Sam Altman: OpenAI's Stance on Open Source is Historically Incorrect

OpenAI Addresses Concerns and Future Plans in Reddit AMA

Following a series of product announcements, key figures from OpenAI – including CEO Sam Altman – participated in an extensive question-and-answer session on Reddit on Friday.

Currently, OpenAI is navigating a complex landscape. The company is facing scrutiny regarding its competitive standing against Chinese AI firms like DeepSeek, with OpenAI alleging potential intellectual property theft. Simultaneously, OpenAI is working to strengthen its ties with Washington D.C. and advance a substantial data center project, alongside preparations for a potentially record-breaking funding round.

Acknowledging the Competition and Open Source Debate

Altman conceded that DeepSeek has narrowed OpenAI’s advantage in the field of artificial intelligence. He also expressed a belief that OpenAI may have erred in its approach to open-sourcing technologies, stating, “I personally think we need to figure out a different open source strategy.”

While OpenAI has previously released open-source models, its general preference has been for a proprietary development model. However, Kevin Weil, OpenAI’s chief product officer, indicated a willingness to consider open-sourcing older, non-state-of-the-art models, stating, “We’ll definitely think about doing more of this.”

Transparency and Reasoning Models

The emergence of DeepSeek has also prompted OpenAI to re-evaluate its approach to transparency in its reasoning models. DeepSeek’s R1 model displays its complete chain of thought, while OpenAI’s models currently conceal their reasoning process to deter competitors from utilizing the information for training their own models.

Weil announced that OpenAI is actively working on increasing the visibility of its models’ thought processes, saying that showcasing this will be available “very very soon.” A balance must be struck, however, as full transparency could lead to competitive distillation.

ChatGPT Pricing and Compute Power

Altman and Weil addressed concerns about potential price increases for ChatGPT, the platform used to launch many of OpenAI’s models. Altman stated his desire to make ChatGPT “cheaper” if possible.

Previously, Altman revealed that OpenAI incurs a loss on its most expensive ChatGPT plan, ChatGPT Pro, which is priced at $200 monthly.

Weil emphasized the correlation between increased compute power and improved model performance. This is a key driver behind projects like Stargate, OpenAI’s ambitious new data center initiative, as well as the growing demand from its expanding user base.

Recursive Self-Improvement and Potential Risks

Regarding the possibility of recursive self-improvement – where AI systems enhance their own intelligence without human intervention – Altman suggested that a “fast takeoff” scenario is now more likely than he previously thought.

However, it’s important to note Altman’s history of making optimistic predictions, including a recent adjustment to OpenAI’s timeline for achieving Artificial General Intelligence (AGI).

Collaboration with the U.S. Government and Nuclear Defense

A Reddit user inquired about the potential for OpenAI’s models to be used in the development of destructive weapons, specifically nuclear arms. This question arose following OpenAI’s recent partnership with the U.S. government to provide its models to U.S. National Laboratories for nuclear defense research.

Weil expressed confidence in the government scientists involved, stating they possess both AI expertise and a rigorous, evidence-based approach to their work, minimizing the risk of reckless application of model outputs.

Future Model Releases and DALL-E 3 Successor

The OpenAI team also responded to technical questions regarding upcoming model releases. Altman indicated that the next reasoning model, o3, is expected to launch “more than a few weeks, less than a few months” from now.

The timeline for GPT-5, OpenAI’s next flagship model, remains uncertain, as Altman stated, “don’t have a timeline yet.” Furthermore, Weil confirmed that a successor to DALL-E 3, the company’s image-generating model, is under development.

DALL-E 3, released approximately two years ago, is showing its age as image-generation technology has rapidly advanced. Weil assured users that the next iteration will be “worth the wait.”