The minds behind Oboe are addressing a common concern regarding artificial intelligence: “Will AI diminish our ability to think for ourselves?” This question was posed in a recent advertisement from the company.
Oboe’s founders assert that the answer is a definitive no, and they aim to demonstrate this through their innovative startup.
Launched this month, the AI education platform utilizes artificial intelligence to create “courses” on any subject of interest to users. Oboe comes from the team behind Anchor, a platform for DIY mini-podcasts, which was purchased by Spotify for $150 million. Nir Zicherman, Oboe’s CEO and co-founder, previously led Spotify’s audiobooks division, while fellow co-founder Mike Mignano managed the podcast team at Spotify.
Zicherman expresses a desire to make high-quality learning experiences more accessible and affordable through the use of AI, as he shared with Technology News. This ambition is particularly noteworthy considering the saturated market of AI-driven educational tools, along with the ongoing challenges of AI inaccuracies.
According to Zicherman, people usually rely on platforms like ChatGPT, Google, YouTube, and Wikipedia for information, which often forces them to “piece it together” on a uniform learning path. In contrast, Oboe seeks to consolidate learning into a singular destination, where users can access information more effectively.
The interface of Oboe resembles familiar chatbots, featuring a textbox for users to input their learning queries. Unlike ChatGPT, however, Oboe provides responses in the form of AI-generated “courses” that mimic traditional educational structures, such as comprehensive texts akin to introductory textbook chapters or concise lists of “key takeaways.”
Users can explore various topics with Oboe, including complex subjects like the origins of artificial intelligence or forms of intelligence found in nature. An example highlighted in Oboe’s promotion.
Alternatively, users might opt for more niche subjects, such as the intricacies of concrete manufacturing and its ecological impacts, a topic I chose for my exploration.
This selection produced a “deep dive” article curated for a 10-minute reading time. It opened with an engaging image description: “Imagine the handful of fresh concrete in your palm.” The text was enriched with headers like “environmental footprint” and included tables listing various materials known as “SCM.” A quick search for the definition of “SCM” yielded no results on the page. The article also integrated images from Wikimedia Commons at regular intervals.
Though I gained a basic understanding of concrete manufacturing, the platform did not alleviate my concern about the accuracy of the information presented. Users are entrusted with verifying facts. Oboe does not provide links to original sources for statements, such as the one that “aggregates are granular materials such as sand, gravel, or crushed stone that constitute the majority of concrete’s volume (commonly 60–75%).” A check with the American Cement Association confirmed this figure, but I had to seek it out independently.
Plans are underway to enhance Oboe by adding citations in the near future. “Citations and other methods for accessing additional online resources are aspects we’re actively working on and hope to incorporate into the platform shortly,” Zicherman commented via email to Technology News.
In contrast, other platforms like Google’s Gemini offer citations upon request. When I inquired about concrete and specified my need for citations, I received a link to a 2018 article in Nature Sustainability, which detailed that global concrete production is responsible for nearly 10 percent of worldwide industrial water use.
While citations can create an illusion of reliability, they provide a valuable resource for those seeking accurate information.
Oboe does not develop its own foundational AI models, according to Zicherman, as it sources models from other entities optimized for specific tasks, though he did not disclose their identities. To mitigate the risk of inaccuracies, Oboe employs some language models to verify the outputs generated by other models.
Zicherman explains, “For example, one model may output facts intended for inclusion in a course, while another model trained on a different dataset evaluates this information to identify inaccuracies. This approach aims to reduce the chances of errors or fabrications.”
At this stage, there is no feature for users to flag inaccuracies directly in the results; instead, they must report issues to Oboe’s team, which has a protocol for reviewing and integrating necessary adjustments into its content pipeline.
Additionally, Oboe offers the option of AI-generated podcasts. For instance, a pair of lively, AI-created “hosts” explained concrete’s ecological impact in a style reminiscent of NPR’s Planet Money.
While the podcast feature may appear similar to Google’s NotebookLM, users of Oboe do not need to upload materials to generate content, although they can if they choose. However, this flexibility also raises trust issues; I find myself more inclined to rely on NotebookLM, knowing its information stems from my own documents, while Oboe’s podcasts are generated from unspecified models without citations.
As user engagement grows, Zicherman believes Oboe will evolve in alignment with how users request information, much like a personal tutor enhances its approach over time. He notes, “Oboe isn’t just a platform providing tailored experiences; it improves as you engage with it, similar to how a human tutor becomes more attuned to a student’s needs.”
For the moment, I find greater value in insights directly from human experts, even though identifying them amid the overwhelming information online presents its own challenges.
0 Comments
