Pat Gelsinger returned to Intel in 2021 to lead its turnaround and become a cornerstone of the artificial intelligence sector. Last week, the company’s board forced him out. WSJ reporter Asa Fitch explains what happened and where it leaves the United States’ chip manufacturing industry. Plus, two university students made a website to rate how well AI chatbots perform tasks. We’ll hear why Chatbot Arena’s rankings have become so important to leaders of companies like OpenAI and Google. Danny Lewis hosts.
Sign up for the WSJ's free Technology newsletter.
Learn more about your ad choices. Visit megaphone.fm/adchoices
[00:00:00] Exchanges. The Goldman Sachs podcast featuring exchanges on the forces driving the markets and the economy. Exchanges between the leading minds at Goldman Sachs. New episodes every week. Listen now.
[00:00:18] Welcome to Tech News Briefing. It's Monday, December 9th. I'm Danny Lewis for The Wall Street Journal.
[00:00:24] The people running the biggest names in the artificial intelligence sector, from OpenAI to Google, have their eyes on one website, Chatbot Arena.
[00:00:34] It's a place where people can go and put AI chatbots head-to-head to find out which give the best results.
[00:00:40] We'll hear how a rankings website made by university students became so closely watched by the AI industry.
[00:00:46] And then, last week, Intel CEO Pat Gelsinger abruptly announced he was stepping down from his job
[00:00:52] and the board of the company where he'd worked for most of his career.
[00:00:56] Gelsinger's departure from Intel raises big questions about the ambitious and costly turnaround plan he devised for the company.
[00:01:04] WSJ reporter Asa Fitch explains what this means for the future of Intel and of the United States chip-making industry.
[00:01:14] But first, in just the last few years, there's been an explosion of artificial intelligence chatbots.
[00:01:21] There's OpenAI's ChatGPT, Anthropics' Claude, Google's Gemini, and XAI's Grok, just to name a few.
[00:01:29] But it can be really hard for people who aren't AI researchers to know which ones are the best.
[00:01:34] Enter Chatbot Arena, a website created by two students at the University of California, Berkeley,
[00:01:41] that's quickly become the most-watched ranking of AI systems.
[00:01:45] WSJ tech reporter Miles Krupa has been looking at how they assess these systems.
[00:01:49] He joins us now.
[00:01:50] Miles, how does Chatbot Arena determine its AI rankings?
[00:01:54] What they've created is basically a battle zone or an arena, as they call it, for these chatbots.
[00:02:03] So, you know, anybody can come to the website, ask a question, and get side-by-side responses from two different chatbots,
[00:02:11] and then rank which one they think is the best.
[00:02:14] So what chatbots can you find on there?
[00:02:16] They have most of the ones that people care about, and what they're testing is not so much like the ChatGPT product that we all consume every day,
[00:02:26] or Google's Gemini consumer chatbot that might be on your phone.
[00:02:32] What they're testing is the basic technology, the large language models that power these chatbots.
[00:02:38] And so on that front, they have multiple ones from OpenAI, multiple from Google, Anthropic, Meta, Mistral,
[00:02:48] all of the major players, basically.
[00:02:51] But in addition to, like, a lot of the big names that our listeners might be familiar with,
[00:02:55] there's a ton of startups with their own chatbots and models that have cropped up as the AI industry has grown.
[00:03:01] How does the team behind Chatbot Arena decide which ones to test?
[00:03:05] Yeah, they basically accept any and all submissions for so-called proprietary models that aren't open-sourced.
[00:03:13] They require that the makers of those models give them free access,
[00:03:17] because this is still just a graduate student project after all.
[00:03:22] They have some funding from Berkeley, but these things are still very expensive.
[00:03:26] And so they rely on the companies that provide access to these models to do it free of cost.
[00:03:31] What have they been finding as they test some of these lesser-known chatbots?
[00:03:35] Well, it's interesting. I mean, we've had a few surprises over time on the leaderboard.
[00:03:41] One recently was this Yi model from China, Yi, placed sixth on the leaderboard a few months ago, right after it came out.
[00:03:50] That was a surprise to many in the U.S. who might not have been keeping as close of an eye on the Chinese AI scene
[00:03:58] to have a model kind of come out of nowhere and be immediately competitive with OpenAI and Meta and the rest.
[00:04:05] Are all the models you can test in Chatbot Arena available to the public?
[00:04:09] Yeah. So one thing that they do is they allow these companies to test the chatbots in the arena before they're released publicly.
[00:04:17] Even if they're not already a fully-fledged product, you can test them out on the arena and see what kind of feedback they get.
[00:04:26] And one of the earliest cases of this was OpenAI testing this model.
[00:04:31] It was called something like, I'm also a good GPT-2 bot.
[00:04:36] And so it kind of used OpenAI's naming conventions,
[00:04:40] and people caught on that this was some unreleased new OpenAI model.
[00:04:47] And it generated a ton of buzz.
[00:04:49] It was actually one of the biggest events in terms of traffic for Chatbot Arena and really helped them take off even further.
[00:04:57] Thousands of people were playing with this bot in the arena.
[00:05:00] And then a few weeks later, it was released as GPT-40, which is the model that currently powers ChatGPT.
[00:05:08] That was WSJ reporter Miles Krupa.
[00:05:11] Coming up, former Intel CEO Pat Gelsinger spent most of his career at the Semiconductor Pioneer and was a true believer in its work.
[00:05:19] But when his rescue plan for the storied company failed, its board lost faith in him.
[00:05:25] What does his abrupt exit mean for Intel's future?
[00:05:28] That's after the break.
[00:05:38] In 2021, Intel hired its former chief technology officer, Pat Gelsinger, to lead its turnaround efforts as the company raced to catch up to the exploding demand for the chips that power artificial intelligence.
[00:05:51] He devised an ambitious and expensive plan.
[00:05:54] But it didn't pan out.
[00:05:56] And last week, Gelsinger abruptly resigned, raising big questions about Intel's future.
[00:06:02] WSJ reporter Asa Fitch has been covering this story.
[00:06:05] He joins us now.
[00:06:06] Asa, let's talk about now former CEO Pat Gelsinger.
[00:06:10] What characterized his career at Intel?
[00:06:13] How did he go from an entry-level job to CEO?
[00:06:16] So Gelsinger had two careers at Intel, basically.
[00:06:19] His first career at Intel began when he was 18 years old.
[00:06:21] He was very sharp, very energetic.
[00:06:24] Some at the company said a little bit arrogant.
[00:06:26] But he fit really well in that culture.
[00:06:29] And he became a really important employee of Intel, an important engineer who helped vault Intel's chips to the dominant position that they enjoyed for many, many years.
[00:06:43] He helped design a chip called the 386 and the successor to the 486.
[00:06:47] These chips were essentially ubiquitous in the personal computers of the 80s and 90s.
[00:06:53] So Gelsinger became a protege of the CEO at the time, Andy Grove, who's renowned within the industry and generally in corporate history.
[00:07:02] So Gelsinger rose to the ranks under Grove.
[00:07:05] Eventually, he became the company's first chief technology officer, first CTO in 2000, a position he held for a number of years.
[00:07:14] But eventually, he was forced out of the company in 2009.
[00:07:18] This was over a project that he oversaw to produce a graphics processing unit and try to compete with NVIDIA.
[00:07:26] And it just essentially didn't work out and he was ousted.
[00:07:30] So that was his first career.
[00:07:31] His second career at Intel, of course, was his return as CEO in 2021.
[00:07:36] And that tenure was certainly not as successful as his first one.
[00:07:41] He had a bold vision to transform the company, return Intel to the glory days that enjoyed Under Grove, his mentor.
[00:07:47] And it didn't happen for a lot of reasons.
[00:07:50] But here we are now.
[00:07:51] He's left the company and Intel is searching for a new CEO.
[00:07:55] So why did the company tap him for leadership in 2021?
[00:07:59] Well, Gelsinger was seen as and for sure is a blast from the glorious past of Intel.
[00:08:06] And so the company had struggled a bit under its previous CEOs, former CFO Bob Swan and Brian Krasanich, who came up through the company's manufacturing operations.
[00:08:16] But the company under those two CEOs had kind of struggled with its manufacturing.
[00:08:21] It struggled to continue to be on top and making the best, fastest chips in the world.
[00:08:26] And it ceded a lot of that ground to the likes of Taiwan Semiconductor Manufacturing Corporation and Samsung.
[00:08:32] So when Gelsinger came on, Intel wasn't doing that well.
[00:08:36] It was in a tight spot.
[00:08:38] And Gelsinger had a very bold plan to revive the company and bring it back.
[00:08:44] I mean, it was very costly and ambitious.
[00:08:46] But the idea was to do a couple of things.
[00:08:49] Gelsinger wanted to open a contract chip manufacturing business like what TSMC had done so successfully in the past number of decades.
[00:08:59] And the other main part of the transformation was to essentially catch up with TSMC and Samsung in the race to make the best chips possible in the world on the cutting edge.
[00:09:09] A race that is very, very expensive.
[00:09:12] He was going really counter to where the industry had gone, which is the industry in the past couple of decades, really, maybe a little bit longer, has essentially broken up.
[00:09:23] People usually, they only make chips or they only design chips.
[00:09:28] But Intel under Pat Gelsinger was going to do both and was going to go full steam ahead doing both.
[00:09:35] And unfortunately for Gelsinger, it didn't turn out so well for him or Intel.
[00:09:40] Why did this plan fail?
[00:09:42] One issue was the foundry business, this contract manufacturing business, simply did not gather a ton of customers.
[00:09:48] The other main problem was the dynamics of the market for personal computers and PC chips and server chips.
[00:09:55] You know, at first in Gelsinger's tenure, those businesses did fairly well.
[00:09:59] I mean, the COVID epidemic was still on.
[00:10:02] People were working from home.
[00:10:03] They needed more computers at home and they were using the Internet more.
[00:10:07] And that drove sales of Intel's chips.
[00:10:09] But then came AI.
[00:10:11] And AI changed everything for Intel because the profits from the AI boom went pretty much exclusively to NVIDIA.
[00:10:19] NVIDIA was designing and having these GPUs made by TSMC that became essentially the hottest commodities in tech.
[00:10:27] And everybody needed them to produce models like ChatGPT and other things like that.
[00:10:31] And tech budgets at companies were being spent on NVIDIA's chips, not Intel's.
[00:10:37] What have Gelsinger and Intel said about his departure?
[00:10:40] Gelsinger has said that it was his dream job to work at Intel.
[00:10:43] And obviously it didn't go the way he wanted it to, but clearly very proud of the time that he spent as CEO.
[00:10:49] But the board hasn't said that much about what their view of the situation is.
[00:10:54] My understanding is that they simply lost confidence in Gelsinger and they gave him the option to either retire or be removed.
[00:11:02] And he chose to retire abruptly.
[00:11:04] So what does Intel's success or failure mean for the American chipmaking industry?
[00:11:10] It's huge for the U.S. semiconductor industry broadly because Intel really is the only American chipmaker that is close to being capable of manufacturing chips at the cutting edge.
[00:11:23] Cutting edge chip manufacturing largely takes place in South Korea and Taiwan.
[00:11:27] Although, of course, Samsung and TSMC are building factories in the U.S.
[00:11:31] They're not U.S. companies.
[00:11:33] So for the U.S. and the U.S. government, Intel is kind of a point of national pride and national security importance.
[00:11:41] And Intel is getting a lot of money actually to help it build new chip plants under the 2022 Chips Act.
[00:11:48] It's getting nearly $8 billion for that.
[00:11:50] If Intel can't manufacture chips at the cutting edge, then the U.S. loses a lot of sort of prestige or capability in what has become an extremely critical industry and a critical realm of investment from a government perspective.
[00:12:09] That was our reporter Asa Fitch.
[00:12:11] And that's it for Tech News Briefing.
[00:12:14] Today's show was produced by Julie Chegg with supervising producers Melanie Roy and Catherine Millsup.
[00:12:19] I'm Danny Lewis for The Wall Street Journal.
[00:12:21] We'll be back this afternoon with TNB Tech Minute.
[00:12:24] Thanks for listening.