Meta’s “open-source” misstep & Nvidia’s chip troubles

Welcome back to a new edition of AI Odyssey!

TSMC posted record profits for the third quarter, with net profit surging 54% driven by strong demand for AI and smartphone chips, beating analyst expectations.

1: Meta criticized for misusing "open-source" label

The news: Meta is under fire from the Open Source Initiative (OSI) for labeling its Llama AI models as "open-source," with OSI arguing the term is being misused and confusing users.

Why it matters: OSI’s Stefano Maffulli claims Meta’s models don’t meet true open-source standards, lacking transparency and restricting competition, even as the European Commission promotes truly open technologies.

Driving the news: Meta offers Llama for free but doesn’t disclose key elements like training algorithms, limiting developers from fully building on the models—unlike traditional open-source software.

What they’re saying: Maffulli calls Meta’s approach "polluting" the open-source term, while Meta insists they’re committed to open-source AI but argue current definitions don’t cover modern AI complexities.

The bottom line: Meta’s approach is raising questions about transparency and competition, with critics warning it could undermine open AI development as regulators push for clearer standards.

2: AI video tools are going mainstream

For years, tech companies have teased AI video-making tools, but now they’re finally hitting the public. Adobe just launched a beta of its Firefly Video Model, giving Creative Cloud subscribers the ability to turn photos or prompts into short video clips.

Why it matters: AI-powered video creation opens up huge creative possibilities, but also raises concerns over deepfakes and misinformation.

Driving the news: Adobe’s Firefly Video Model and Generative Extend feature for Premiere Pro are now in public beta. Meta, OpenAI, and Google are also ramping up their video AI tools for broader release.

The big picture: As AI video tech grows, companies are racing to develop safeguards, like Truepic’s content verification system, to limit misuse.

Yes, but: Running these services is costly and risky. Adobe is restricting video generation to paid users and banning certain content, while others push for broader adoption of verification tools across devices like smartphones.

What’s next: The industry is watching for Apple and Google to integrate authentication tech into Android and iOS cameras, which could significantly impact how AI-created content is verified.

3: The New York Times demands AI search engine Perplexity stop using its content

The New York Times has issued a cease-and-desist letter to AI search engine startup Perplexity, accusing it of using Times content without permission. This follows similar claims made by Forbes and Condé Nast earlier this year. The Times, already suing OpenAI and Microsoft over AI model training using its articles, argues Perplexity has unjustly benefited from its journalism.

Perplexity’s stance: The startup denies scraping content for AI training, stating it only indexes web pages to surface factual content. A company spokesperson emphasized that "no one organization owns the copyright over facts." Perplexity has made deals with publishers like Fortune and Time, and its CEO, Aravind Srinivas, expressed willingness to collaborate with the Times and other publishers.

Perplexity plans to respond to the Times' notice by October 30th.

4: Nvidia and TSMC’s AI chip partnership faces strain

Nvidia’s crucial relationship with its chipmaker, Taiwan Semiconductor Manufacturing Co. (TSMC), is feeling the pressure as demand for AI chips surges. Their partnership, vital to the AI boom, recently hit a bump with delays in the production of Nvidia’s new Blackwell chips, designed for powerful AI software.

Why it matters: Nvidia relies heavily on TSMC to build its cutting-edge chips, but tensions are rising as both companies struggle to meet the demands of the fast-growing AI market.

The issue: After Nvidia announced its Blackwell chip family in March, testing revealed that the chips faltered in high-voltage data center environments, leading to finger-pointing between the two companies. Nvidia partially blamed TSMC’s rushed production process, while some TSMC employees suggested design flaws were at fault.

The stakes: The hiccup highlights the immense pressure on both Nvidia and TSMC as they try to keep up with the explosion in AI demand. Nvidia, now the world’s second most valuable company, needs TSMC to expand its chip manufacturing capabilities to meet the skyrocketing needs of AI developers.

Yes, but: Despite these tensions, both companies are deeply interdependent. Nvidia accounts for 10% of TSMC’s revenue, and TSMC is essential to Nvidia’s production of AI chips. But Nvidia is also exploring alternatives, including working with Samsung for future chip production.

What’s next: Nvidia has made tweaks to the Blackwell design, with mass production expected to start in the fourth quarter of 2024. However, significant shipments may not begin until early 2025.

AI NOTES 🗒️

Watch: Google signs deal to build seven nuclear reactors to satisfy AI energy hunger (DW News)

Extra read: This LLM framework takes a first stab at benchmarking Big AI’s compliance with the EU AI Act (TechCrunch)