AI in brief Nvidia’s second-generation A100 GPUs, currently the subject of export controls that mean they’re not for sale in China, can fetch up to $20,000 in black markets – double the regular price.
Last year, the Department of Commerce blocked top US hardware companies, including Nvidia and AMD, from selling their most advanced chips to China and Russia in a bid to harm the countries’ technological prowess. Under the new rules, Nvidia was prohibited from exporting its A100 and H100 GPUs.
But a Reuters report reveals that developers in China could still procure small batches of A100s – by importing them from other Asian countries like India, Taiwan or Singapore, in black markets, and even in some electronics stores. A single GPU can reportedly go for $20,000 in Shenzhen, if you ask discreetly at specialized stores.
Ivan Lau, co-founder of Hong Kong’s Pantheon Lab, said he was trying to buy two to four A100 chips to develop AI models. He said he was in talks with two sellers that had bought the gear outside the US, and had set a price of HK$150,000 ($19,150) per card. “They told us straight up that there will be no warranty or support,” he said.
Buying or selling these types of chips isn’t illegal within China, but is frowned upon. They are difficult to find in large quantities, since US companies are not allowed to export them directly, making it more difficult for Chinese developers to train large, powerful models.
“If we receive information that a customer is breaching their agreement with us and exporting restricted products in violation of the law, we would take immediate and appropriate action,” Nvidia said in a statement.
US government misses chance to address AI-generated political adverts
The Federal Election Committee is torn over whether it should prohibit political adverts created using generative AI tools. In a petition, the FEC was asked to clarify if potentially misleading AI-made ads are considered “fraudulent misrepresentation of campaign authority.” Half of the six-member committee voted to fulfil the petitioner’s request, while the remainder are opposed to any action.
“The Commission discussed a Notice of Availability seeking comment on whether the Commission should initiate a full rulemaking on a proposal in the Petition for Rulemaking from Public Citizen but did not approve publication of the notice by the required four affirmative votes,” the regulator confirmed in a somewhat convoluted statement this week.
Concerns that AI-generated content could undermine democracy by spreading disinformation to manipulate people’s views of political candidates are real. The GOP recently published a fake anti-Joe Biden ad created with AI, and another one made for Ron deSantis’s campaign targeted his main rival Donald Trump.
Should these be regulated or banned outright? It’s not clear how officials plan to tackle this issue, and social media platform owners like Meta seem unsure too.
Senator Chuck Schumer unveils AI regulatory framework
Senate Majority Leader Chuck Schumer launched his AI policy: SAFE Innovation in the AI Age this week.
At an event hosted by the think tank Center for Strategic and International Studies, Schumer said the regulatory framework aims to mitigate AI safety risks whilst supporting US innovation.
“Our framework must never lose sight of what must be our North Star: innovation,” he said in his speech. “But if people don’t think innovation can be done safely, that will slow AI’s development and even prevent us from moving forward. So my SAFE Innovation framework calls for Security, Accountability, protecting our Foundations, and, lastly, Explainability – one of the most important and most difficult technical issues in all of AI.”
For security, Schumer recommends enforcing guardrails that ensure bad actors cannot use AI to harm others, as well as protecting people’s jobs to prevent mass unemployment and income inequalities. He believes that companies building and deploying software must be held accountable for its impacts – which includes subjecting them to audits.
But trying to study models and algorithms under the hood requires them to be explainable. People need to be able to understand how a machine arrives at a particular decision. Schumer described this as the most difficult technical issue problem in AI.
“Now let me share my second proposal: a new legislative approach for translating this framework into legislative action. Later this fall, I will convene the top minds in artificial intelligence here in Congress for a series of AI Insight Forums to lay down a new foundation for AI policy,” he said.
IBM builds AI commentator for Wimbledon
A generative AI system built by engineers at IBM will provide commentary at this year’s Wimbledon tennis tournament from July 3–16.
The All England Club, which manages the Wimbledon venue, tapped Big Blue to develop new AI-powered features for the official Wimbledon app and website. Using tools from the IBM watsonx platform, the system will generate audio commentary and text captions summarizing the best bits of a tennis match captured in video highlights.
“The tool has been designed to give fans a more insightful experience when catching up on key moments from matches with highlights videos on the Wimbledon app and wimbledon.com,” IBM said in a canned statement. “Its introduction this year is a step towards making commentary available in an exciting way for matches outside of Wimbledon’s Show Courts, which already have live human commentary.”
The AI commentator was trained on tennis lingo uttered by experts, but does not replace them. “I see AI as very much complementing the human element, rather than replacing,” said Kevin Farrar, IBM sports partnerships leader, according to The Telegraph.
“You can’t replace John McEnroe doing commentary, that human element always needs to be there. It’s very much supplementing and complementing. For Wimbledon, it’s about providing commentary in the future on matches that don’t currently have human commentary – like the seniors, juniors, wheelchair [events].” ®
