Phew, That Was A Week

A reminder for new readers. That Was The Week collects the best writing on critical issues in tech, startups, and venture capital. I select the articles because they are of interest. The selections often include things I disagree with. The articles are only snippets. Click on the headline to go to the original. I express my point of view in the editorial and the weekly video below.

This Week’s Video and Podcast:

Content this week from @Om, @Isabellesarraf, @pmarca, @elonmusk, @lexfridman, @jessicalessin, @ajkeen, @nickgrossman, @krishnanrohit, @kyle_l_wiggers, @demishassabis, @levie, @databricks, @@madewithmosaic, @mustafasuleymn, @inflectionAI


Editorial: Phew, That Was A Week

Essays of the Week

Why Apple’s Vision Pro Will Change Movie-Watching

Spotify’s Video-Podcast Surge

Linda Yaccarino’s vision for Twitter 2.0 emerges

‘Pressures Remain’: Coatue Prepares Tech Founders for the Road Ahead

Three Reasons to Look Past the 40% Drop in VC Funding

Big News Publishers Look to Team Up to Address Impact of AI

‎Matt Higgins on How the Publishing Industry Will Feel the Coming AI Storm

Amazon, Friction, and the FTC

Video of the Week

Marc Andreessen on what makes Elon Musk special | Lex Fridman

AI of the Week

AI + Crypto: Best and Worst Cases

The US Senate Wants to Rein In AI. Good Luck With That

Building god

Inflection lands $1.3B investment to build more ‘personal’ AI

Snowflake and Databricks are putting the data stored in their services to work

Google DeepMind’s CEO Says Its Next Algorithm Will Eclipse ChatGPT


News Of the Week

Cash Is Drying Up in the Late-Stage VC Market

ThoughtSpot acquires Mode Analytics, a BI platform, for $200M in cash and stock

Startup of the Week

Why Lightspeed is Leading Redpanda’s $100 Million Series C

Tweet of the Week

Aaron Levie – Databricks Buys Generative AI Startup MosaicML For $1.3B

Editorial: Phew, That Was A Week

Last week we had no AI section due to slow news. This week, wow!

DataBricks bought for $1.3bn. Inflection, developer of the Pi AI chatbot (try it, it’s good, if a little too friendly) raised $1.3bn, MidJourney released version 5.2 of its platform and, DeepMind stated that its next algorithm (strange word for this tech) will eclipse OpenAI’s ChatGPT.

Billion-dollar acquisitions are rare. Databricks is a private company valued at around $38 billion, so it paid 3.5% of its value for Why? Because data wants to be queried and data users would like to avoid writing code. It is a smart acquisition.

Any large company facing business customers or consumers will need to buy or develop AI. Early movers will lead the way.

But the real focus this week is on the media business, broadly defined. Om Malik is the top Essay of the Week because he boldly goes where nobody has gone before and predicts that the Apple Vision Pro will change movie watching. Read his story, it’s great. Not only is he right, so look out the theatres, but I will go one further and predict it will own the live sports experience within a few years also. There are very few things that would make me want to wear a face covering for two or more hours, but a great movie or a 3D seat at a live sport, sitting in a virtual stadium, is one of them. And if it’s Manchester United I will do it every game, 60 times a season or so.

Ai is going to have a huge impact on the media. Big publishers are already trying to slow it down with accusations of copyright infringement.

Barry Diller, accusing AI companies, said this week:

“Firstly, our content is being harvested and scraped and otherwise ingested to train AI engines,” he said at a recent industry event. “Secondly, individual stories will be surfaced in specific searches. And, thirdly, our content will be synthesized and presented as distinct when it is actually an extracting of editorial essence.” 

Now, read that again, but imagine it is accusing a human being of the same. This human brazenly read a lot, listened to a lot, and watched a lot. She then has the audacity to form views that she could not have formed prior to those efforts. Every opinion or fact she utters has been harvested and ingested to train her brain. She spits out parts of her learning every time she speaks or writes. She sometimes refers to specific stories without crediting them, often because the source is not relevant in the context. Or she simply can’t remember the source. She presents herself as having original thoughts that are, in part, synthesized from her consumption of other people’s work. Oh, wait. That is every human being.

AI is more like a human than a thief. It reads, listens, learns, and forms words. It is not copying or plagiarising. It is learning from what came before. That is not infringement. indeed the very purpose of publishing is so others can consume and ingest and learn.

So on this issue, I have to tell Diller (and Congress) that they are smoking up the wrong tree here. Andrew Keen has a good interview with Matt Higgins on the challenge AI poses to the publishing industry.

And finally, a lot this week on venture capital. Take a look at Coatue’s slides from its East Meets West event. They are a good summary of the current state of startups and venture capital.

Oh, one more thing. Video of the Week with Marc Andreessen is wonderful.

Essays of the Week

Why Apple’s Vision Pro Will Change Movie-Watching

Om Malik

This weekend I did something I had not done in a long time — I went to the movies. I wanted to see the new Wes Anderson movie Asteroid City, and going across town was my only option. I took a long Uber ride and then sat through about 30 minutes of pointless commercials and trailers of movies that seem to be remakes or the second or third movie in a series. And then there were people looking for their seats in the dark and unable to find them — the movie theaters now work on minimal staff in the guise of being “contactless.”

I went with a friend, so we talked a little before the movie, but not much, as there were people around us — and I didn’t want anyone to silence me. And once the movie started, I was fixated on the screen and didn’t care if I was with someone or was in a social space. However, watching the movie was a solitary experience — as it has always been. The total cost of the evening — Uber ride and the cinema ticket — made me wonder: why couldn’t I pay a premium and watch the movie at home? 


Photo credit: The Life Picture Collection / Library of Congress

Every time I make that argument, people look at me as if I am some kind of an anti-social degenerate. Does no one else notice that our idea of what a social space is ever-changing? Remember drive-in movie theaters? They were all over the country, but now they exist in our imagination or as a nostalgia act. 

We went from drive-in theaters to indoor theaters to multiplexes. And that is a good arc to follow to understand where we are going. The driving forces of change in the movie experience have been the human desire for comfort and convenience. And above all, property values — each change in movie viewing maybe be driven by technology, but it has been about deriving maximum value from the land. 

Indoor theaters meant more shows, which meant more revenue per square foot. Multiplexes meant more screens which attracted more people, which meant more revenue per square foot. Smaller multiplexes inside shopping malls, too, are all about the moolah. As a property developer, you can make a lot more money selling apartments than you can from half-filled theaters.

I do admit movies at one time were a social experience. We all used to have a shared moment when cinema was the social epicenter of the community. We enjoyed coming together to jointly participate in emotions inspired by the movie — laughter, horror, or simply the shock. With the death of rom-com and R-rated comedies, we can’t experience collective (humor or) horror. Today, all we are left with are big-budget sequels of superhero films that are as satisfying as a super-sized fast food meal. 

Times have changed, and so has our approach to socialization. Now we can’t even have a conversation without looking up from our computer screens. With the rise of social media and digital platforms, our interactions and the movie-watching experience have evolved. On-demand on our screens is the future, no matter how much traditionalists want to fight that.….

Spotify’s Video-Podcast Surge

By Isabelle Sarraf

Alex Cooper, host of Call Her Daddy, at Spotify’s Art of the Interview Session on June 20 in Cannes, France. Photo by Getty.

June 28, 2023 2:23 PM PDT

For the past three years, Spotify has been trying to beef up its video podcasts as it tries to keep up with the TikTok-driven explosion in short-form video. The efforts are starting to show some progress. The music streaming service now hosts more than 100,000 podcast shows that publish video episodes, a 40% increase from March though still a sliver of its 5 million total shows, according to new data from Spotify.

Reflecting its interest in video, the media streaming site has been inking deals with creators that highlight their videos on the app. Starting earlier this month, Emma Chamberlain’s “Anything Goes” podcast has been distributed across several platforms as an audio-only show, but its video podcasts are exclusive to Spotify. The company also recently tapped YouTuber Markiplier to exclusively host the video episodes of his podcast on its platform. 

For less well-known creators, such as the duo behind “Friends From Work,” who chat about the Marvel Cinematic Universe, Spotify has promoted a video version of their podcast, which they distribute only on the platform, co-host Kyle Schonewill told me.

Spotify’s push into video podcasts reflects TikTok’s influence over how people are consuming content. As short-form video has taken off on the app and rivals Instagramand YouTube, creators and media publishers have moved to publish more of their content as video that fans may repurpose and share as short-form video. 

For instance, one-minute video clips of Alex Cooper’s “Call Her Daddy” podcast, which is a Spotify exclusive for audio and video, frequently go viral on TikTok—where they’re posted by listeners, not just Cooper herself.

The focus on video podcasts follow an overhaul of Spotify’s podcasting operations. Earlier this month, Spotify cut 200 jobs, or about 2% of its total staff, mostly from its podcasting unit. The company said it was reorganizing its podcasting business to work with more top creators. 

“There really is not enough mobility in the top tier of creators. In podcasts, you have the top, let’s say 2,000 creators, that get the lion’s share of audience,” Julie McNamara, Spotify’s head of global podcast studios, told Kaya last week at the Cannes Lions advertising conference in France. “Something that we’re working on is both that sort of combination of algorithm and editorial to really feature where there’s continuous engagement.”

Some podcasters say it’s still hard for fans to find their work on Spotify, even with video. “Sometimes we see some of our longform videos pop off in views [on YouTube] because people come in via the [YouTube] Shorts,” said Eric Wei, co-CEO and co-founder of Karat, which offers credit cards and financial services for creators and co-hosts a video podcast where he interviews creators. “Spotify is trying to do this too, they’re just behind.”

Wei said his podcasts perform more than 30 times better on YouTube than on Spotify in terms of viewership and subscriber count, but Spotify is better for retaining listeners. After a YouTube Short clip of an older video podcast received over a million views, he said views of the original long-form video on YouTube jumped from 2,000 to 20,000 views.

Large media companies like Bloomberg as well as creator-focused startups have taken notice, as my colleague Alex reported Tuesday. London-based Veed, a startup in our Creator Economy Database, offers video editing tools popular with podcasters because of its free tool to convert videos to audio. Others, like creator monetization startup Kajabi, allow creators to repurpose longform videos into short-form clips for social media.

The Takeaway: Given how much video is dominating people’s attention, Spotify needed to make big moves in video. But it’s still not a platform where video clips can go viral—unless they get passed around on TikTok—a key way creators find new fans.

Linda Yaccarino’s vision for Twitter 2.0 emerges

CEO hits ground running with plans for new advertisement offering while telling staff to ready for ‘hand-to-hand combat’

Elon Musk and Linda Yaccarino, at a marketing conference in April, shortly before the Twitter owner asked her to come on board as the company’s chief executive © AP/Getty Images

Hannah Murphy in San Francisco

Twitter’s new chief executive, Linda Yaccarino, is preparing a series of measures to bring back advertisers who had abandoned the platform under Elon Musk’s ownership, including introducing a video ads service, wooing more celebrities and raising headcount. 

The former NBCUniversal advertising head, who started as chief executive on June 5, is seeking to launch full-screen, sound-on video ads that would be shown to users scrolling through Twitter’s new short-video feed, according to three people familiar with the situation. 

Yaccarino also plans to meet media partners, publishers and talent agencies in a bid to bring celebrities, political figures and other content creators to the platform.

The hope is that with content from high-profile names, Twitter can sell more advertising, and also facilitate sponsorship and brand deals between advertisers and creators.

In a meeting earlier this month with Twitter’s global sales team, Yaccarino told staffers the company was going to have to work hard to win back the trust of advertisers, according to one person with knowledge of the matter.

She said they would need to deploy “hand-to-hand combat” — persuading big brands in person rather than from behind their desks — as part of the effort, the person said.

“I do believe that they’re going to have superior ad product instantly [under Yaccarino],” said Lou Paskalis, an advertising veteran and chief executive of AJL Advisory. “The issue is content moderation. And it exists in two dimensions. For everybody else. And for Elon.”

“The hope is Linda will create a buffer between what he does, and a brand-safe environment on Twitter,” he added.

This account of Yaccarino’s vision for the platform is based on interviews with Twitter staffers, people familiar with her thinking and brands, advertisers and ad agencies. Twitter declined to comment.

The plans are designed to attract more ad dollars as Twitter seeks to reverse its fortunes under the direction of Yaccarino, an advertising heavyweight known for her industry relationships.

‘Pressures Remain’: Coatue Prepares Tech Founders for the Road Ahead

By Jessica E. Lessin

June 29, 2023 11:26 AM PDT ·

Comment by Anand Shah

Earlier this month, I had the opportunity to attend a founder and investor conference hosted by Coatue, called East Meets West. While the conference, held in Montecito, Calif., was off the record, I asked the firm for permission to share one of my favorite parts of the event: the keynote its partners gave on the state of the finance and tech markets.

Coatue estimates the overall market capitalization of private tech companies worth more than $1 billion is around $5 trillion based on where these companies last raised money. But based on marking those investments to comparables today, the number would sit at half that: $2.5 trillion.

In a 46-page slide deck, investment firm Coatue lays out advice for startups confronting today’s vastly changed private tech market.

To be sure, the haircut is in some respects healthy. It puts those companies as a percentage of the market capitalization of companies in Coatue’s public markets coverage at around 8%—as opposed to 16% based on the last-round valuation.

But how do the private tech markets get through that comedown at a time when mergers and acquisitions is anemic, the initial public offering window is closed and investors can earn a 5% return for no risk?

I asked Coatue co-founder Thomas Laffont to elaborate in a follow-up conversation this week.

“We are not going to get through it by applying the same rules we used over the last 10 years,” he said. “There is no longer one-size-fits-all advice.”

A slide deck that accompanied Coatue’s presentation laid out the different paths Laffont and Coatue—backers of more than 200 companies including ByteDance, Chime and Databricks—advise companies to think about based on the state of their business. They range from getting IPO ready as soon as possible to looking for a partner to weather the storm with. Given its decades of experience in the public markets and a track record in private tech over the last decade, Coatue has an interesting perspective.

The deck also presents some other sobering facts: An index of 68 cloud software companies that were growing net new annual recurring revenue at 20% from 2017 to 2020—and then 50% during the pandemic—have slowed to negative 20% net new ARR growth this year. Ouch.

The deck presents mountains of data about the overall economy and what it means for the likelihood of a recession. It also offers some interesting analysis about how the big tech companies’ stock performance compares to the overall market and what we can learn from the deviations.

Lastly, it wouldn’t be a tech presentation without a heaping dose of artificial intelligence. Coatue has plenty of optimism, calling Nvidia’s blowout second-quarter forecast report the “breakout” moment of the next tech supercycle. But the firm also encourages founders to really think through what AI could mean for their business. Does it allow them to be more efficient? Reveal a potential product they hadn’t thought of yet?

Here are the slides

Final Emw 2023 Macro Keynote 062823
5.98MB ∙ PDF file




Jessica Lessin founded The Information in 2013 after reporting on Silicon Valley for eight years for the Wall Street Journal. She regularly writes about all things tech, media and the wild ride both industries are in for. She can be found on Twitter at @jessicalessin.

Three Reasons to Look Past the 40% Drop in VC Funding

By Kate Clark

June 29, 2023 12:23 PM PDT

To my surprise, the recent boom in artificial intelligence investments has done little to dig U.S. venture capital funding out of its hole.

VC funding decreased in the second quarter, which ends tomorrow, according to new data provided to The Information by Crunchbase. U.S. startups raised just $27.6 billion compared to $45.2 billion in the first quarter, representing a roughly 40% drop. Compared to the same period last year, it’s an even more substantial plunge of 55%.

In total, U.S. startups have raised $73 billion so far this year, indicating that 2023 fundraising won’t even come close to last year’s $215 billion raised, which itself saw a big drop from 2021, a record-setting year. There are reasons to be positive, though. 

For one, at least six new early-stage funds were announced this week alone. That should be welcome news for entrepreneurs, especially after we reported on the fundraising struggles for megafunds. Seed-stage firms Uncork Capital, Kindred Ventures, Correlation Ventures and Unshackled Ventures raised nearly $1 billion in aggregate. 

It was an incredibly busy week for M&A announcements too. Three major M&A deals, coupled with Mediterranean restaurant chain Cava’s successful initial public offering earlier this month, has given investors hope the exit floodgates could finally be reopening.

This week’s acquirers include Databricks, which on Monday said it would pay $1.3 billion for two-year-old AI startup MosaicML (though the value of that deal is likely much less since Databricks’ own valuation has fallen, my colleague Aaron reports). Thomson Reuters said it plans to buy AI legal startup Casetext for $650 million in cash. And Visa said it would pay $1 billion for fintech startup Pismo, also in cash. 

Maybe most importantly, the shame of raising money in a down round seems to have mostly dissipated. Startups can correct their valuations and get back to work without fear of the negative PR cycle that once came with these sorts of financings. That’s a good thing, because it means more deals are likely to get done. 

A large reason for the continued slowdown in funding has been price mismatch. Startups have wanted to maintain their boom-era valuations, while investors have insisted on a steep discount. 

But a reset is already in motion, even for Silicon Valley’s most cherished startups. Investors valued Stripe, for instance, at $50 billion in fundraising earlier this year, a 47% discount to its last private fundraising. 

In fact, Coatue Management anticipates the overall market capitalization of private tech companies worth more than $1 billion has likely fallen to about $2.5 trillion—or about half of their worth based on where they last raised money, according to the firm’s 46-slide presentation we published today. Slowly, both sides are finding a middle ground. 

The downturn isn’t over, but the second half of the year is poised to be far busier than the first half.

Big News Publishers Look to Team Up to Address Impact of AI

New York Times, Wall Street Journal parent News Corp and Dotdash Meredith owner IAC are among companies discussing forming a coalition

IAC Chairman Barry Diller says that publishers, in some cases, should ‘get immediately active and absolutely institute litigation.’ PHOTO: GLENN CHAPMAN/AGENCE FRANCE-PRESSE/GETTY IMAGES

By Alexandra Bruell

June 28, 2023 at 4:24 pm ET

Several large news and magazine publishers are discussing the formation of a new coalition to address the impact of artificial intelligence on the industry, according to people familiar with the matter. 

The possibility of such a group has been discussed among executives and lawyers at the New York Times NYT 0.03%increase; green up pointing triangle; Wall Street Journal parent News Corp NWSA 0.00%increase; green up pointing triangle; Vox Media; Condé Nast parent Advance; Politico and Insider owner Axel Springer; and Dotdash Meredith parent IAC, the people said.  

A specific agenda hasn’t been decided, and some publishers haven’t yet committed to participating, the people said. It is possible a coalition may not be formed, they said.

While publishers agree that they need to take steps to protect their business from AI’s rise, priorities at different companies often vary, the people said. Such differences could create a hurdle when it comes to setting an agenda in a coalition, they said. 

Collaboration among competitive large publishers is rare, and the talks are indicative of the existential threat generative AI technology represents both to the industry and society. 

Generative AI, a type of artificial intelligence that can create various types of content including text, images and audio, has become a buzzword since OpenAI’s ChatGPT was launched late last year. The chatbot—which can eloquently answer seemingly any question, but is sometimes spectacularly wrong—became an overnight global phenomenon, fueling concerns that AI could reshape many industries.

In recent months, publishing executives have begun examining the extent to which their content has been used to “train” AI tools such as ChatGPT, how they should be compensated and what their legal options are, the Journal previously reported.  

Of particular concern is that the tools provide information directly to users, eliminating the need to follow links to information sources, such as articles. 

Even before AI’s rise to prominence, publishers have sought to get compensated for tech platforms’ use of their content, but traditionally structured deals on their own. For instance, News Corp and the Times each have reached sizable deals with Alphabet’s Google, the Journal previously reported. 

News publishers are being represented as a group by News Media Alliance, which has been behind a push for legislation in the U.S. that would allow publishers under a certain size to collectively negotiate for compensation from tech giants such as Google for the use of their content.  

A handful of executives in publishing have been vocal about AI and its implications for the news industry.  

Matt Higgins on How the Publishing Industry Will Feel the Coming AI Storm

In Conversation with Andrew Keen on Keen On

Hosted by Andrew Keen, Keen On features conversations with some of the world’s leading thinkers and writers about the economic, political, and technological issues being discussed in the news, right now.

In this episode, Andrew talks to Matt Higgins, the author of Burn the Boats, about how the publishing industry and writers will be impacted by the coming AI storm.

Matt Higgins is a guest shark on ABC’s Shark Tank, an executive fellow teaching at Harvard Business School, and, through RSE Ventures, the private investment firm he cofounded, an investor in some of America’s most beloved brands. A high school dropout at age sixteen, Higgins ultimately received his law degree from Fordham University School of Law and became the youngest press secretary to the mayors in New York City history. He then helped lead the effort to rebuild the World Trade Center site before becoming an executive for the New York Jets and later vice chairman of the Miami Dolphins. Passionate about human rights, Higgins works on behalf of the Global Solidarity Fund in furtherance of Pope Francis’s mission to support refugees and migrants around the world.

Amazon, Friction, and the FTC

Posted on Monday, June 26, 2023

It was Friday morning, and I needed sunglasses — specifically the nerdy ones that fit on top of a pair of prescription glasses. I wasn’t sure where to buy them — my dad (and who else would know better) suggested Walmart — but Amazon had a few; the only problem is that I was leaving early Saturday morning on a fishing trip, and surely that wouldn’t be sufficient time for e-commerce!

In fact, it was more than enough: Amazon had delivery options of 12-4pm, 4-8pm, or 4-8am the next morning; four hours later I had extra sunglasses in hand (and Walmart, for the record, didn’t have any).

This wasn’t the first time I’d leveraged Amazon’s same-day delivery: I was shocked to even see that it was an option when I arrived back in the U.S. and needed an ethernet cable at 4am; it showed up at 9:30am. It is fairly new, though; from the Wall Street Journal earlier this year: Inc. is expanding ultrafast delivery options, a sign that it remains committed to pushing its logistics system for speed as it scales back plans in other areas. The tech giant is continuing to devote resources to facilities and services structured to deliver packages to customers in less than a day. The expansions are happening at a crucial point for Amazon, which faces competition for fast-delivery options while Chief Executive Officer Andy Jassy puts a renewed focus on profits.

A central part of Amazon’s ultrafast delivery strategy is its network of warehouses that the company calls same-day sites. The facilities are a fraction of the size of Amazon’s large fulfillment warehouses and are designed to prepare products for immediate delivery. In contrast, the larger Amazon warehouses typically rely on delivery stations closer to customers for the final stage of shipping.

Amazon has opened about 45 of the smaller sites since 2019 and could expand to at least 150 centers in the next several years, according to MWPVL International Inc., which tracks Amazon warehouse operations. The sites have primarily opened near large cities and deliver the most popular 100,000 items in Amazon’s catalog, MWPVL said. New locations recently opened in Los Angeles, San Francisco and Phoenix, according to Amazon, which declined to provide information on how many of the same-day sites it has.

The reason to bring this program up now is to provide some personal context about the FTC’s latest lawsuit, this time against Amazon. Again from the Wall Street Journal:

The Federal Trade Commission sued on Wednesday, alleging the retail giant worked for years to enroll consumers without consent into Amazon Prime and made it difficult to cancel their subscriptions to the program. The FTC’s complaint, filed in federal court in Seattle, alleged that Amazon has duped millions of consumers into enrolling in Amazon Prime, a $139 annual subscription service with more than 200 million members worldwide that has helped Amazon become an integral part of many American households’ shopping habits.

“Amazon tricked and trapped people into recurring subscriptions without their consent, not only frustrating users but also costing them significant money,” FTC Chair Lina Khan said. The complaint, which is partially redacted, is the culmination of an investigation that began in March 2021. The FTC, a federal agency tasked with enforcing antitrust laws and consumer protection laws, seeks monetary civil penalties without providing a dollar amount.

I started with my own anecdote to explain why I am not personally familiar with the FTC’s complaints about the ease of signing up for Prime and the difficulty of cancelling: I haven’t had even a thought of going through either process for years. Indeed, even though I only live in the U.S. for a part of the year Prime is still worth it (and you get international shipping considerations as well).

This, to my mind, is the chief reason why this complaint rubs me the wrong way: even if there is validity to the FTC’s complaints (more on this in a moment), the overall thrust of the Prime value proposition seems overwhelmingly positive for consumers; surely there are plenty of other products and subscriptions that aren’t just bad for consumers on the edges but also in their overall value proposition and reason for existing.

Dark Patterns

The FTC makes two primary allegations in its complaint; the first is about the use of “dark patterns” to sign up for Prime:

Video of the Week

Marc Andreessen on what makes Elon Musk special | Lex Fridman

AI of the Week

AI + Crypto: Best and Worst Cases

June 28th, 2023by Nick Grossman

I think of AI and crypto as two very different, but very much related, elements of society moving from the industrial age to the digital age.

At USV, we (along with lots of others over the years) have used the Carlota Perez framework, which studies how techno-economic paradigms unfold over eras. Ben Thompson has a good summary of her ideas here. But the basic pattern looks like this:

It feels to me like we are somewhere in phase 3 or 4 of the “age of information & telecommunications” which Perez defines as starting in 1971. As much as it feels like information technology and the internet are fully woven into our daily lives, I don’t think it’s true that we’ve fully crossed over into a “digitally native” society, which is fully transformed into the new paradigm.

AI and crypto are both big missing pieces in the transition. AI is digitally-native knowledge, and crypto is digitally-native “proof”. The two can and will work together in many ways over time.

Major transitions are powerful and scary, and so are both AI and crypto. I have been struggling to calibrate between what I view as two poles in thinking about them together, kind of a best case hope and worst case fear.

Best Case:

AI finally unlocks knowledge from data. For decades we’ve been producing data (especially in digitized industries like media, finance and software), but making sense of it has been near impossible. AI systems solve the job of integrating, synthesizing and interpreting all of the data we have. AI accelerates the development of software systems, and makes it easier to digitize more industries and make them vastly more productive and efficient. Large Language Models, having turned human language into a programming language / API, make interacting with software and information as easy as typing or speaking, and as a result, we use software for infinitely more things, and get infinitely more value out of any data we produce. The pace of progress across everything (health, learning, climate, etc) increases exponentially.

At the same time, AI introduces new problems. First, a fundamental trust problem: it becomes difficult to tell what is real, what is fake, who said what, and who did what. And second, AI compounds the market power problem in the tech industry, where the large companies with the most data + compute + capital + distribution can generate insurmountable advantages.

Crypto (e.g., blockchain networks, web3, etc) addresses both of the issues introduced by AI. First on trust, crypto becomes the “real” yin to AI’s “fake” yang, as blockchain records and digital signatures become ground truth for everything digital: assets, transactions, media, etc. Anything digital that must be trusted (including the software we run and rely on) will need to be grounded in the best source of digital trust we have: crypto network security and unalterable digital histories. Crypto also addresses the tech consolidation issue by spreading compute across companies, individuals and geographies, and also by providing “open” alternatives to the big tech app store and online identity monopolies. (not related to AI, but not to be forgotten: crypto also finishes the job of upgrading the financial system)

Something like the above is my hope, and is the future we’re investing towards at USV.

Worst Case:

AI gets quickly beyond human control, pursuing its own goals (aka the terminator scenario). Concerns about job loss are quickly replaced with concerns about extinction. Everyone wonders how to “turn it off”.

The US Senate Wants to Rein In AI. Good Luck With That

With a poor track record on tech regulation, do lawmakers stand a chance?


AI IS DEFINING the future, even as many US senators struggle to understand it in the present.

“It would have been better if it had been held in a room where the acoustics were better,” Senator Chuck Grassley, an Iowa Republican, says of a much-anticipated—if overdue—All-Senators AI briefing orchestrated by Senate Majority Leader Chuck Schumer earlier this month.

The shoddy acoustics of the first of three closed-door meetings—kept private to insulate senators from electoral pressure to perform before cameras—were far from Grassley’s biggest complaint. “I would say that the next [one] will be more valuable, because this was a very general overview,” he says.

As AI expands its foothold across industries, households, and legislative bodies—including amongst some at the Capitol itself—Congress is under pressure to act quickly, even though many lawmakers still don’t know what they’re being asked to regulate. While Schumer, the White House, and industry leaders are spotlighting the revolutionary power of artificial intelligence, it’s still unclear if this hyper-dysfunctional Congress—currently consumed by the 2024 election cycle—can address AI before it remakes our world in its generative image.

For now, AI seems to be the least partisan issue in Washington, even as today’s bipartisan optimism is coupled with bicameral fear. This otherwise divided Congress is tuned in to AI—from Nafta flashbacks as some imagine AI upending today’s already upended job market to persistent Cold War fears now trained on AI’s potential to launch nuclear strikes. And that’s to say nothing of the electoral threat posed by generative AI and increasingly sophisticated deepfakes. These dauntingly high stakes may explain the Senate’s collective shrug after Schumer’s first big, closed-door AI reveal.

“There wasn’t much there that I hadn’t heard, and that’s a pretty low bar,” says Senator John Hickenlooper, a Colorado Democrat. “I wish it was more substantive.”

Senators inside the room when the doors closed say Massachusetts Institute of Technology professor Antonio Torralba was informative, especially when answering basic—yet seismic—questions, like how does AI learn? While the briefings are secret, earlier this week Schumer delivered a highly publicized AI address in which he laid out his shiny new SAFE Innovation Framework for AI Policy at the Center for Strategic & International Studies.

“In many ways, we’re starting from scratch, but I believe Congress is up to the challenge,” Schumer told the crowd. “AI moves so quickly and changes at near-exponential speed, and there’s such little legislative history on this issue, so a new process is called for.”

As a once-distant future rapidly becomes our present, the overarching question is: Can the current Congress learn fast enough to adapt?

For what is likely this briefest of windows, many US senators are gathering before a vetted few AI intellectuals, unified by their earnest search for answers. The thing is, some of those answers may not exist, even as humanity—and the Senate—journeys into the algorithmic unknown.

Lobbyists on the Sidelines—for Now

Over the Silicon Valley tech boom of the past two decades, Congress has held many hearings—some less embarrassing than others—but when it comes to actual regulations, lawmakers have been mostly hands-off.

Schumer is now vowing to do what he and his colleagues have failed to do thus far: regulate the tech titans who have spent tens of millions of dollars all but painting the Senate in their trademarked hues. That pressure is yet another reason the doors are closed for these All-Senators AI briefings.

Building god

Or at least Talos


JUN 28, 2023

Recently I started writing an essay about AI development. Somehow it got away from me and has become a small book (40k words or so). I’m still figuring out what to do about it. But one part of it is topical, related to building a self-improving, reflective, Agent which is good enough to run scientific experiments and how we might build one. Here is that section. Would love thoughts or feedback!

Oh and warning: this one’s a little wonky.

Blueprint for an agent that could actually do complex tasks end to end autonomously

Intelligence unbound

Intelligence is important1

.You might even claim it drove much of human achievement in the past. In the physical world, when we figured out how to dig out rocks from under the earth, to use coal and iron and steel, we had an industrial revolution put together with human ingenuity.

Even outside the physical realm, Jay Gould in the gilded age made his fortune because he could communicate with his business partner through telegraph during the time of the American Civil War. Reuter’s fortune was made during the Franco-Prussian War as he used his news agency to provide news and information to both sides of the conflict, which made him a valuable asset to both governments.

Intelligence helped collect, collate, condense, integrate the information about the world, and to use them to craft new futures.

That’s both the opportunity and the fear about AI. That it’s unbridled intelligence.

When people talk about AI eventually taking over the world’s GDP or replacing humans, everyone has an idea of what they’re able to do. For some it’s like HAL, from 2001 A Space Odyssey, a being which has a mind of its own and is stubborn and refuses to let anyone tamper with its instructions.

For some it’s Terminator, without Arnold’s cool glasses or his one-liners you read in the subtitles and laugh out loud. For some it’s like Iron Man, where you get to say “Jarvis, build me a time machine” and Jarvis says “Yes, sir, right away” in an impeccable British accent and then does exactly that.

A lot of these essentially include imagining that the AI can, basically, combine almost anything any human can and almost anything any computer can. Like it can think through new situations and make silly jokes like humans, and it can calculate the product of two primes faster than a supercomputer, and most things in the middle. Is this realistic?

So there are a bunch of things people wonder about. Like, is there a particular size of the training when the network simply just wakes up and goes “hello”? We can make it say so of course, after all Blake Lemoine managed to do so in Google, and Microsoft publicly managed to do it with Sydney, when it apparently wanted to seduce a man away from his wife2.


So if we want to define what AGI is, we might have to keep being a bit vague. The ability to harness “intelligence”, that utterly nebulous concept that in Wittgensteinian terms can only be defined through usage, and to put it to work much the same way we turned the power of horses into horsepower.

The core capability question

Instead the question I like is if we were to drop a machine off in a random spot on Earth, whether that’s in the trading floor of Goldman Sachs or the remote jungles of Congo, will it be able to understand where it is and somehow take care of itself?

Inflection lands $1.3B investment to build more ‘personal’ AI

Kyle Wiggers @kyle_l_wiggers / 7:00 AM PDT•June 29, 2023

Image Credits: NicoElNino / Getty Images

There’s still plenty of cash to go around in the generative AI space, apparently.

As first reported by Forbes, Inflection AI, an AI startup aiming to create “personal AI for everyone,” has closed a $1.3 billion funding round led by Microsoft, Reid Hoffman, Bill Gates, Eric Schmidt and new investor Nvidia. A source familiar with the matter tells TechCrunch the tranche, which brings the company’s total raised to $1.525 billion, values Inflection at $4 billion.

CEO Mustafa Suleyman, who previously co-founded the Google-owned AI lab DeepMind, says that the new capital will support Inflection’s work to build and design its first product, an AI-powered assistant called Pi.

“Personal AI is going to be the most transformational tool of our lifetimes. This is truly an inflection point,” Suleyman said in a canned statement. “We’re excited to collaborate with Nvidia, Microsoft, and CoreWeave as well as Eric, Bill and many others to bring this vision to life.”

Palo Alto, California-based Inflection, which has a small team of around 35 employees, has kept a relatively low profile to date, granting few interviews to the media. But in May, Inflection launched the aforementioned Pi, which is designed to provide knowledge based on a person’s interests and needs. Available to test via a messaging app or online, Pi’s intended to be a “kind” and “supportive” companion, Inflection says — offering “friendly” advice and info in a “natural, flowing” style.

Inflection’s personal AI assistant, which offers “friendly” advice. Image Credits: Inflection

Inflection recently peeled back the curtains on Inflection-1, the AI model powering Pi, asserting that it’s competitive or superior with other models in its tier — namely OpenAI’s GPT-3.5 and Google’s PaLM-540B. According to results from the company, Inflection-1 indeed performs well on various measures, like middle- and high school-level exam tasks and “common sense” benchmarks. But it falls behind on coding, where GPT-3.5 beats it handily and, for comparison, OpenAI’s GPT-4 smokes the competition.

Snowflake and Databricks are putting the data stored in their services to work

Both companies are helping customers build generative AI apps

Ron Miller, Alex Wilhelm/ 9:30 AM PDT•June 28, 2023

Image Credits: Yuichiro Chino / Getty Images

Snowflake and Databricks are surely similar companies. While each positions itself a bit differently, both provide data storage, processing and governance in a cloud context. Both are holding customer conferences this week, and both are looking for ways to help customers build generative AI and other intelligent applications on top of the data stored in these platforms.

If that wasn’t clear before, it became even more apparent this week when Databricks announced it was acquiring MosaicML for a cool $1.3 billion. That’s a lot of money for a startup, even a well-capitalized one like Databricks. The move came weeks after the company announced it was releasing Dolly, an open source LLM, and another acquisition in AI governance tool Okera.

Snowflake announced last month that it was buying Neeva, giving it a search tool and some high-end AI engineering talent. The company also bought Streamlit last year, which lets companies build applications from the data stored in Snowflake, and on Wednesday, it announced a new container service and partnership with Nvidia, giving customers a way to build generative AI applications and run them on Nvidia GPUs.

All of these moves (and others) are designed with one thing in mind: to use the data stored in these services as fuel for machine learning models, especially large language models. Both companies want to help customers take advantage of all this data stored on their platforms.

Nvidia’s VP of enterprise computing, Manuvir Das, speaking in the context of Wednesday’s partnership announcement with Snowflake, sees the move toward more practical use of the data as a logical progression for Snowflake.

“The fact that Snowflake is now moving in this next step where they’re saying, OK, not only can you keep your data here and do sort of the obvious data processing things on it, but this is the place where you can build all the applications that drive your company because your data is right here. That’s a very powerful thing,” Das told TechCrunch+.

Similarly, Databricks is increasingly seeing itself as a place where you can not only store data and do the various data tasks associated with that, but you can also be part of a whole data stack, where you build applications on top.

This week’s MosaicML acquisition was part of this broader strategy to put the data to work in an AI context, said Ray Wang, founder and principal analyst at Constellation Research. That’s something that was hard for Databricks to do, even with Dolly.

“The AI angle is all about making it easy to acquire, manage, train and deploy LLMs with ease,” Wang said.

Both companies are clearly moving hard toward AI through acquisitions, partnerships and product development. But what does that mean from a potential revenue perspective for the future of these companies, one of which is already public and one that surely will be there eventually?

Google DeepMind’s CEO Says Its Next Algorithm Will Eclipse ChatGPT

Demis Hassabis says the company is working on a system called Gemini that will tap techniques that helped AlphaGo defeat a Go champion in 2016.


IN 2016, AN artificial intelligence program called AlphaGo from Google’s DeepMind AI lab made history by defeating a champion player of the board game Go. Now Demis Hassabis, DeepMind’s cofounder and CEO, says his engineers are using techniques from AlphaGo to make an AI system dubbed Gemini that will be more capable than that behind OpenAI’s ChatGPT.

DeepMind’s Gemini, which is still in development, is a large language model that works with text and is similar in nature to GPT-4, which powers ChatGPT. But Hassabis says his team will combine that technology with techniques used in AlphaGo, aiming to give the system new capabilities such as planning or the ability to solve problems.

“At a high level you can think of Gemini as combining some of the strengths of AlphaGo-type systems with the amazing language capabilities of the large models,” Hassabis says. “We also have some new innovations that are going to be pretty interesting.” Gemini was first teased at Google’s developer conference last month, when the company announced a raft of new AI projects.

AlphaGo was based on a technique DeepMind has pioneered called reinforcement learning, in which software learns to take on tough problems that require choosing what actions to take like in Go or video games by making repeated attempts and receiving feedback on its performance. It also used a method called tree search to explore and remember possible moves on the board. The next big leap for language models may involve them performing more tasks on the internet and on computers.

Gemini is still in development, a process that will take a number of months, Hassabis says. It could cost tens or hundreds of millions of dollars. Sam Altman, OpenAI CEO, said in April that creating GPT-4 cost more than $100 million.

Playing Catch-Up

When Gemini is complete it could play a major role in Google’s response to the competitive threat posed by ChatGPT and other generative AI technology. The search company pioneered many techniques that enabled the recent torrent of new AI ideas but chose to develop and deploy products based on them cautiously.

Since ChatGPT’s debut Google has rushed out its own chatbot, Bard, and put generative AI into its search engine and many other products. To juice up AI research the company in April combined Hassabis’ unit DeepMind with Google’s primary AI lab, Brain, to create Google DeepMind. Hassabis says the new team will bring together two powerhouses that have been foundational to the recent AI progress. “If you look at where we are in AI, I would argue that 80 or 90 percent of the innovations come from one or the other,” Hassabis says. “There are brilliant things that have been done by both organizations over the last decade.”

Hassabis has experience with navigating AI gold rushes that roil tech giants—although last time around he himself sparked the frenzy.


Jun 26, 2023 by Dunja Djudjic 2 Comments

Midjourney has just launched its new version, 5.2, and it’s a game-changer! There are several new features in the new update, but the hottest one is the “Zoom Out” feature, which creates a wider view of your image and automatically fills in the rest. I played with it to test it out, and while there are still some hiccups – it’s pretty impressive overall. So, let’s dive right in and see how it works.


Midjourney’s new features in v.5.2

New Aesthetic System

New “High Variation Mode”

New “/shorten” command

New “Zoom Out” Feature

Testing Midjourney’s “Zoom Out” feature

Final thoughts


Before we dive deeper into the Zoom Out feature, let me briefly introduce you to all the new additions to Midjourney 5.2. But if you’re really curious, scroll down to the “New ‘Zoom Out’ Feature” heading. 🙂


Midjourney has improved aesthetics and gives you sharper, more detailed images. From what I saw in my quick test, they are indeed a bit sharper, and I am seeing more details in textures. Coherence and text understanding has been slightly improved, and Midjourney also claims that diversity has been increased! However, they note that sometimes you still may need to roll more than once to get what you want.

The –stylize command has also been fixed to have a strong effect on the amount of stylization applied to your image. It goes from –stylize 0 to –stylize 1000, and the default value is –stylize 100.


Another new addition is turned on by default and makes all variation jobs more… well, varied. You’ll find it under settings: type /settings and click a different variation mode. You’ll also be able to choose which strength of variation you want underneath all upscales.

News Of the Week

Cash Is Drying Up in the Late-Stage VC Market

Companies seeking Series C and D round funding are likely to experience a down round in the second half of 2023, according to PitchBook.

June 27, 2023

By Hannah Zhang

Getty images

The second half of the year is looking bleak for companies backed by venture capital firms, especially for those at later stages.

The companies that are most starved for capital are those that are currently in Series C and D round fundraising. These are most likely to raise at lower valuations compared to their previous funding rounds, according to PitchBook’s latest VC outlook. For such companies, which are defined as late-stage companies by PitchBook, the current demand for capital exceeds the available supply by approximately 2.8 times. In the second quarter, 12.3 percent of late-stage companies that raised funds experienced a down round, up from 8.5 percent in the first quarter.

The fundraising landscape also appears challenging for companies at Series E or later, known as the venture-growth stage. These companies face a demand for capital that exceeds the available supply by 1.3 times. In the second quarter, 34.3 percent of companies in the venture growth stage experienced a down round, up from 31.9 percent in the previous quarter.

“Venture-growth companies, which missed their IPO window in 2020 and 2021 and likely raised funds at soaring valuations, found themselves particularly exposed to the market downturn,” according to the report. “Their failure to initiate an IPO could signify a lack of the fundamental strength necessary to attract public markets.”

PitchBook estimates that deal value for venture-growth companies will fall below $50 billion in the U.S. this year. In both 2021 and 2022, deal value for venture-growth companies surpassed the $50 billion threshold. “While this hurdle miss might seem unimportant, there have never been more companies in need of venture-growth or pre-IPO capital,” the report said. “There are now more than 2,000 venture-growth companies. Outright failure of venture-growth companies is an uncommon event, but for those companies that may need to accept a low exit value because of current conditions and the lack of venture-growth capital, it may feel like one.”

Things look a lot better for companies at the seed stage. According to PitchBook, the median valuation of seed-stage companies reached a record high of $12.9 million in the first quarter. The robust valuation of seed-stage companies is due in part to the rising interest in early-stage deals from nontraditional investors, including hedge funds, corporate venture capital funds, and mutual funds.

“Not only has the more formulaic nature of dealmaking at the seed stage prevented the valuation compressions seen in later stages, but the elevated participation of corporate VCs and larger investors at this stage has helped boost deal metrics,” the report said. It added that the deals sizes of seed-stage companies have been climbing past the broader market’s median as the larger investors put their dry powder to work.

ThoughtSpot acquires Mode Analytics, a BI platform, for $200M in cash and stock

Kyle Wiggers @kyle_l_wiggers / 9:00 AM PDT•June 26, 2023

Image Credits: Andrey Suslov / Getty Images

ThoughtSpot, an AI-powered analytics platform last valued at $4.5 billion, today announced that it’s entered into a definitive agreement to acquire Mode Analytics, a business intelligence startup, for $200 million in cash and stock.

Mode will become a wholly owned subsidiary of ThoughtSpot once the deal closes later this year, subject to customary closing conditions and the approval of Mode’s shareholders.

It’s ThoughtSpot’s fourth acquisition following (most recently) the company’s purchases of SQL-based analytics firm SeekWell in March 2021 and data integration company Diyotta in May of that same year. Enabling all the deals is ThoughtSpot’s massive war chest, which totaled over $663 million as of August 2019.

As for the Mode acquisition, ThoughtSpot CEO Sudheesh Nair says that it’ll bolster ThoughtSpot’s generative AI apps while doubling the company’s customer base and growing its annual recurring revenue to more than $150 million.

“With this acquisition, we’re giving both data teams and business users the tools they need to efficiently and quickly turn data into insights and those insights into actions,” Nair said in a press release issued this morning.

San Francisco-based Mode, which TechCrunch last covered in August 2020, was co-founded by Derek Steer, Benn Stancil and Josh Ferguson in 2013. All three previously worked at Yammer (they were early employees and stayed on after the Microsoft acquisition), where they were a part of a larger team building custom data analytics tools for the Yammer platform.

Startup of the Week

Why Lightspeed is Leading Redpanda’s $100 Million Series C

The Developer’s Journey in a Streaming World

Coffee was followed by a sushi lunch a couple of days later, and we dove deeper into the technical details of how he planned to deliver on his core insight: Kafka, the incumbent open-source streaming event platform, was built for a different time — a world of spinning disk drives, big iron data centers, and single-core CPUs. A world of Java. Fast forward to today: It’s a cloud-first world. Computation and data storage have been decoupled, and software optimized for modern environments and services can deliver massive improvements in speed, reliability, and cost efficiency. Redpanda was built from the ground up for this world, written in C++ with cloud primitives, massive scalability, and performance in mind.

Core to Alex’s vision was to not throw out the baby with the bathwater. Redpanda has been Kafka-API compatible from the beginning. By changing a single line of code, developers can easily switch to Redpanda and enjoy immediate benefits: higher performance, less infrastructure, a simpler back-end to manage, and much lower costs to run and maintain.

The final piece of the puzzle is the most obvious but essential: flexibility. Customers who want to can bring their own cloud (or on prem system) to run Redpanda, while others can choose to run their instance on Redpanda’s own dedicated, secure, isolated cloud, for maximum ease and flexibility. No matter how big or small a team or application, there’s a solution that fits.

Tweet of the Week

Databricks Buys Generative AI Startup MosaicML For $1.3B

Chris Metinko

June 26, 2023

Chris Metinko

Databricks, the data storage and management startup last valued at $38 billion, signed a definitive agreement to acquire OpenAI competitor MosaicML for $1.3 billion.

MosaicML, a San Francisco-based generative AI platform, had raised $64 million to date since launching in 2021, per reports. Its investors included DCVC, Lux Capital and Playground Global.

MosaicML allows customers to build generative AI tools using its own proprietary data — eliminating the need for users to incorporate their own data with OpenAI’s proprietary models that use public data.

“Every organization should be able to benefit from the AI revolution with more control over how their data is used. Databricks and MosaicML have an incredible opportunity to democratize AI,” said Ali Ghodsi, Databricks’ co-founder and CEO, in a release.

Getting bigger

San Francisco-based Databricks last raised in 2021, locking up a $1.6 billion Series H led by Morgan Stanley’s Counterpoint Global at a $38 billion valuation — making it one of the most valuable private companies in the world.

The deal for MosaicML could be another step toward the public market for the decacorn, as it looks to expand its portfolio of offerings and cash in on an exploding AI market.

While the Databricks deal is one of the first big M&A transactions in the AI space this year, the sector has seen a flood of funding dollars invested in it. 

In January, Microsoft invested $10 billion in OpenAI. In March, San Francisco-based Adept AIraised $350 million in a Series B at a reported post-money valuation of at least $1 billion. In May, Anthropic — a ChatGPT rival with its AI assistant Claude — raised a $450 million Series C that reportedly valued the company at $5 billion.

And that is only a handful of the funding deals the AI has seen since the beginning of the year.

Databricks just recently talked about surpassing a milestone of $1 billion in annual revenue and has talked about a potential IPO in the past. 

Founded in 2013, Databricks has raised more than $3.5 billion, per Crunchbase. Its investors include Andreessen Horowitz, NEA, T. Rowe Price, BlackRock and many others.

How Databricks CEO Justifies Paying $1.3 Billion for a Young AI Startup

By Aaron Holmes

Naveen Rao (left), CEO of MosaicML, and Ali Ghodsi (right), CEO of DataBricks. Photos by Getty.

June 28, 2023 3:31 PM PDT ·

Comments by Zen Zen, Bradford Harries, and 3 others

When enterprise software firm Databricks revealed on Monday it would pay $1.3 billion for a two-year-old artificial intelligence startup, MosaicML, the deal looked overpriced. Databricks is paying 65 times Mosaic’s $20 million in annual recurring revenue, a measure of customer commitments to pay for its software, according to Databricks CEO Ali Ghodsi.

In reality, the value of the deal is much less. Databricks is paying for Mosaic in stock at the same share price as Databricks’ last equity financing round, in 2021, which valued it at $38 billion, Ghodsi said. That period was the peak of startup valuations, so Databricks’ current valuation may be closer to half that. If Databricks’ share price were to be cut in half, the Mosaic deal would be worth closer to $650 million, or 32.5 times revenue. That’s well above the value put on most enterprise software deals outside AI, which lately have been done at less than 10 times next year’s revenue.

• The all-stock deal would be worth closer to $650 million based on today’s valuations
• The deal reflects the battle of open-source AI models versus closed ones from OpenAI
• Microsoft could resell Databricks’ AI-building tools alongside OpenAI software

The episode injects more optimism for a boom in AI startup acquisitions after a previously dormant period. ChatGPT creator OpenAI has prompted seemingly every software provider to place new bets on AI—especially the kind that can understand conversational commands—or risk being left behind. Snowflake, a Databricks rival, earlier this month acquired search startup Neeva, whose engineers had experience with large-language models that handle such commands, in a deal valued around $150 million, said a person briefed about the deal. And financial data firm Thomson Reuters this week bought Casetext, a 10-year-old firm that helps law firms automate document reviews, for $650 million in cash. Castext in March began using OpenAI software to boost its services.

Ghodsi said the price for Mosaic is “easy to justify” because of the startup’s revenue growth—its ARR was just $1 million in January—and the “pent-up demand” for custom AI models at large enterprises. Mosaic’s software helps businesses develop their own ChatGPT-like apps for automatically writing software code for internal apps or recommending ways to save money on cloud software. Mosaic’s customers are primarily firms that help businesses implement this type of AI.

“MosaicML has about three folks doing sales. Our [sales] organization is about 3,000 people. I believe the combination can accelerate our revenue significantly,” he said.

Databricks sells a cloud database and other software to help companies apply machine-learning models to their data. It wants to use the Mosaic acquisition to offer customers an easier way to customize large-language models, a type of machine-learning software that powers chatbots, than generalized software from OpenAI can provide.

Open Versus Closed

The acquisition is a sign of the battle developing between companies like OpenAI, Anthropic and Cohere that develop the largest LLMs, which are proprietary or closed source, and software providers like Databricks (as well as Meta Platforms) that hope businesses will want to do the legwork of training smaller, open-source LLMs on their own corporate data to get better performance and to prevent the data from leaking to OpenAI. OpenAI may hedge its bets, however: The Information has reported the company has been preparing to launch a new open-source LLM.

%d bloggers like this: