For the field of generative AI (GenAI), this month has posed significant challenges. Amazon Web Services (AWS) made a splash with the introduction of Amazon Q, positioning it as a competitor to Microsoft’s Copilot. However, internal warnings about “serious illusions” and data leaks from AWS staff cast a shadow over this launch. Similarly, Google unveiled Gemini, its answer to ChatGPT, amid much excitement and a dazzling demo, only to later concede that the demo was misleading. Meanwhile, Meta made waves by releasing new open-source AI health tools. Despite these advancements, a critical flaw of GenAI tools—their susceptibility to quick shot attacks—remains a glaring concern.
The recent setbacks in the GenAI sector do not signify its demise or insignificance. Rather, they highlight how the allure of AI potential has sometimes overshadowed reality. The truth is, there is substantial promise in this field, and we need not resort to hyperbole.
While open source may not be a perfect fit for GenAI, it could offer valuable solutions.
Urgency in Action
The launch of Amazon Q by AWS was a pivotal moment for the company, presenting an opportunity to bridge gaps and even surpass competitors. However, despite attempts to position Q as more robust than rivals like ChatGPT, it falls short. This discrepancy, highlighted by Duckbill Chief Economist Corey Quinn, is just one of many issues. AWS’s eagerness to assert superiority before proving it undermines its credibility. As AWS ventures into the realm of AI programs, it faces significant challenges, lacking the historical expertise in this domain, as noted by Quinn.
This rush to meet the AI hype is not unique to AWS. Google, despite its impressive AI track record, resorted to shortcuts in a recent demo. The video showcasing Gemini implied capabilities that the actual product lacks, as pointed out by Parmy Olson and Grady Booch. The pressure to position themselves as AI leaders drives companies like AWS and Google to exaggerate their offerings.
The same trend extends beyond these tech giants, with executives across various industries touting AI in earnings calls. While GenAI technology shows immense potential, the reality often falls short of the hype. By pushing companies to adopt inherently insecure software, we exacerbate the pressing issue of quick shot attacks.
Perhaps, open source collaboration could provide a way forward.
The Role of Open Source
Embracing open source does not guarantee flawless AI systems. However, considering the impact of open sourcing in other IT sectors, it raises questions about why GenAI has been slow to follow suit. The definition of open source in the context of AI remains ambiguous, as highlighted by Meta’s Purple Llama initiative, which sidesteps critical issues like quick shot attacks, as noted by Simon Willison.
Complex systems like Gemini harbor numerous intricacies that may not be fully understood without open sourcing, as Professor Xin Eric Wang suggests. The Open Source Initiative (OSI) grapples with challenges related to access rights, modifications, and redistribution of AI models. While open source offers transparency, it also prompts companies to position themselves more authentically.
Greater transparency through open source could lead to a more honest portrayal of AI capabilities. By shifting the focus from flashy demos to code scrutiny, companies can build trust among developers, who are key users of GenAI. While open source may not solve all issues faced by GenAI vendors, it fosters a culture of accountability and transparency that is sorely needed in the industry.
