Newsletter
GenAI vendors expect you to get sued
Today: why Microsoft is extending legal protections to its customers, the rise of LLMs that weren't trained primarily on English, and the quote of the week.
Welcome to Runtime! Today: why Microsoft is extending legal protections to its customers, the rise of LLMs that weren't trained primarily on English, and the quote of the week.
(Was this email forwarded to you? Sign up here to get Runtime each week.)
Send lawyers guns and money
The potential legal issues generated by the generative AI boom are the kinds of pesky things that most of those in the arena trying stuff like to dismiss out of hand. Microsoft, with a long history of legal wranglings with the government and private industry, isn't taking chances.
Microsoft will essentially pay any legal damages imposed upon customers using its Copilot generative-AI assistants, including GitHub Copilot's coding tool. As long as those customers use the built-in content filters that Microsoft ships with those products, "we will defend the customer and pay the amount of any adverse judgments or settlements that result from the lawsuit," the company announced in a blog post this week
Those lawsuits have already arrived.
- OpenAI and Microsoft have already been sued twice for training large-language models on personal information found on the internet.
- While the issues involved with enterprise generative AI tools aren't as direct as the ones involving image generators like Stable Diffusion and Midjourney, they're going to get a hearing.
- Existing copyright law doesn't clearly indicate whether or not code generated by an LLM trained on huge repositories of other code infringes on the rights of its original creator.
- And former GitHub CEO Nat Friedman anticipated these issues right from the launch of Copilot: "We expect that IP and AI will be an interesting policy discussion around the world in the coming years, and we're eager to participate!" he wrote at the time.
Microsoft is attempting to walk a very fine line with this policy, and it won't be the only vendor on that path.
- It's clear that after six months of hype, most enterprises are very much in test and evaluation mode when it comes to generative AI, and a substantial number are sitting on the sidelines.
- There are many reasons for that reluctance, including security and product quality, but the legal issues have been clear for a while and very few people enjoy getting sued.
- But Microsoft needs more companies to adopt its expensive generative AI tools to show a return on the investments it has made in the technology over the past few years.
- While this new policy will probably encourage some companies to take the plunge, acknowledging that you expect a certain percentage of your customers to get sued merely for using your products is a little dicey.
Generative AI tools market themselves up as magic tools that will finally unlock your business's potential, promising greater productivity and efficiency. But what happens when the first software company gets sued for selling software using code arguably protected by copyright?
- There's no guarantee the courts would allow the defendant to keep selling that software while the case plays out.
- Few companies can afford to have some of their most valuable lines of business in limbo for what could be years.
- It's going to take a long time to settle some of these disputes and odds suggest one case will likely reach the Supreme Court.
- At least Microsoft has one of the most experienced and well-resourced legal teams in technology, and can afford to litigate such disputes on behalf of its customers.
Who wants to go first?
AI in India
If the LLM approach to AI is really going to make an impact on the world, it will need to confront the immense diversity of languages spoken on this planet. Nvidia and Jio Platforms took a big step toward making that happen this week.
Nvidia and Jio Platforms, which owns India's largest wireless carrier, announced a partnership Friday that will provide Jio with Nvidia's highly coveted AI hardware in order to build a new large language model "trained on India’s diverse languages," Techcrunch reported. The plan would see Jio build an AI cloud for Indian researchers and developers over an unspecified time frame.
As noted earlier this week, for AI to be successful outside the Cerebral Valley it will have to be trained on local data, far outside the massive cloud computing complexes that have powered its rise. And given that Nvidia is unable to ship any of that pricey and powerful hardware to China, India seems like a natural place to establish a regional foothold.
Quote of the week
“From a client perspective, what we hope to do through these events is paint a picture of what’s possible — bring a new perspective to each of our clients about how they could be applying these capabilities to their business." — Noah Syken, IBM’s VP of sports and entertainment partnerships, as told to AdWeek about the U.S. Open TV ads that make it out to be at the forefront of technology during a week in which it also raised prices across the board.
The Runtime roundup
Square went down for several hours Thursday afternoon and evening in the U.S., forcing me to pay cash for a burrito.
SAP acquired LeanIX, a startup that helps companies map their IT assets, for more than $1 billion according to reports.
Thanks for reading — see you Tuesday!