Arvind Narayan has an excellent retort to the pausing AI “development”.
Here’s this:
A misleading open letter about sci-fi AI dangers ignores the real risks
The letter refers to a common claim: LLMs will lead to a flood of propaganda since they give malicious actors the tools to automate the creation of disinformation. But as we’ve argued, creating disinformation is not enough to spread it. Distributing disinformation is the hard part. Open-source LLMs powerful enough to generate disinformation have also been around for a while; we haven’t seen prominent uses of these LLMs for spreading disinfo.
I disagree. Those with means to create disinformation clearly have the means to distribute. If it’s not happening now, it will, subsequently. For example, CNET and Buzzfeed (both run by opaque financial entities) masquerade as “mainstream media”) and generate several factual errors, in their rush to show profitability and create SEO laden spam. That’s also one reason why I avoid the generative AI to create blog posts because you can smell the BS from far away.
Here’s another interesting point:
GPT-4 was released to much hype around its performance on human exams, such as the bar and the USMLE. The letter takes OpenAI’s claims at face value: it cites OpenAI’s GPT-4 paper for the claim that “contemporary AI systems are now becoming human-competitive at general tasks.” But testing LLMs on benchmarks designed for humans tells us little about its usefulness in the real world.
From their earlier blog-post:
GPT-4 and professional benchmarks: the wrong answer to the wrong question
In some real-world tasks, shallow reasoning may be sufficient, but not always. The world is constantly changing, so if a bot is asked to analyze the legal consequences of a new technology or a new judicial decision, it doesn’t have much to draw upon. In short, as Emily Bender points out, tests designed for humans lack construct validity when applied to bots.
On top of this, professional exams, especially the bar exam, notoriously overemphasize subject-matter knowledge and underemphasize real-world skills, which are far harder to measure in a standardized, computer-administered way. In other words, not only do these exams emphasize the wrong thing, they overemphasize precisely the thing that language models are good at.
I agree, but the intent has been demonstrated. Those are the initial first steps. It is true that you can’t predict the nature of a person as a baby, but in the real world interactions give a fair idea of how a person addresses challenges. The critique is on their current shortcomings. It has nothing to do with how the potential evolves with funding or intent.
Long term risks:
Long-term catastrophic risks stemming from AI have a long history. Science fiction has primed us to think of terminators and killer robots. In the AI community, these concerns have been expressed under the umbrella of existential risk or x-risk, and are reflected in the letter’s concerns about losing control over civilization. We recognize the need to think about the long-term impact of AI. But these sci-fi worries have sucked up the oxygen and diverted resources from real, pressing AI risks — including security risks.
The authors conclude by:
A better framework to regulate the risks of integrating LLMs into applications is product safety and consumer protection. The harms and interventions will differ greatly between applications: search, personal assistants, medical applications, etc.
Regulation, therefore, has been a recurring theme in the past couple blog posts (please note that I bunch them together while writing, and all are scheduled for the “future”). Hence, you’d notice a consistent theme for generative AI, at least. Regulation.
Good counter points, but I’d tend to broadly disagree. Nation states will deploy any means necessary. WhatsApp, for example, is the tool that uses generated nonsense to allow forwarding, and remains one of the most widely accepted spyware. Large Language Models (even if it’s not ChatGPT), are proliferating like cancer. Someone will weaponize any one of them.
RadOncNotes 