Good morning and welcome to Tech News Now, TheStreet's daily tech rundown.
I've been covering Nvidia's blowout earnings since they started last year, and my eyes still popped out of my head at the revenue numbers the chipmaker posted yesterday. The Nvidia story seems far from over.
In today's edition, we're covering Nvidia's enormous earnings beat, Reddit's reported content licensing deal with Google, Google's launch of a new artificial intelligence model, the fixing of ChatGPT and criticism of Meta's AI training practices.
Tickers we're watching today: (NVDA) .
Check out our coverage yesterday of Pindrop, a cybersecurity firm that has developed an AI tool designed to detect and flag deepfake audio in real time.
Let's get into it.
Related: How the company that traced fake Biden robocall identifies a synthetic voice
Nvidia delivers 'game-changing moment'
Shares of Nvidia, which spent all of last year surging to mountainous heights, spiked around 11% in premarket trading following another enormous earnings beat from the semiconductor giant.
Nvidia reported revenue of $22.1 billion for its fiscal fourth quarter, a 265% year-over-year increase that, in the words of Wedbush's Dan Ives, crushed Wall Street's expectations of $20.4 billion.
The company's data-center division — which was a focus for investors heading into the report — reported sales of $18.4 billion, 409% year-over-year growth.
The company again guided strongly for the current quarter, forecasting revenue of $24 billion, far beyond Wall Street's expectations.
“Fundamentally, the conditions are excellent for continued growth in 2025 and beyond,” CEO Jensen Huang said.
Related: Nvidia to Wall Street: Artificial intelligence is just getting started
Nvidia analysts respond: 'The story is intact'
Deepwater's Gene Munster said that in the near term, the stock is up because of strong revenue guidance; long term, the "story is intact."
"The business is powering along despite the headwind from China restrictions," he said. "The reason is we are still early in the first wave of the AI infrastructure wave, selling to hyperscalers and AI startups."
Ross Gerber, CEO of Gerber Kawasaki Wealth & Investment Management, called Nvidia "the most impressive company" he's ever seen.
Even at its current valuation — the stock closed at $674 a share yesterday — Gerber thinks the stock is "wildly undervalued."
Ives wrote in a note that this "robust" quarter from the chipmaker proves his view that the "AI revolution is just starting." He called it the "biggest moment for the market and tech sector in many years."
"This was a game-changing moment for the tech bulls and puts jet fuel in the tech bull market thesis," Ives said.
Nvidia's massive growth comes as concerns about the environmental impact of AI remain unaddressed. Sasha Luccioni, AI researcher and ethicist, highlighted the ongoing lack of "information about the manufacturing process" of Nvidia's chips, asking the company to report the carbon footprint of its graphics-processing units.
Related: Artificial Intelligence is a sustainability nightmare - but it doesn't have to be
Reddit's $60 million licensing deal is with ... Google
Reddit's $60 million licensing deal with an unnamed AI company — first reported Tuesday by Bloomberg — is reportedly with Google (GOOG) , according to Reuters.
The contract with the tech giant is valued at about $60 million annually, a source told Reuters.
Neither Reddit nor Google was immediately available for comment.
W/ the new Reddit data licensing deal, some important questions remain:
— MMitchell (@mmitchell_ai) February 21, 2024
-How will they treat content that infringes the copyright of others?
-Is user consent accounted for?
-Do users have a way to opt out?
-Is all user content since 2005 included?https://t.co/TB0dw16tjt
The licensing deal comes as the social media platform is looking to make its initial public offering, a move that sources told Reuters could come this week.
With copyright issues a central component of such licensing deals, it's worth noting that Reddit's terms of service make clear that users grant the company a "worldwide, royalty-free, perpetual, irrevocable, non-exclusive, transferable and sublicensable license to use, copy, modify, adapt, prepare derivative works of, distribute, store, perform, and display Your Content ... in all media formats and channels now known or later developed anywhere in the world."
Open, not open-source: Google's Gemma
Google on Wednesday introduced two new open, not open-source, AI models christened Gemma. The models were built from the same research that underpins Google's larger Gemini models.
Open, Google DeepMind researcher Lucas Beyer said, means that the model is directly accessible; it's not behind an application-programming interface. Open-source, on the other hand, allows for open access (with the opportunity to modify and redistribute) the source code.
ChatGPT going haywire + Google launching a new, accessible LLM = a great day for open-source AI 🤗
— Sasha Luccioni, PhD 💻🌎🦋✨🤗 (@SashaMTL) February 21, 2024
Gemma, according to Google, is "built for the open community of developers and researchers powering AI innovation."
Google plans to make the model available to developers for free
The tech giant said in a blog post that the smaller Gemma models outperform larger models on key benchmarks. In a technical report, Google estimated the carbon footprint of training the Gemma models to be around 131 tons of CO2 equivalent, though the company does not mention the environmental cost of running the models.
Related: ChatGPT maker has a strong response to New York Times lawsuit
The AI Corner: The fixing of ChatGPT
Yesterday, users on X pointed out the breaking of ChatGPT, highlighting instances in which the model ended "each reply with hallucinated garbage."
OpenAI fixed the issue yesterday, noting that a bug was affecting the way the model processes language.
"Large Language Models generate responses by randomly sampling words based in part on probabilities. Their 'language' consists of numbers that map to tokens," OpenAI said. "In this case, the bug was in the step where the model chooses these numbers. Akin to being lost in translation, the model chose slightly wrong numbers, which produced word sequences that made no sense."
Yes, ChatGPT has been generating some weird stuff.
— Sasha Luccioni, PhD 💻🌎🦋✨🤗 (@SashaMTL) February 21, 2024
No, it's not schizophrenic.
The thing with ML-based systems of this size is that a simple component update can create completely unintended ripple effects in the whole codebase.
That's why we need more transparency! https://t.co/EvbX8BF8JS
AI researcher Gary Marcus yesterday said that regardless of how quickly this specific issue gets fixed, it is indicative of a wider problem.
"The reality, though is that these systems have never been stable. Nobody has ever been able to engineer safety guarantees around them," Marcus wrote. "The need for altogether different technologies that are less opaque, more interpretable, more maintainable and more debuggable — and hence more tractable — remains paramount."
Related: Senate Judiciary Committee seeks to build new framework to rein in Big Tech
Privacy CEO calls on FTC to investigate Meta
Luiza Jarovsky, CEO of Implement Privacy, yesterday called on the Federal Trade Commission to investigate Meta's (META) AI-training practices.
Jarovsky pointed out a recent blog post from the FTC in which the agency said that companies quietly changing their terms of service could be "unfair or deceptive."
She pointed out several instances in which Meta or its CEO, Mark Zuckerberg, have said that the company trains its AI models on user data; at the same time, Meta's privacy policy makes no mention of AI in its section "How do we use your information."
"Whatever is going on here deserves a further look from data protection authorities," she said.
Meta did not immediately return a request for comment.
Contact Ian with AI stories via email, ian.krietzberg@thearenagroup.net, or Signal 732-804-1223.
Related: The ethics of artificial intelligence: A path toward responsible AI
