Oboe AI: Learning’s Credibility Gap

Oboe, a new AI-powered learning platform from former Spotify executives, promises to streamline education with personalized courses. However, its current lack of verifiable sources and citations raises significant concerns about the credibility of its information.

In an era increasingly defined by the dazzling, often disorienting, ascent of artificial intelligence, a fundamental question echoes through our collective consciousness: Will these powerful algorithms elevate us, or will they ultimately dull our minds, making us intellectual dependents?

It’s a query that Oboe, a new AI-powered learning platform, directly confronts, positing an emphatic “no” to the notion of AI-induced intellectual atrophy.

Born from the minds of former Spotify executives Nir Zicherman and Mike Mignano, Oboe steps onto the stage with a bold promise: to democratize access to a superior learning experience, affordably, and all powered by AI.

Zicherman, Oboe’s CEO and co-founder, alongside Mignano, who previously helmed Spotify’s podcast team after their DIY podcast platform Anchor sold to the audio giant for $150 million, are no strangers to disrupting digital landscapes.

Now, they aim to revolutionize how we acquire knowledge.

Their vision, as articulated by Zicherman, is to streamline the current fragmented learning process – the endless tabs of ChatGPT, Google searches, YouTube videos, and Wikipedia articles – into a “single destination” for effective learning.

It’s an alluring proposition, especially when framed against the backdrop of an internet increasingly flooded with disparate, often unreliable, information.

Upon entering Oboe’s digital domain, users are greeted by a familiar chatbot-like interface, a simple text box inviting them to declare their intellectual curiosity.

The difference, however, is stark.

Unlike the conversational back-and-forth of ChatGPT, Oboe delivers a structured “course” in response.

This could manifest as a comprehensive essay akin to an introductory textbook chapter, a concise bulleted list of key takeaways, or even an AI-generated podcast.

One could, for instance, delve into the origins of AI or the multifaceted forms of intelligence found in nature. Or, like this observer, one might venture into the seemingly mundane yet critically important realm of concrete manufacturing and its environmental footprint.

The resulting “deep dive” on concrete manufacturing was presented as a 10-minute read, peppered with headers, tables, and images sourced from Wikimedia Commons.

It offered a surface-level understanding, a quick skim of a complex industrial process.

Yet, as the digital pages turned, a familiar unease began to creep in.

The text mentioned “SCM,” an acronym crucial to the subject, but offered no immediate definition.

A quick search within the generated course yielded no answers.

More significantly, a gaping void emerged: the absence of citations.

This isn’t a minor oversight; it’s a profound challenge to the very foundation of credible learning.

Oboe, in its current iteration, does not provide links to original source materials.

While a statement like “aggregates are granular materials such as sand, gravel, or crushed stone that make up the majority of concrete’s volume (typically 60–75%)” might indeed be accurate – and in this case, a quick external check confirmed it with the American Cement Association – the onus of verification falls entirely on the learner.

This effectively negates the promise of a streamlined, efficient learning experience, transforming it instead into a new form of digital archaeology, where users must dig for the truth that the platform itself omits.

Zicherman acknowledges this critical gap, stating that “citations and other means of accessing additional online resources are therefore things we are actively working on and hope to add to the platform in the coming months.”

But “coming months” feels like a long time to wait when the integrity of information is paramount.

Other AI platforms already offer a degree of transparency.

Google’s Gemini, when prompted for citations, readily provided a link to a 2018 Nature Sustainability paper detailing concrete’s global water usage.

This isn’t just about academic rigor; it’s about trust.

In a world awash with “AI slop,” as some have termed it, the ability to trace information back to its source, however imperfectly, provides a much-needed anchor.

Without it, Oboe risks becoming another conduit for plausible-sounding but unverifiable content.

Oboe’s technical approach to mitigating the pervasive issue of AI hallucinations is equally intriguing, if not entirely reassuring.

Zicherman reveals that the platform doesn’t train its own foundational models but rather selects and optimizes models from unnamed third-party providers.

The solution to potential inaccuracies? Tasking some large language models (LLMs) with the job of correcting the outputs of other LLMs.

It’s a fascinating, almost recursive, approach – an AI checking another AI.

But what happens if both LLMs hallucinate, or if the “checking” LLM itself introduces new inaccuracies?

And for now, users have no direct mechanism to flag inaccurate content; corrections must be reported directly to the company, a process Zicherman assures is in place for “reviewing and incorporating” feedback.

Beyond text, Oboe also ventures into audio, generating podcasts narrated by chipper, AI-generated hosts.

My concrete manufacturing podcast, for example, broke down the topic in a conversational style reminiscent of NPR’s Planet Money.

While similar to Google’s NotebookLM in generating audio from content, Oboe’s freedom from requiring user-uploaded material comes at a cost.

With NotebookLM, there’s a degree of trust because the content originates from your documents.

With Oboe, the unnamed models and lack of citations left me struggling to fully trust what I heard.

Zicherman likens Oboe’s continuous improvement to that of a human tutor, learning a student’s unique learning style over time.

“Oboe is not just a product that gives you a great personalized experience, it’s also a product that gets better the more you use it,” he explains.

It’s a compelling analogy, hinting at a future where AI truly adapts to individual needs.

Yet, for all its futuristic promise, the present reality of Oboe underscores a persistent tension in the age of AI.

While the allure of instant, personalized knowledge is undeniable, the fundamental requirement of verifiable truth remains paramount.

Until platforms like Oboe can robustly address the critical issue of source attribution and accuracy, many of us will likely continue to prefer the sometimes harder, but ultimately more reliable, path of seeking out human experts, even amidst the vast, ever-expanding digital ocean increasingly churned by AI.