|
Each Tuesday, we turn over Consequence Daily to Features Editor Wren Graves, who writes about trends in the entertainment and media industry and publishes a pop-culture crossword puzzle...
It’s been a wild week, and when the history books are written TikToks are danced, Elon Musk’s... uhh salute is unlikely to be top story. Generative AI is sprinting to change the world and it’s getting harder to keep up.
First there was Trump’s decision to rescind a Biden executive order on AI safety, which was quickly followed by a massive pledge to accelerate AI -- up to $500 billion for data centers and electricity generators, part of a project called Stargate. With all of this plus [gestures vaguely at everything], many people missed DeepSeek’s rollout of their Large Language Model (LLM). But by Friday the implications took over the news cycle, just about erasing Trump’s announcements from social media’s goldfish-sized memories.
Last Monday, the Chinese startup DeepSeek unveiled R1, an open-source LLM that matched or exceeded America’s top products, including OpenAI’s GPT-4.o1, in a series of third-party tests. More eyebrow raising was DeepSeek’s claim that it only took two months and less than $6 million to build it -- a tiny fraction of the soon-to-be trillions spent by American companies. Granted, it hasn’t yet outperformed OpenAI’s newest model, o3 (CEO Sam Altman apparently took lessons on how to name software versions from Microsoft "10 comes after 8" Windows). But DeepSeek’s paper explaining the methodology has upended our understanding of AI, either casting doubt on how Silicon Valley has been spending their billions, or pointing out how to use all that processing power with much more force.That’s without even mentioning yesterday’s AI-driven stock market tumble or the cyber attack against DeepSeek. It’s enough to make anyone dizzy, plus if you cover music like I do, you’re already wondering how early is too early to pre-game the Grammys (Wednesday is too early, Thursday is up for debate).
To help us make sense of it all is Maryam Meseha, founding partner at the tech-focused law firm Pierson Ferdinand LLP. Among other things, Meseha counsels businesses using AI on how to navigate regulatory frameworks, implement AI governance, and ensure that what they’re doing is... you know... ethical. In other words, she’s really good at breaking down the technical world of AI into language the rest of us can follow.The following interview has been lightly edited.
DeepSeek is challenging conventional wisdom about the cost of AI. Does this change our understanding of how AI scales? For example, could OpenAI accomplish even more if they learned from DeepSeek, or do upcoming investments like Stargate now seem like a waste of resources?
DeepSeek’s achievements challenge the narrative that AI scalability requires massive investments. The efficiency demonstrated by R1 suggests that innovation in training techniques and hardware optimization can significantly lower costs without sacrificing performance.OpenAI and other developers could potentially learn from these advancements, rethinking resource allocation and exploring cost-effective methodologies. However, it’s essential to note that innovation isn’t just about cost; responsible scaling must also prioritize security, ethical safeguards, and transparency. Investments like Stargate remain relevant if they push the boundaries of what AI can achieve responsibly, but DeepSeek’s model undoubtedly raises the bar for justifying such expenditures.
Given the open-source nature of R1, will it continue to be vulnerable to cyber attacks? Are there any other potential unintended consequences of open-source AI?
Open-source AI inherently comes with greater cybersecurity risks. The transparency that drives innovation also exposes the model to manipulation by bad actors, making it easier for them to identify and exploit vulnerabilities. The recent cyberattack on DeepSeek underscores this challenge. Without robust monitoring and frequent updates, open-source AI will remain a prime target for cyber criminals.Beyond cybersecurity, open-source AI amplifies risks of misuse. Bad actors could repurpose models like R1 for malicious purposes, such as disinformation campaigns, automated cyberattacks, or the creation of deepfakes. These unintended consequences highlight the need for safeguards that balance innovation with accountability. Open-source developers must take proactive steps to mitigate these risks, including partnerships with cybersecurity experts and the adoption of ethical governance guidelines.
Last week Trump rolled back a Biden executive order meant to curb some potential dangers of AI. How will that decision impact the development of responsible AI?
The rollback of Biden’s executive order represents a step backward in the effort to create a unified framework for responsible AI development in the US. The original order emphasized addressing AI bias, transparency, and misuse, areas critical to maintaining public trust and mitigating risks as AI technologies proliferate.Without federal leadership, the regulatory landscape in the US remains fragmented, with businesses navigating conflicting state-level rules. This lack of cohesion weakens the country’s ability to lead on the global stage, particularly as the EU has established more robust governance frameworks through their EU AI Act going into effect in February. The decision creates a vacuum in AI leadership, potentially paving the way for unchecked innovation that prioritizes speed and profitability over safety and ethics.To ensure responsible AI development, the U.S. must recommit to creating comprehensive regulations that prioritize transparency, accountability, and ethical safeguards. Failure to do so risks not only public trust but also the country’s standing as a leader in the AI space. |
