Google used to be a company where experimentation was encouraged -- or at least it felt like that from the outside. Now it's hard to remember when Google last launched a new product that was an immediate hit. But with NotebookLM and its AI podcasts, Google finally scored an AI hit.

By now, you're probably familiar with NotebookLM and the AI podcasts it can generate. I remember seeing the demo page a few weeks ago (probably via Hacker News) and how surprised I was by how good those examples were. Yet, I'm also skeptical of every demo that seems too good to be true.

Now with this same functionality open to everyone, it's clear that Google got this right and I've actually been using the service quite a bit to prepare for meetings and big interviews. While doing my research, I bookmark a bunch of sources in NotebookLM, no matter whether those are Wikipedia pages, news articles, interviews on YouTube or scientific papers. Then I let it generate a podcast from those.

I'm a habitual reader and I've likely read (or at least skimmed) all of those sources before I bookmarked them in NotebookLM, but there is something about hearing the conversation between the two AI hosts that gives me a different perspective, or maybe puts a different emphasis on the sources. It's an easy 10-minute exercise and has been worth it every single time I've tried it.

While the podcasts are a nice gateway into NotebookLM, I think many pundits are forgetting that this is just one feature. A lot of the value I've been getting out of it is using its chat function that uses these same sources as the core of a Gemini/ChatGPT-like experience -- just like any similar chat tool that you've fed with a large amount of context -- and while it can go beyond those sources (but not the topics they are about), it'll always try to refer back to those documents. Ask it about the meaning of life, though, and it'll politely refuse.

In the open-source world, all eyes remain on the WordPress/WP Engine feud. A lot has been written about that already, so there's no need to rehash the details. In the end, it's always about ego and money.

Talking about money and open source: Broadcom's acquisition of VMware, and the way Broadcom has treated VMware's customers ever since, has become a boon for OpenStack, the massive open source project that originally set out to offer enterprises an on-prem version of the core AWS services. Over the last decade, OpenStack and the foundation that backed it, went through a well-chronicled hype cycle. I've been covering the project and talking to the leaders of what is now the OpenInfra Foundation since I first met them in 2014.

“I did not have ‘VMware sparks OpenStack resurgence’ on my 2024 bingo card,” OpenInfra Foundation Executive Director Jonathan Bryce told me when I interviewed him a few weeks ago. “It’s definitely been something that has driven just an incredible amount of interest."

Some old OpenStack vendors are now coming back into the ecosystem, and some that had put their services on ice, only serving their legacy customers, are doubling down on the platform again.

OpenStack is ready for the VMware refugees | TechCrunch

Broadcom’s acquisition of VMware has left a lot of customers uneasy (and with rising bills). For the longest time, VMware was the de facto standard for

TechCrunchFrederic Lardinois

Earlier this week, I talked to AWS CEO Matt Garman. We talked about a lot of things (generative AI being at the forefront, of course), but also open source.

Here is the key quote from my story: “I think our view is pretty straightforward,” Garman said when I asked him how he thinks of the relationship between AWS and open source going forward. “We love open source. We lean into open source. I think we try to take advantage of the open source community and be a huge contributor back to the open source community. I think that’s the whole point of open source — benefit from the community — and so that is the thing that we take seriously.”

You can read more about his thoughts on open source here.

AWS CEO Matt Garman on generative AI, open source, and closing services | TechCrunch

It was quite a surprise when Adam Selipsky stepped down as the CEO of Amazon’s AWS cloud computing unit. What was maybe just as much of a surprise was

TechCrunchFrederic Lardinois

I don't think anyone outside of Matt Mullenweg had the recent WordPress/WP Engine drama on their bingo card for this year. After a bit of early confusion, I think it's now clear that this is, in many ways, an extension of the open source discussions we've had in the enterprise world over the last few years. Mullenweg/Automattic is unhappy that WP Engine is profiting from the open source WordPress platform and not giving back enough. So he asked them to pay a licensing fee or "make in-kind contributions to the open source project."

We've seen this play before, typically around AWS commercializing an open source project that is also backed by a commercial entity (which, to be clear, they have the legal right to do). In the enterprise world, the remedy has been licensing changes that specifically prevent the large clouds from offering hosted services based on specific open source projects.

Typically, most people would sympathize with Mullenweg here, but he didn't help himself by first arguing this whole spat had something to do with how WP Engine manages revisions in the editor when this was very much about money.

All of this, of course, is complicated by the fact that Mullenweg isn't simply the benevolent overlord of open source WordPress. He is a director of the WordPress Foundation and also runs Automattic, which offers its own commercial WordPress hosting service that competes with WP Engine. So when WordPress.org cut off access to vital community repositories to WP Engine users (it has since restored it), it's hard to see that as purely a discussion about open source values.

What a mess.

The WordPress vs. WP Engine drama, explained | TechCrunch

The world of WordPress, one of the most popular technologies for creating and hosting websites, is going through a very heated controversy. The core issue

TechCrunchIvan Mehta

The Largo Embargo

I wrote a short post on LinkedIn a few weeks ago about founders who ping reporters with news that is already a few weeks old. It's no secret that I have a tendency to make fun of bad PR pitches (I think all journalists do – and it's only fair for PR to push back sometimes), but many founders genuinely don't know how embargoes work. Sadly, quite a few PR professionals don't either, or I wouldn't have an inbox full of embargoes that I never accepted.

So let's do a quick Embargo 101:

The general idea here is that you share information with one or more journalists before it becomes public. You set the embargo date and time for when that information can be published. Some writers may push for an exclusive when you reach out to them, which may be worth considering.

Here is the process for that: you ping the writer with a vague outline of the news and ask if they will accept the embargo ("Hey, we have some financial news/product release/etc. coming up. Would love to share it with you. Can you accept the embargo of October 4 at 7am PT?").

This part is key: you can't just send someone an email with your press release and say it is under embargo.

This is a two-way street: unless the writer has accepted the embargo, they are not under embargo! If the journalist didn't accept it, they are free to write at any time.

All of this works under the honor system. We're not VCs. If you ask for an NDA, don't expect to ever hear back.

It's worth noting that some writers simply don't work with embargoes. I don't mind them. The advantage is that the writers have time to look at your news, set up interviews, ask questions, and spend a bit more time analyzing it.

My workflow is simple: accept the embargo, set up an interview, write. The vast majority of my stories are based on what we talk about in that interview.

One more thing to note: there are no guarantees. A journalist may accept an embargo and not write about your company.

Fun things I wrote about:

United's new Wi-Fi deal: last week, I had a good chat with the Chief Customer Officer of United Airlines, Linda Jojo, after the company announced its plans to replace its current Wi-Fi providers with Elon Musk's Starlink. 

Why United chose SpaceX’s Starlink to power its free Wi-Fi | TechCrunch

Late last week, United Airlines announced that it signed an agreement with Elon Musk’s SpaceX to bring its Starlink internet service to its entire fleet

TechCrunchFrederic Lardinois

AWS bringing OpenSearch to the Linux Foundation: talk about open source drama... Back in 2021, AWS forked Elasticsearch after the company behind the open source project changed its license to explicitly disallow the large cloud providers – and most specifically AWS – from using its code to power their own hosted Elasticsearch services.

Now, AWS is launching the OpenSearch Foundation under the Linux Foundation umbrella. As it turns out, OpenSearch became far more popular than even AWS expected and there is now enough of a community to finance a foundation and keep the project going independently. And as is the case with many similar projects, being so closely related to a single company is also likely holding others back from contributing. You're not likely to see Walmart, for example, contributing to an AWS project.

AWS brings OpenSearch under the Linux Foundation umbrella | TechCrunch

AWS today announced that it is transitioning OpenSearch, its open source fork of the popular Elasticsearch search and analytics engine, to the Linux

TechCrunchFrederic Lardinois

Meeting bots: I also wrote about two AI meeting bots in recent weeks, courtesy of Fathom raising $17 million and Atlassian acquiring Rewatch to improve the Loom video messaging tool it acquired last year. Atlassian's strategy is to use meetings as another data source for its larger AI initiative, which makes a lot of sense given that this will help augment data from Confluence wikis and Jira. Loom AI could, for example, automatically generate a Jira issue based on what is discussed in a meeting. 

Exclusive: Atlassian acquires Rewatch as it gets into AI meeting bots

Atlassian plans to integrate Rewatch into its recently launched Rovo AI platform so that transcripts become searchable within the overall business context.

TechCrunchFrederic Lardinois

In my discussion with Fathom co-founder and CEO Richard White, we talked about the same idea: meeting transcripts are underused and undervalued right now, but once the AI tools are smart enough, they can use those transcripts to help employees with a lot of the busy works that comes after a meeting, be that writing up action items or updating an issue tracker. 

AI notetaker Fathom raises $17M | TechCrunch

In many meetings today, it sometimes feels like there are more AI notetaking and transcription bots than people. There are seemingly dozens of options to

TechCrunchFrederic Lardinois

Everybody loves Postgres: Supabase raised an $80 million Series C for its Postgres-centric Firebase alternative.

Exclusive: Supabase, a Postgres-centric developer platform, raises $80M Series C

With this, Supabase has now raised a total of $196 million, including a Series B round in 2022, which was also an $80 million round.

TechCrunchFrederic Lardinois

I've followed the development of quantum computing for the better part of the last decade. For the longest time, it's been "just around the corner" and with the advent of generative AI, any of the hype around the technology has receded into the background. But as with so many technologies, progress has been steady, with new error-correction techniques for building more durable, noise-resistant quantum circuits slowly making it out of the theoretical research arena and into the hardware.

IBM thinks it'll have a system with 1,000 logical qubits and low error rats ready to go in 2030. Earlier this summer, Ionq said it would be able to offer a system with 100 logical qubits next year (and a 1,000-qubit machine after that, which would then achieve "broad commercial advantage." Microsoft, which has been working closely with Quantinuum (all while working on building its own quantum computer), recently said that it believes a machine with 100 logical qubits could "potentially solve scientific problems that are unsolvable on classical machines."

All of this goes to show that there is now a reasonable chance that we will see useful quantum computers within the next five to ten years. And that has some interesting consequences for enterprises today: at some point in the not-so-distant future, quantum computers will be able to break the standard public-private cryptographic algorithms that secure virtually all data today.

A couple of months ago, I talked to Dario Gil, the director of research at IBM who is responsible for much of Big Blue's quantum computing program. IBM and others have long worked on creating post-quantum cryptography techniques that are hard to break for even the most powerful quantum computers.

Current encryption algorithms are secure because the calculations required to break them would take an impractically long time on traditional computers. But Shor's algorithm shows that it'll be easy for a quantum computer to break these security schemes. Exactly how many logical qubits it would take to break today's public-private key systems is up for debate, but chances are a machine with 2,000-3,000 stable qubits will likely suffice.

For my TechCrunch story, I talked to Dario about what this means for businesses today. After all, it'll take years -- if not decades -- to switch out all of the existing cryptography systems for ones that are resistant in a post-quantum world.

There's also another scenario to keep in mind: you can reasonably assume that if an adversary (and we're likely talking nation-states at this point) gains access to your encrypted data today, they won't be able to do much with it today, but if they hoard it, they'll be able to decrypt that information at some point.

This week, the National Institute of Standards and Technology (NIST) published the first set of standards for post-quantum cryptography. Now that these are available, Gil argued, it's time for developers to start implementing them to replace the existing systems.

Ideally, there'll be drop-in replacements for existing libraries, but it probably won't be that easy – especially given the long tail of applications today that use encryption.

You can read my full story here.

Also from me this week

• Definity raises $4.5M as it looks to transform data application observability: a look at data observability startup Definity, which aims to give businesses better real-time insights into their data pipelines while their data is still in motion.
• I also had a fun conversation with the founders of CannonKeys, a mechanical keyboard company out of Providence, Rhode Island, founded by a former Microsoft and Klaviyo engineer.
  
  A few weeks ago, CannonKey's CCO reached out to me and we met up for coffee in Portland. He brought along the company's latest keyboard, the Sat75 X, a fun, affordable keyboard with a cute little LCD screen. Building that board reminded me of how much fun it can be to build something physical again.
  
  Full story here.
  
• (Bonus: I also reviewed the Keychron K2 HE earlier this month)

It's been a week... Given everything that's happened lately, it's hard to believe that the CrowdStrike outages hit only a week ago. We're now deep in the clean-up phase of that particular disaster and while the blame for this particular incident falls squarely on CrowdStrike, it's been interesting to watch how Microsoft is reacting as well. It was Windows machines, after all, that were affected by this issue and quite a bit of media attention focused on that, not CrowdStrike, which is (or was?) pretty much unknown to a mainstream audience. A headline that mentions Microsoft is always going to get more clicks than one that mentions CrowdStrike.

CrowdStrike offers a $10 apology gift card to say sorry for outage | TechCrunch

Several people who received the CrowdStrike offer found that the gift card didn’t work, while others got an error saying the voucher had been canceled.

TechCrunchLorenzo Franceschi-Bicchierai

The company line at Microsoft is that the company wasn't able to make Windows more secure because of a 2009 agreement with the EU that ensured that third-party security tools would get the same access to the Windows kernel as Microsoft's tools. It looks like Microsoft will use the CrowdStrike incident to either push back on this requirement or figure out ways to secure its kernel better. Let's see how that plays out in the next few weeks.

My colleague Ron also talked to a few experts about best practices in software testing to avoid issues like this in the first place:

How to prevent your software update from being the next CrowdStrike | TechCrunch

There is no fool proof way to prevent a buggy update like CrowdStrike’s, but there are best practices that could mitigate the fallout.

TechCrunchRon Miller

Adv-VANTA-ge

I spent more time editing than writing this week, but I did get a few stories out. The most important of these is the $150 million in funding for Vanta (now valued at $2.45 billion), a startup that helps businesses automate their security and compliance processes.

Is that the most exciting of problems to solve for a startup? Obviously not, but if you're selling into an enterprise today, chances are you need to be able to show SOC 2 compliance and your adherence to a few more similar standards as well.

What was interesting in my conversation with Christina Cacioppo, the co-founder and CEO of Vanta, was that she is starting to position the company to focus more broadly on 'trust.'

“Vanta today, we still do a lot of SOC 2, but a lot of what we’re building is around how do you help companies build out their security programs?” Cacioppo told me. “And then how do they go get credit? There’s a compliance piece, there’s the trust centers, there’s real-time security status pages and questionnaire automation, but the thesis behind a lot of that is: if you can give people credit — which really means revenue — for showing off all the good security work they’ve done, they will do more good security work. … When we talk about trust, a lot of trust in software, especially B2B software, it’s around: Can I trust you with my customers’ data?”

Vanta raises $150M Series C, now valued at $2.45B | TechCrunch

Vanta, a trust management platform that helps businesses automate much of their security and compliance processes, today announced that it has raised a

TechCrunchFrederic Lardinois

SPACs still exist

Another story I wrote this week was about Cyabra, a startup that helps businesses and governments track down fake social media accounts that spread mis- and disinformation. Cyabra went public by merging with a SPAC, something I hadn't seen for a long time – and something that's gone rather out of favor in recent years.

For that story, I got a comment from Mike Pompeo – yes, that Mike Pompeo. "Cyabra is a crucial partner in the fight against disinformation," he said. "Their capabilities in uncovering inauthentic accounts, false narratives, and manipulated AI content are vital for protecting democracy and safeguarding national security. Cyabra’s dedication to these goals makes them an essential ally in our efforts to defend against threats to our nation."

I couldn’t mention this in the TechCrunch story, but there’s irony in someone who defended a habitual liar like Trump now fighting misinformation.

Cyabra, a startup helping companies and governments detect disinformation, plans to go public via SPAC | TechCrunch

The transaction values Cyabra at $70 million, and the company expects the merger to close by the end of the year.

TechCrunchFrederic Lardinois

And that's that for this week. I've got about half a dozen new mechanical keyboards under my desk that I need to test and write about, so expect a lot of content about switches, keycaps and thock in the next few weeks.

Renting servers in the cloud isn't cheap. Traditionally, the argument has been that it's still cheaper than buying your own hardware and having a dedicated staff managing servers. Plus, you get to benefit from the innovations that Amazon, Microsoft, Google and Co. are bringing to market. And then there's the uptime! Think of all the nines!

But at least for the last two or three years, I've been hearing more and more complaints about how pricey the cloud has become and barely a week goes by without seeing a new startup pitching me a new tool for bringing cloud costs down.

This week's newsletter was typed on the Akko MOD 007B PC Santorini.

The cloud price wars are most definitely over -- with maybe one exception: egress fees. Google stopped charging its customers any fees for transferring their data out of its cloud in January. This week, AWS followed suit. This isn't just out of the goodness of their hearts, of course. The European Data Act more or less forces them to do so. The missing hyperscaler here is Azure, but that's just a matter of time.

There's a lot more going on here, of course, including enterprises trying to consolidate their cloud spending while times are leaner, AI training and inferencing taking off (and being quite expensive) and more. But that's for another week. Let's talk about what's new in private cloud instead.

For a long time, private cloud options were either closed and expensive, like vSphere, or open and complex to manage, like OpenStack – and since virtually nobody in enterprise IT knows how to deploy and manage a giant open platform like OpenStack, you end up paying a vendor for that, too.

This week, I wrote about Ubicloud, a new open-source project from the team behind the Microsoft-acquired Citus Data. Ubicloud is building a new cloud system that businesses can run in their own cloud or on bare-metal servers from affordable providers like Hetzner and OVH. Right now, there's only a compute, storage and networking piece to this, as well as a PostgreSQL database.

But the mission here is to build an alternative to OpenStack by building an open alternative to AWS. A Kubernetes platform is coming next, which will then make it very easy to deploy virtually any application on the platform.

"We're designing for simplicity," Ubicloud writes in its docs. "With OpenStack, you pick between 10 hypervisors, 10 S3 implementations, and 5 block storage implementations. The software needs to work in a way where all of these implementations are compatible with each other. That leads to consultant-ware. We'll take a more opinionated approach with Ubicloud."

Them's fighting words.

Ubicloud already offers a managed service, too, making getting started quite easy. It's very early days, obviously, but I rather like the idea of somebody trying to build a new platform like this.

A couple of other stories I wrote this week:

• Numbers Station launches its cloud service
• Brevian raises funding for its no-code AI agent platform for enterprises.

Another busy week in tech news, but mostly in areas outside of my immediate beat (goodbye, Apple Car), allowing me to focus on just a few stories and even get out of the house for a few in-person meetings.

This week, I got to chat with GitHub CEO Thomas Dohmke. Among all of the AI services out there -- except ChatGPT, I guess -- GitHub Copilot has had the largest impact on developers' day-to-day work. Now, GitHub is launching its enterprise version of Copilot into general availability and the main feature here is that it's now making internal code bases and documentation available in the Copilot experience. That's huge for large businesses that have a ton of internal docs and libraries.

🗞️ This newsletter is also available on Ghost. Sign up here.

Looking ahead, Dohmke told me, we will see Copilot in more places where developers are already doing their work. When I talked to him, I didn't realize the importance of what was almost a throw-away comment. Dohmke told me that part of this mission to bring Copilot to more places involves its partner program, which the company originally announced last year. That program allows partners to build plugins for Copilot. At the time, GitHub said those partners included the likes of Datastax, LaunchDarkly, Postman, Hashicorp, and Datadog.

During our conversation, Dohmke also mentioned developer Q&A site Stack Overflow. "We are currently working with this early partner program with many partners, including StackOverflow, on integrating these destinations into corporate and corporate chat," he said. Seemed reasonable and I didn't really think twice about it at the time.

Then I got a call from Google to talk about their partnership with Stack Overflow and intentions to bring Stack Overflow's data to Google's Gemini model and build Stack Overflow's Q&A capabilities right into the Google Cloud console.

You can read all about what that will look like here, but I think the more interesting question is about the value of content in the age of large language models.

⌨️ This week's newsletter was written in iA Writer on the AM Relic 80.

We've seen Reddit make deals with OpenAI, publishers selling their archives to various AI firms -- and there are even brokers who will handle all of this for publishers now. So that's obviously no new. There are no LLMs without content, after all, and those content owners want to get paid.

But for Stack Overflow, this creates a really interesting conundrum. The purpose of Copilot and its brethren is to keep developers in a flow -- a flow that explicitly avoids having to go to Stack Overflow to look for answers. However, Stack Overflow needs people on the site to ask questions and answer them (for free).

The site boasts almost 60 million questions and answers. That's clearly a lot, but those Q&As can also quickly grow stale and outdated.

Google is explicitly combining a more old-school Stack Overflow experience in addition to feeding the data into its models. Maybe that will work, but are developers really going to move back and forth between those two modes when code completion and in-IDE chat is good enough? That chat, after all, will also be trained on all kinds of other programming docs, too.

It'll be interesting to see what a GitHub/Stack Overflow integration will look like and how Stack Overflow will chat a path into this future.

Productivity Scrapbooking

Something else I've been thinking about a lot in the last few weeks is productivity tools. I don't know why I went down that path, but my best self-diagnosis is that as the media industry spirals downward, my workflow is something that I have control over.

For years, I used Trello as my go-to task manager to keep track of stories and deadlines. Combined with Google Calendar to handle meetings, that worked fairly well. But I grew frustrated by having to juggle personal and work calendars -- and corporate security rules that make it impossible to combine the two or use something like Notion Calendar.

In the end, I decided to use a good old hard-bound daily planner. I found a 24-hour one that is grouped by weeks and that's been working really well for me (I need 24 hours instead of a more typical work day-only planner to keep track of embargoes). Now I transfer every meeting right into my planner, which has the added benefit that I'm more conscious about what's ahead.

Before making that decision, I went down the YouTube productivity rabbit hole -- and it wasn't pretty. A lot of the YouTube productivity economy seems to be about creating pretty bullet journals and color-coding and stickering daily planners. I can't help but think that most of those people don't use the tools they talk about. It's productivity scrapbooking.

One good outcome of that, however, was that it led me down the path of rethinking some of my note-taking tools, too. The vast majority of my meetings are interviews. Ten years ago, I wrote those notes by hand, then moved to OneNote. With the advent of transcription services like Otter, I abandoned note-taking except for writing down the occasional keywords to remember something to follow up on during a meeting.

Yet the constant barrage of information we're all faced with and the ever-shrinking attention spans that has led to (or maybe that's just me?), I've found that I needed a place to collect my thoughts again. I tried going back to writing notes by hand, but while that works during interviews, it's not a viable way to keep notes on things I read online during the day. I tried a few tools like Notion (felt too much like work) and Obsidian (felt like even more work), but nothing really stuck until I happened upon Capacities. Somehow that clicked for me, even as I only use maybe half of its features and miss some (like a web clipper).

In a way, that's weird, because all of the "personal knowledge management" hype doesn't resonate with me at all. It feels like busy work masquerading as productivity, too. But having a single place to store links, story ideas, task lists etc. -- and a simple tagging system to keep it all organized -- is rather nice.

Let's see if I stick with it.

Bonus: if you need a good laugh (or cathartic cry, depending on if your current job involves deploying Kubernetes clusters):

One nice thing about writing a newsletter is that it’s a break from the day-to-day hustle to get stories out of the door. It’s also a chance to go back to some older stories and see how they played out – and sometimes it only takes a week for those stories to change entirely. Here is a good example:

What’s next for Mozilla?

When I first wrote about the leadership changes at Mozilla a few weeks ago, I admit that I was skeptical. Replacing Mitchell Baker with Laura Chambers, who despite Mozilla’s claims, never really held a product role and doesn’t have a background in open source, seemed at best like an odd move and, for quite a few pundits, it looked like another Mozilla blunder. Now, I think it may have been the right move. Sometimes you need a new leader to come in to make the hard decisions.

Today’s newsletter was written on the Keychron Q8.

About two weeks ago, a week after the leadership change, I got a tip that there would be layoffs at Mozilla, with about 60 people impacted. Mark Gurman over at Bloomberg beat me to that story, but luckily I also got the internal memo about what that meant for Mozilla’s product portfolio. Since Bloomberg didn’t have that context, I quickly retooled my story to focus on that instead.

As it turns out, Mozilla is shutting down projects like its 3D virtual world Hubs, scaling back its investments in its Mastodon instance and some of its security and privacy tools (which are often just repackaged and re-branded versions from partners anyway). Instead, Mozilla wants to focus on bringing trustworthy AI to Firefox and it’s combining the Pocket, Content and AI teams with the Firefox organization.

Mozilla is great at getting distracted. Refocusing on Firefox is a necessity before it loses what is left of its market- and mind-share.

Mozilla downsizes as it refocuses on Firefox and AI: Read the memo | TechCrunch

After installing a new interim CEO earlier this month, Mozilla, the organization behind the Firefox browser, is making some major changes to its product

TechCrunchFrederic Lardinois

And let’s be real. I have no idea if there’s a future for Firefox, but Mozilla’s future is tied to the success of its browser, despite its best efforts to change that. Sure, the Google deal will get renewed for the foreseeable future. Google needs it as much as Mozilla to keep it out of yet another anti-trust lawsuit (even as everyone knows that it’s just playing a game here by propping up a competitor to Chrome). Yet, once Firefox becomes completely irrelevant, there's no value in keeping up the charade for Google anymore either.

Mozilla is betting heavily on (small) local AI models and the natural place for the organization to bring that to market is within Firefox. It’s unclear to me what that’ll look like, but the organization regularly points to Fakespot as an example.

Either way, I do feel more optimistic about the future of Mozilla today than I did a month ago.

What's next for digital media news organization?

Talking about layoffs... The media bloodbath continues: Vice is essentially shutting down, BuzzFeed is cutting 16% of its employees, and our Yahoo cousins at Engadget laid off 10 people, including its top leadership team on the editorial side.

The Engadget layoffs hit close to home, of course, especially after we had our round of layoffs at TechCrunch a few weeks ago. We do both belong to the same private equity-owned parent organization, after all. Nobody has ever accused a private equity firm of doing what’s in the best interest of employees (of the future of journalism).

Going forward, Engadget is going to be split into a News and Features team, that will focus on news and “traffic growth” and a Reviews and Buying Advice team, that will focus on evergreen content and guides. The second part is code for commerce posts – the current bane of the Internet.

When advertising rates fall and subscription businesses don’t work out, what’s left? The current answer: commerce posts that generate affiliate fees from Amazon and the like. The logical result of that is what’s left of Cnet today or (because Apple News keeps recommending this to me), the Travel Products section of Travel+Leisure. Some of the smarter publishers hide this stuff on their homepages but know exactly how to perform the necessary SEO voodoo to have it rank high on Google.

For a while, Wirecutter got this model right and felt trustworthy – and maybe it still is in some areas – but the fact that every random site now features product recommendations that are clearly not based on actual reviews makes every other site that uses this same model feel less trustworthy. It also devalues the journalism on those sites.

One thing I do know: if you're a journalist and your leadership starts unironically referring to your work as 'content,' start thinking about a plan B.

What's next for open-source in the age of AI?

Let’s get back to tech, though. Google launched Gemma this week, a set of open models that perform quite well against Meta, Mistral and others competitors.

That is interesting by itself, but it got me thinking once again about what this means in the context of open source. Because these models are ‘open weight models’ that may have permissive licenses but are explicitly not open source. Google stressed that multiple times in its press conference before it launched the Gemma models and noted that it did so to ensure it could release these models in a responsible way (Google’s own Gemini model also showed what happens when it’s done in a way that’s maybe a bit too responsible). To me, that feels like a bit of a cop-out. It allows you to claim to be ‘open’ without having to go all the way and, among other things, disclose your training data and other details.

I know the folks over at the Open Source Initiative are deep in the weeds on what an open-source definition would look like. One of their core tenets, best I can tell, is that to be open source, a model’s creator would have to allow users to study how the system works and be able to inspect its components.

Right now, though, there’s still a discussion around what the word ‘system’ actually refers to in this context. I know. I know. Fascinating stuff, but also of vital importance to what AI development will look like in the future. More about that in a future newsletter.open-weight

Google launches two new open LLMs | TechCrunch

Barely a week after launching the latest iteration of its Gemini models, Google today announced the launch of Gemma, a new family of lightweight

TechCrunchFrederic Lardinois

One lesson to learn from this edition: I'm much better at asking questions than providing answers...

Note: This newsletter is a personal project that is not affiliated with Yahoo and TechCrunch.

The last few months have been rough for the news business. I was going to write something about that last week, but then it was TechCrunch's time to get hit with a round of layoffs (my team and I got spared, thankfully). The company also shut down our subscription product, TechCrunch+ (a.k.a. Extra Crunch).

There isn't much more I can say about the details in public, but it's no secret that the entire industry is struggling. In tech media, you can add the overall pullback in marketing spend to that, given how dependent most sites are on ads from major tech brands. It's good to have a conference business.

I think in the long run, niche sites will be just fine. I worry more about general news sites. The Messenger clearly couldn't make it, despite getting traffic (though it surely had plenty of other problems, too). But traffic isn't what it used to be when it comes to monetizing a site. Sidenote: I finally read Ben Smith's Traffic, so this has been top of mind for me, even as I think the book is a bit too myopic in its focus.

News brands come and go. I've been lucky to have a very stable career so far, but I'm also very aware how unusual that is in this business. For the individual journalist, the publication you work for is often everything. It's a paycheck, for sure, but there's also the umbrella of that publication's reputation.

Unless you have an established brand, you can't just go and create a newsletter and charge $10 a month for that – and not everybody can build that brand anyway. Plus, not every topic lends itself to the newsletter format and even if it does, it's going to take a very long time to build up that business. Not everyone gets a 27,000 subscriber list from their former employer as a parting gift.

Enough about that for now, though. I'm sure I'll have more to say about it in the future.

Here is what I wrote about this week, with a bit of extra color for good measure:

Google Gemini Ultra

Google launches Gemini Ultra, its most powerful LLM yet | TechCrunch

Google Bard is no more. Almost exactly a year after first introducing its (rushed) efforts to challenge OpenAI’s ChatGPT, the company is retiring the name

TechCrunchFrederic Lardinois

Google finally launched its best large language model to the public this week. I don't have a good sense of how well this landed for Google, but having used it for a day now, Gemini Ultra 1.0 does feel like it's on par with GPT-4. That's the expectation Google had set when it launched Gemini Pro and it lives up to that.

Still, it's a bit uneven at times – sometimes it'll happily summarize my inbox, for example, other times it'll tell me that it doesn't have access to it – but it's exactly this integration with the rest of the Google products that gives it an edge, at least in the consumer and SMB space.

Gemini always had a bit of a tendency to hallucinate and I've found that with Ultra as well.

I'm also glad that Google retired the old Bard and Duet AI brands in favor of just 'Gemini.' That branding was always confusing and Bard didn't have the best of starts, so a fresh name may just help here. What is still confusing, though, is that you need a Google One subscription to access Gemini Ultra, but that also helped Google push its One business to over 100 million subscribers now.

On Android, you can also have Gemini replace the Google Assistant. What does that mean for the future of the Google Assistant? Google won't say, but the writing is on the wall.

Google Assistant is now powered by Gemini -- sort of | TechCrunch

Google Assistant users now have the choice of switching to one of Google’s Gemini AI models to answer queries, generate artwork and more.

TechCrunchKyle Wiggers

AWS and Amazon remain nowhere to be seen in this business (and yes, I know they have the Titan model, but be honest, until you read this sentence, you had forgotten about that, too).

New Mozilla CEO

A New Chapter For Mozilla | The Mozilla Blog

Today marks a significant moment in our journey, and I am thrilled to share some important news with you. After much thoughtful consideration, I have decid

The Mozilla BlogMitchell Baker

I didn't write about this one, but I do have some thoughts:

Mozilla has been looking for a second act for a very long time. Firefox – and the Google search deal – keep it afloat, but Firefox is dwindling in its relevance. I would like Mozilla to focus on that, but instead, the organization (with its complicated relationship between the Foundation and the commercial side) is more focused on other projects that have far less defined of a mission. And yes, people want a browser that respects their privacy, but clearly, that isn't enough of a selling point.

When I talked to Mozilla president and executive director Mark Surma earlier this year, we talked a lot about the organization's plans around AI – and specifically the responsible use of it. I think Mozilla has a chance to play a role here, but it hasn't really established itself as a player here (and it doesn't have the money to build its own models that could compete with the open models of Meta, for example).

I don't know anything about the new interim CEO Laura Chambers. The announcement was big on corporate speak (" Outstanding Execution: Focus, Processes, Capabilities") but low on specifics.

All of this doesn't make me feel more optimistic about the future of Mozilla.

Also from me in the last two weeks:

Angry Miao only goes slightly over the top with the Relic 80 TKL keyboard | TechCrunch

To write about an Angry Miao keyboard always means to write about the design as much as the actual typing experience. Just look at the AM AFA or AM Hatsu.

TechCrunchFrederic Lardinois

Attack surface management platform Ionix adds another $15M to its $27M Series A round | TechCrunch

Ionix (formerly Cyberpion) helps enterprises reduce their attack surface by giving them a better view of their overall security posture and software

TechCrunchFrederic Lardinois

Aim Security raises $10M for its GenAI security platform | TechCrunch

Securing generative AI-based applications from prompting attacks, ensuring that they don’t leak sensitive data or create malicious output or results that

TechCrunchFrederic Lardinois

Nile raises $11.6M in seed funding to build a Postgres-powered data layer for SaaS applications | TechCrunch

SaaS needs a new data system. That’s the driving idea behind Nile, a startup that aims to create this data system with serverless Postgres at its core.

TechCrunchFrederic Lardinois

Inngest raises $6.1M as it expands its workflow engine | TechCrunch

Inngest, a startup that makes it easier for developers to build and manage serverless backends, today announced that it has raised a $6.1 million funding

TechCrunchFrederic Lardinois

It's been a week. After snow and ice paralyzed Portland over the long MLK weekend and we lost power for well over a day, more ice and wind wreaked havoc throughout the week. America's infrastructure is brittle, but with more extreme weather by the year, maybe it's at least time to bury power lines underground... With all of those distractions, plus a bit of a lull in my news cycle after the holidays and CES, I didn't write as much as I normally would. Judging by my inbox, that'll change in the coming weeks.

A better calendar

My main story this week -- and the one our readers at TechCrunch seemed most interested in -- was the launch of Notion's stand-alone calendar app, based on their acquisition of Cron. It's a nifty calendar and if it wasn't for our IT department keeping me from using third-party apps without Google Workspace subscription, I'd probably move to it full-time. Ahead of the announcement, I talked to Notion CEO Akshay Kothari, whom I hadn't talked to since his Pulse days almost a decade ago. While Notion is a hit now -- and has a bit of a cult following -- he mentioned how the team didn't get things right in the early days, in part because users couldn't quite figure out what they should use Notion for. It was simply too much of a blank canvas. That worked for some but left too many users scratching their heads. The way they fixed that was also simple: add starter templates. As for Notion Calendar, you can read my review here.

Vectors everywhere

I also covered Pinecone's update this week. Vector databases are everywhere now and virtually every database company I talk to wants to talk about them, which makes sense, given that they are the ideal instrument to provide context and long-term memory to large language models. Pinecone didn't invent this space but the team leaned into the LLM market early and hard, which isn't a surprise since the founding team came from Amazon's AI divisions. The new Pinecone Serverless architecture promises to be up to 50x cheaper to run than the previous generation. That's a large number and those tend to make me skeptical, so I didn't make a big deal out of it in my story.

Also from me this week:

• Snyk acquires Helios to bolster its AppSec platform
• Infield wants to make open source dependency management trivial

I hope to keep the power on next week (likely, since I finally ordered a largish (2kWh) solar power station), so I'll hopefully get to write a bit more.

Google has been very conservative about releasing APIs for Google+ and may not even release a full read/write API for its social network before the end of this year. The one API the Google+ team has put its weight behind, though, is the Hangouts API, which gives developers access to Google+’s video chat features. Today, the company announced that it is taking this API out of preview. This means developers can now launch and share their hangout apps with the Google+ community. To launch this feature, Google has partnered with a number of developers, including Aces Hangout, Cacoo, Scoot & Doodle, Slideshare and Clubhouse Challenge by Bravo.

A simple click on the Google+ hangout button on these sites opens up a standard Google Hangout with the respective application running inside the same window. You can now also find these apps in the new “Apps” pane in Google+ Hangouts.

Google is rolling this feature out right now, so it may take a bit before it becomes available in your account.

Google, which has been struggling to get users to adopt Google+ as a social network, has long been pushing these video chats as a differentiating feature for its service. It’s not clear how much adoption this feature has seen on Google+ itself (even as video chats become more common, most people still feel very self-conscious on camera, after all). By effectively decoupling hangouts from the social networking aspects of Google+, however, I wouldn’t be surprised if a number of productivity apps and games will adopt this feature now. Google, as far as I can see, isn’t charging developers for this tool and adding video chat tends to be a costly feature for developers.

Mozilla, in cooperation with French developer Little Workshop, launched a new MMORPG called BrowserQuest today to demonstrate what developers can do with HTML5, WebSocket, Canvas and other advanced web technologies, including Node.js. The game, which is actually quite fun in its own right, should work with virtually every modern desktop browser (except for Internet Explorer), as well as Safari on iOS and Firefox on Android. The mobile version is, in Mozilla's words, "more experimental," but should be seen as "an early glimpse of what kind of games will be coming to the mobile Web in the future."

The one technology Mozilla really wants to showcase here – besides the multiplatform nature of using web technologies over native apps – is WebSocket. With this, developers can set up a system to communicate back and forth between the browser and the server. In the case of BrowserQuest, this means that the server can keep your actions and those by your fellow players in sync without much effort. 

Here is the total list of web technologies BrowserQuest uses: 

• HTML5 Canvas, which powers the 2D tile-based graphics engine.
• Web workers, allowing to initialize the large world map without slowing down the homepage UI.
• localStorage, in which the progress of your character is continually saved.
• CSS3 Media Queries, so that the game can resize itself and adapt to many devices.
• HTML5 audio, so you can hear that rat or skeleton die!

Just in time for the cherry blossom season in Japan, Google today launched what it calls a "new visual travel guide" for the country. This new interactive guide, which Google created in cooperation with the Japan Tourism Agency and the Japan National Tourism Organization, is based on Google's street view images and allows you to visit eight distinct areas in Japan (though it's worth nothing that four of these are actually in Tokyo, including the city's famous fish market and the Ahikabara and Ginza shopping districts).

The guide, it is worth noting, also features some indoor imagery, as well as the ability to virtually stroll through some of the country's famous gardens. Besides the obvious Street View feature and sightseeing suggestions, the guide also features restaurant, hotel and shopping recommendations – all of which feature indoor Street View images and a bit of information about the establishment.

According to Google, the guide features 339 locations – including "26 great cherry blossom viewing spots." The guide is available in English, Japanese, Korean and Chinese.

It will be interesting to see if Google plans to expand this program to cover other countries as well. While there are a number of third-party sites that use Street View to augment their travel guides, this guide for Japan is actually Google's first foray into using this feature for a home-grown travel guide. The fact that the company calls this an "edition," though, provides a hint that more of these guides may be in the works.

Car manufacturers are slowly but surely morphing into technology companies and it's not just upstart manufacturers like Tesla who are trying to piggyback on Silicon Valley's deep pool of engineering and academic talent. General Motors, for example, already has offices in Palo Alto and today, Ford announced that it has also chosen Palo Alto for its new Silicon Valley lab. The company already has research labs in Dearborn, MI; Aachen, Germany; Nanjing, China and Tel Aviv, Israel. The lab will be lead by TJ Giuli, a Ford research engineer and Stanford-alumni who has been working on the forefront of the company's tech efforts for the last few years. 

The company hopes that the new lab will "serve as a hub for independent technology projects and identification of new research investments and partners located along the West Coast." It also hopes to develop deeper partnerships with local tech firms and universities. Ford is, for example, already a member of Stanford's Automotive Research at Stanford (CARS) affiliate program.

Ford will hire Silicon Valley talent for its new offices, but also plans to rotate-in engineers from its other locations to work on new in-car technologies and ideas.

The company first announced its plans to open a Silicon Valley office earlier this year, but at that time, it hadn't chosen a location yet. According to the city of Palo Altos' economic development manager Thomas Fehrenbach, the city itself reached out to Ford when it heard about Ford's plans, though chances are that it was already pretty high on Ford's list of possible locations.

About three quarters of U.S. teens (77%) now have a cell phone. This number is, according to a new report by the Pew Internet & American Life Project, up slightly from September 2009 (75%). Looking back to 2004, though, it’s clear how this number has increased dramatically over the last few years. In late 2004, only 45% of teens had cell phones. The interesting trend here, though, is that 80% of older teens (14-17) now own mobile phones (and 31% of these older teens own smartphones), but that the number of younger teens with cell phones has actually dropped a bit (57% vs. 66%).

Unsurprisingly, teens from households with a higher income are more likely to own cell phones and while 83% of suburban kids now own a mobile phone, only 69% of kids in urban areas and 73% in rural areas own one.

The Pew study did not find a statistically relevant difference between boys and girls when it comes to cell phone ownership, but when it comes to texting, girls are still far heavier users than boys. The median number of texts per girl in this study was closer to 90, while boys only send about 50 (the mean numbers, it’s worth noting, are far higher and also far closer to each other: 165 for girls and 168 for boys).  Unsurprisingly, those teens who send the most texts are also more likely to own a smartphone.

As for old-fashioned voice calls, teens – just like the rest of us – are slowly making fewer calls and most teens report that they mostly use text messages to socialize with the people in their lives: