Tech giants and billion-dollar startups are duking it out over large language models in AI. But who will win?

Welcome back to LinkedIn News Tech Stack, which brings you news, insights and trends involving the founders, investors and companies on the cutting edge of emerging technology.

As always, let me know if you have ideas or feedback on something that could be a fit by sending me an InMail. Click subscribe to be notified of future editions and follow me on LinkedIn for other tech updates.

Zooming In

A deep-dive into one big theme or news story every week.?

OpenAI – the artificial intelligence startup behind ChatGPT – is the buzziest company in Silicon Valley at the moment, the frenzy around it only growing by the day.

OpenAI has been a pioneer in advancing the AI field of natural language processing by employing large language models (LLMs) at scale, in which AI is trained on gargantuan amounts of text to understand the nuances of writing and speech.

Cashing in on the momentum, last week Microsoft (LinkedIn’s parent company) invested $10 billion into the company . But it’s not alone.

谷歌 , Meta and a host of smaller startups including Adept , Anthropic , Cohere and Models Lab (formerly Stable Diffusion API) have all developed their own LLMs thanks to notable advances in computational power and mathematical techniques, making it possible to train AI on larger datasets than ever before. The startups among the group have collectively raised more than $1 billion in funding.

“We have seen step-function improvements in processing,” Colin Beirne , a partner at venture capital firm Two Sigma Ventures, who was featured on this week’s #VCWednesdays , told me. “That’s what we’re seeing with ChatGPT and other applications – they’re possible due to improvements in LLMs that are simply a product of training AI with large datasets.”

In other words, large language models are the building blocks on top of which vertical natural language applications like Jasper Ai (for AI-generated copywriting) or Contentsquare (digital customer analytics) are built for specific industry use cases. Without these LLMs, there would be no cool applications.

Who’s ahead of the curve when it comes to building these powerful LLMs? The big guns have an obvious advantage, given how expensive the computing power required to train these models is, experts say. Beirne called it a “classic problem” that requires a ton of money, and is therefore “accessible to only a few companies.”

“LLM models – because of the expense to create them and to run them – are, without question, an advantage for the largest players, the mega tech giants,” said Sameer Dholakia , partner at Bessemer Venture Partners .

Moreover, it’s a strategic necessity for the likes of Amazon and Microsoft that have thriving cloud businesses, Dholakia said. In Microsoft’s case, for example, the company would make money anytime a query is run on OpenAI since it runs on top of its Azure Cloud service. Other cloud providers could also sell artificial intelligence–based applications developed alongside their other products, he said.

While Google has an LLM called LaMDA , it’s keeping the bulk of its efforts under wraps for now. But the company has reportedly declared a "code red" to ship AI projects faster, as its search dominance has come under threat. Last week, it also posted a paper with samples showing the work of MusicLM, which is able to generate music from text prompts.

But don’t count the startups out yet. While they may trail the tech giants in resources, they make up for it with their models and talent, said experts.

In recent years, top researchers from OpenAI and Google DeepMind have left the companies to found or lead other startups in the field of large language models. Adept CEO David Luan was the former tech lead for Google’s work on large models, and a vice president of engineering at OpenAI prior to that.

In addition, a recent evaluation published by Stanford University’s Center for Research on Foundation Models, for example, found that Cohere and Anthropic’s LLMs were less likely to produce toxic responses than OpenAI’s GPT-3 – which has been called out for hallucinations.?

Ultimately, as both the startups and established players produce better models and bring computing costs further down, it may become a more free-market economy, said Wesley Chan , co-founder and managing partner at FPV Ventures .

“Like in the case of mobile, the underlying technology became secondary and apps are what became table stakes,” he said. “Once AI becomes commoditized, its use cases are what are going to matter more than the LLMs they used.”

Who do you think will come out on top? Let me know in the comments.

TechTok?

Catch up on the tech headlines you may have missed this week and what our members are saying about them on LinkedIn.

• Continuing our coverage of the AI world, ChatGPT is launching a paid subscription service. Parent company OpenAI announced that for $20 a month, subscribers will get an upgrade from the limited, free service provided during busy periods.?
• One of Twitter's biggest shareholders has slashed the value of its stock for the second time in a few months. Fidelity has written down Twitter’s stock by over 60% since Elon Musk acquired the company for $44 billion at the end of October.?
• The U.S. has stopped providing American suppliers licenses to export to Huawei , with the idea of potentially cutting the Chinese telecom giant’s access to U.S. tech altogether, various outlets reported this week . The U.S. has long considered Huawei a national security threat.
• Meta bucked the trend of downbeat tech earnings reports this week. The Facebook parent saw its stock soar in after-hours trading on Wednesday after reporting $32.2 billion in fourth-quarter revenue , about $600 million more than analysts had expected.
• While government data suggests that it’s still a worker's job market , an avalanche of layoff announcements in tech and beyond continue. In the past week, Okta and Pinterest announced that they will cut 300 jobs and 150 employees, respectively (about 5% of their total workforces).

Movers and Shakers

Here’s keeping tabs on key executives on the move and other big pivots in the tech industry. Please send me personnel moves within emerging tech.

• David Cahn , a general partner focused on AI and COO of venture capital at Coatue , is joining Sequoia Capital as a partner on its growth team, Axios reports . Instagram co-founders Kevin Systrom and Mike Krieger are teaming up on a new venture called Artifact, a machine learning-based news app, Platformer reports. Peloton Interactive has hired former Twitter chief marketing officer Leslie Berland as its CMO, as the beleaguered fitness-tech company mounts a comeback.? Docusign has hired former Atlassian marketing chief Robert Chatwani to be president and general manager of growth, and former Google executive Anwar Akram as COO.Real estate software company Built has named James Chen , most recently the CTO of Flexport , as its chief technology officer.?

This Week in AI

A new section where we bring you up-to-speed with the latest advancements from the world of AI.

• In addition to OpenAI releasing the first subscription version of ChatGPT, it’s also rolled out an official tool that detects ChatGPT content . So all you professors panicked about plagiarism and students using the AI generator to write papers: don’t fret!
• MIT researchers say they have developed an AI model that can detect future lung cancer risk based on CT scans.
• Companies are starting to capitalize on the investor frenzy around artificial intelligence.

ICYMI

Here are other top stories of the week from beyond LinkedIn in the broader world of tech.

• Instagram has entered its flop era, and its focus on advertising content as opposed to user-generated content is partly to blame, Chris Stokel-Walker writes for Business Insider .
• TikTok is facing immense scrutiny , so much so that CEO Shou Chew will testify before the U.S. Congress in March. As the Chinese-owned app awaits a government-approved security deal , states across the country have banned the app on government devices and on college WiFi networks, and members of Congress have introduced bills to ban the app entirely.

Thanks for reading. Please share Tech Stack if you like it! And if you have any news tips, find me on InMail .