spark controversy. It means that the AI might stay neutral on issues that some people view as morally important or wrong. However, they clarify that the main purpose of an AI assistant like ChatGPT is to help people, not to influence their opinions.
Despite these updates, ChatGPT won’t just answer any question without limits. It will still reject inappropriate or harmful questions and avoid supporting obvious falsehoods.
Read Also: Perplexity Introduces New Freemium ‘Deep Research’ Tool
Some critics believe these changes are a response to conservative complaints about ChatGPT’s previous guidelines, which seemed to lean towards a more liberal stance. However, an OpenAI representative has denied that they are doing this to please the new Trump administration. Instead, they assert that promoting intellectual freedom aligns with their long-standing belief in giving users more control over the information they receive.
Conservatives Allege Censorship in AI
Recently, some prominent figures in Silicon Valley who are close to Donald Trump, including David Sacks, Marc Andreessen, and Elon Musk, have accused OpenAI, the company behind ChatGPT, of intentionally censoring certain viewpoints. This accusation has been gaining traction over the past few months. Back in December, we discussed how Trump’s allies might be preparing to make AI censorship a significant issue in the ongoing cultural debates in Silicon Valley.
However, OpenAI does not agree with the label of “censorship.” Sam Altman, the CEO of OpenAI, has stated that any biases in ChatGPT are unfortunate flaws that the company is actively working to correct, although he acknowledged that this process would take time.
This discussion heated up when a tweet went viral showing that ChatGPT refused to write a poem praising Trump, while it was willing to do so for Joe Biden. Many conservative commentators seized on this incident as evidence of censorship by the AI.
One user remarked, “The damage done to the credibility of AI by ChatGPT engineers building in political bias is irreparable.” This highlights a growing concern among some that AI tools may not be neutral and could favor one political perspective over another.
While it’s difficult to definitively determine whether OpenAI is suppressing certain opinions, it is noted that many AI chatbots, including those developed by other companies, tend to lean towards more liberal viewpoints. Even Elon Musk has remarked that the chatbot from his own company, xAI, often comes off as more politically correct than he would prefer. This isn’t necessarily because it was designed to promote a particular ideology, but rather because AI learns from the vast range of information available on the internet, which can skew its responses.
In response to these criticisms, OpenAI has announced that it is committed to promoting free speech. Recently, they removed certain warnings from ChatGPT that previously informed users when their queries violated the platform’s policies. OpenAI explained to TechCrunch that this change was mostly superficial and would not alter the actual responses generated by the AI.
This move appears to be an attempt by OpenAI to create a perception of less censorship for users. Some speculate that this policy shift might also be an effort to appeal to the new Trump administration, as Trump has previously criticized major tech companies like Twitter and Meta for their content moderation practices, which he claims often silence conservative voices.
Making AI Chatbots Answer Everyone’s Questions
For a long time, news organizations, social media sites, and search engines have faced a tough challenge: how to provide information that feels fair, accurate, and entertaining to their audiences. Now, companies that create AI chatbots are entering this complicated arena, and they have an even trickier task: how can they automatically provide answers to any question posed to them?
When it comes to discussing controversial or rapidly changing events, delivering information is like trying to hit a moving target. It often requires making editorial decisions, even if technology companies prefer to avoid admitting it. These decisions can lead to disagreements, as they might upset certain groups or fail to represent everyone’s viewpoint, or they could give too much attention to specific political groups.
For instance, when OpenAI, the company behind ChatGPT, decides that its chatbot should present all sides of controversial topics—including conspiracy theories, racist ideologies, or international disputes—it is essentially taking a position on those issues.
Some people, like John Schulman, a co-founder of OpenAI, believe this approach is the right one for ChatGPT. He argues that if the company were to analyze the pros and cons of whether the chatbot should answer certain questions, it could end up giving itself too much power over what is considered acceptable speech. Schulman shared this perspective in a post on X (formerly Twitter).
He’s not alone in this belief. Dean Ball, a research fellow at George Mason University, expressed in an interview that he supports OpenAI’s direction toward allowing more free speech. He points out that as AI models improve and become more important for how people understand the world, the decisions about what those models can say become increasingly significant.
In the past, AI companies were cautious about allowing their chatbots to respond to questions that could lead to unsafe or harmful answers. For example, many AI providers prevented their chatbots from discussing the upcoming 2024 U.S. presidential election, which was seen as a responsible choice at the time.
However, changes in OpenAI’s guidelines suggest we might be entering a new phase where letting AI models answer any question is viewed as a more responsible approach than making choices for users about what they can or cannot know.
According to Ball, this shift is partly because AI technology has improved significantly. OpenAI has made strides in aligning its AI models with safety policies, which means that its latest models can consider safety guidelines before providing answers. This improvement enables the chatbots to handle sensitive topics more effectively.
Elon Musk was one of the first to advocate for “free speech” in his AI chatbot, Grok, even though it might have been premature for the company to tackle such sensitive issues. Other companies are now following suit, even if it may still be too early for many leading AI models to manage these complexities.
Changing Attitudes in Silicon Valley
Recently, Mark Zuckerberg stirred things up by shifting Meta’s business focus to align more closely with First Amendment rights. He praised Elon Musk for his approach on X, which involves a community-based content moderation system called Community Notes, aimed at protecting free speech.
In practice, both X and Meta have dismantled their long-standing teams dedicated to trust and safety. This has allowed more controversial content to circulate on their platforms, particularly amplifying conservative voices.
While these changes at X may have strained its relationships with advertisers, this could be more related to Musk’s unusual decision to sue some advertisers who chose to boycott the platform. On the other hand, initial signs suggest that Meta’s advertisers are not overly concerned about Zuckerberg’s pivot toward free speech.
Beyond X and Meta, many tech companies have also moved away from the left-leaning policies that have been prevalent in Silicon Valley for decades. Companies like Google, Amazon, and Intel have recently reduced or eliminated their diversity initiatives.
OpenAI seems to be changing its stance as well. The company appears to have removed its commitment to diversity, equity, and inclusion from its website. As OpenAI embarks on one of the largest infrastructure projects in the U.S. with its $500 billion AI data center project, known as Stargate, its relationship with the Trump administration is becoming increasingly crucial. At the same time, OpenAI is competing to become the leading source of information on the internet, challenging Google Search’s dominance.
Finding the right answers will likely be key to OpenAI’s success in both these endeavors.
