| In today’s edition, we have the scoop on Sundar Pichai’s memo to employees on the problems found in ͏ ͏ ͏ ͏ ͏ ͏ |
 | Reed Albergotti |
|
Hi, and welcome back to Semafor Tech.
The news never takes a break, especially with everything going on in AI. So while on vacation, I got word yesterday of Google CEO Sundar Pichai’s staff memo addressing the worsening woes of its Gemini AI model. We put out the scoop and you’ll see it below.
We wrote last week about how Gemini’s image generation product had gone off the rails, refusing to depict images of white people and creating inaccurate historical images that inserted people of color. The problems got worse as Gemini’s text outputs also showed signs of trouble. In one case, it said it was a tossup on who had a more negative impact on society: Elon Musk or Adolf Hitler.
The problems led to a $90 billion stock hit for parent company Alphabet on Monday.
Here’s my take: This is not a technology problem. Google is at the very forefront of AI. The underlying Gemini model is comparable in power to other frontier AI models, like GPT-4. Instead, Google messed up on the guardrails of AI.
I’ve seen a lot of reporting about how Google’s progressive culture led to this outcome but I don’t think this is about culture wars. Most companies are trying to tame the LLMs they release publicly. And that includes inserting more diversity, so that if you ask an AI model to generate an image of a doctor, lawyer, or banker, it doesn’t spit out white men 100% of the time.
There is a debate to be had here, but it’s not Google-specific. Some people believe LLMs should have fewer guardrails. They have a point. With every attempt to tame generative AI models, they get a little less useful. I think people who use these tools are largely responsible adults who can determine what results are biased, inaccurate or inappropriate, and disregard those outputs. But let’s face it: No public company looks at it that way.
 Drew Angerer/Getty Images➚ MOVE FAST: Double down. U.S. Federal Trade Commission Chair Lina Khan is taking some credit for Nvidia’s upswing, saying her agency’s 2021 move to block the chipmaker’s Arm deal led to further innovation. Her comments show she’s sticking to her anti-merger stance, despite some losses in the courts. ➘ BREAK THINGS: Give up. Apple ditched its plans to build an electric car as AI becomes a priority. Some of its auto-related employees will now work on artificial intelligence products. Wall Street seemed to like the move, but the company needs to merge into the fast lane if it wants to catch up on AI. |
|
French startup Mistral has launched a new model: Mistral Large. And the reviews are pretty good, with the free, open-source model performing at the same general capabiltiy of OpenAI’s GPT-4. But some people using Mistral’s service have noticed another benefit: It’s less reserved than models from OpenAI and other tech companies. Companies like Perplexity have already begun to offer Mistral Large as an option. (In Perplexity’s case, Mistral’s models can power its generative search product.) In a nod to Mistral’s growing influence in tech, Microsoft recently invested in the company. That’s notable, considering Satya Nadella’s relationship with OpenAI. |
|
Sundar Pichai: Google working ‘around the clock’ to address Gemini problems |
 Reuters/Gonzalo FuentesTHE SCOOP Google CEO Sundar Pichai addressed the company’s Gemini controversy Tuesday evening, calling the AI app’s problematic responses around race unacceptable and vowing to make structural changes to fix the problem. Google suspended its Gemini image creation tool last week after it generated embarrassing and offensive results, in some cases declining to depict white people, or inserting photos of women or people of color when prompted to create images of Vikings, Nazis, and the Pope. The controversy spiraled when Gemini was found to be creating questionable text responses, such as equating Elon Musk’s influence on society with Adolf Hitler’s. Those comments drew sharp criticisms, especially from conservatives, who accused Google of an anti-white bias. Most companies offering AI tools like Gemini create guardrails to mitigate abuses and to avoid bias, especially in light of other experiences. For instance, image generation tools from companies like OpenAI have been criticized when they created predominately images of white people in professional roles and depicting Black people in stereotypical roles. “I know that some of its responses have offended our users and shown bias – to be clear, that’s completely unacceptable and we got it wrong,” Pichai said. He said the company has already made progress in fixing Gemini’s guardrails. “Our teams have been working around the clock to address these issues. We’re already seeing a substantial improvement on a wide range of prompts,” he said. Read here for Pichai's full memo. → |
|
|
 Today, we’ve opened registration for the 2024 World Economy Summit – our boldest venture in live journalism yet and the only major media event to be held against the background of the IMF and World Bank meetings, taking place in Washington, D.C. on April 17-18. Legends of industry, Penny Pritzker and David Rubenstein will serve as our co-chairs, and today we’re thrilled to unveil our speaker line-up of global leaders on the record. Speakers, Sessions & Registration here. |
|
 Stripe’s valuation in its newly announced tender offer, allowing current and former employees to cash in on private shares while the IPO market remains choppy. The fintech’s worth, at least on paper, has also been a bit volatile over the last several years. About a year ago, it was pegged around $50 billion; in 2021 it was at $95 billion. |
|
 Reuters/Elizabeth FrantzThe Biden administration is looking to keep foreign rivals from buying large troves of sensitive data on Americans through third parties, Semafor’s Morgan Chalfant reports. President Biden is signing an executive order today empowering the Justice Department to block certain bulk transactions — like genomic data, personal health information, and geolocation data — to foreign “countries of concern,” officials said, citing risks to U.S. national security. A senior Biden administration official singled out China and Russia. The executive order will also restrict data transfers to North Korea, Iran, Cuba, and Venezuela, a senior DOJ official said. There will be carve-outs to prevent disruptions to commerce — such as for processing financial transactions — and the executive order triggers a lengthy rulemaking process, meaning it won’t go into effect for several months. |
|
 Facebook/ScreenshotMark Zuckerberg is the latest tech leader to make a swing through Asia as the global AI competition heats up. He met with Japanese Prime Minister Fumio Kishida on Tuesday and is now in South Korea. In Seoul, he is meeting with executives of Samsung Electronics and LG Electronics to discuss AI chips, extended reality, and other issues, according to local media. Open AI CEO Sam Altman and Nvidia boss Jensen Huang are among the others who have recently visited the region. It’s not all about business. Zuckerberg also gave rave reviews for his meal at a McDonald’s in Japan and helped make a sword with a Japanese master. He will also go to India to attend celebrations for billionaire Mukesh Ambani’s youngest son, who is getting married. |
|
