Tech Brew Ride Home - Fri. 04/14 - AWS Enters the Gen AI Race

Episode Date: April 14, 2023

AWS has begun offering customers access to LLMs made by Anthropic, Stability AI, and AI21 Labs, as they look to be a neutral platform for generative AI features. WhatsApp is rolling out new verification and protection features, including defenses against SIM jacking and social engineering attacks. Twitter has just increased the character limit to 10,000 for Blue subscribers. And your weekend longreads suggestions. Sponsors: Ramp.com/techmeme Bloomberg.com/careers Links: Amazon Joins Microsoft, Google in AI Race Spurred by ChatGPT (Wall Street Journal) Amazon Is Joining the Generative AI Race (WIRED) WhatsApp makes it harder for scammers to steal your account (Engadget) Key transparency explainer (Matthew Green, Twitter) Twitter Blue’s new 10,000 character limit turns tweets into essays (The Verge) Dril Is Everyone. More Specifically, He’s a Guy Named Paul. (The Ringer) With a wave of new LLMs, open-source AI is having a moment — and a red-hot debate (VentureBeat) Is ‘The Economist’ Making the Best News Podcasts Right Now? (Vulture) Learn more about your ad choices. Visit megaphone.fm/adchoices

Welcome to the TechMeme ride home for Friday, April 14th, 2020. I'm Jackson Bird filling in for Brian McCullough today. AWS has begun offering customers access to LLMs made by Anthropic, Stability AI, and AI-21 Labs, as they look to be a neutral platform for generative AI features. WhatsApp is rolling out new verification and protection features, including defenses against simjacking and social engineering attacks. Twitter has just increased the character limit to 10,000 for blue subscribers and your weekend long read suggestions. Here's what you miss today in the world of tech.

Amazon Web Services has officially entered the Gen A.I. Ring. Yesterday, AWS announced a platform called Bedrock, which will offer customers access to large language models made by Anthropic, Stability, AI, AI-21 Labs, and AWS's own Titan. As Wired put it, quote, AWS doesn't have a chat GPT rival, but it wants to sell you the tools you need to build one, end quote. And quoting the Wall Street Journal, unlike Google and Microsoft, which have announced products for the general public, AWS is targeting corporate customers. In addition to new AI tools, the company is expanding access to custom-made chips that it says can run AI software more efficiently and cheaply than competitors. The largest three cloud companies, AWS, Microsoft, and Google, have put generative AI at the center of their sales pitches recently to try to capitalize on the explosion in interest in the technology.

The entire world is scrambling right now, said Shishir Motra, the CEO of AI Document Startup Koda, and an early tester of AWS's new AI products. He said the current rush for companies to ready themselves for this new technology resembles the shift from computers to smartphones. AWS is forging a different path. So far, avoiding a major investment in an outside AI company or consumer-facing tools, it says it wants to act as a neutral platform for businesses that want to incorporate generative AI features. By not being tied to any one AI startup, AWS is marketing itself as the Switzerland of the cloud giant. AWS is selling access to multiple large language models, allowing companies to build generative features off more than one.

AWS says its AI will be more suited for businesses because it can be trained only on a customer's data, says its internal cache of documents, rather than the broader set of webpages that other models use. That could make it a safer choice for businesses that are nervous their private data could end up shared and mixed up with other companies. Another feature the company is pushing is Code Whisperer, which generates and fixes computer code. It will compete directly with Microsoft's GitHub Copilot, which uses Generative AI. Previously, Amazon had made Code Whisperer available only to a small number of users, end quote.

As Vicki Boykis put it on Twitter, quote, buckle up, we're just getting started. In more verification news, yesterday I told you about LinkedIn's new verification features using employee email addresses, Microsoft Entra, and government IDs through Clear. Now, WhatsApp has announced a handful of new verification and protection features to thwart scammers, including defenses against sim jacking and social engineering attacks. First, there's a new registration check when you download WhatsApp on a new device. You might be prompted to use your old device to confirm that you want to move the account to the new one. This won't always happen, only if the company detects a suspicious registration attempt,

and if you don't have access to your old device, you can request a second one-time passcode. Quoting NGadgett, whether or not you decide to switch devices anytime soon, your WhatsApp account will be safer, thanks to the new introduction of new background checks. You won't need to directly interact with the verification features WhatsApp is adding. Nonetheless, the company says they will help secure your account against malware and better protect you if your WhatsApp is ever compromised. Separately, WhatsApp is also making it easier for users to verify their connection with someone is encrypted. Right now, verifying your connection with someone involves either scanning a QR code or comparing a 60-digit number,

both of which you can find by tapping the encryption tab under a contacts info sheet. Moving forward, tapping the tab will automatically verify whether your connection is secure. The new features will roll out to all WhatsApp users in the coming months. In the meantime, if you want to do the most you can to secure your account, WhatsApp parent company meta recommends you enable two-factor authentication and encrypted backups, end quote. Johns Hopkins cryptography professor Matthew Green posted a great explainer thread on Twitter, diving a bit deeper into key transparency and why WhatsApp's verification update is a big deal. Quoting Green,

In messengers like WhatsApp and Signal, you can detect attacks by having your friend compare a security code, safety number, or QR code, using some other channel. Almost nobody does this. Even my cryptographer friends are like, nah, let's yolo this one. One potential solution to this is key transparency. The idea of key transparency is that you, one, publish a single hash that commits to every key or identity in the system at a given time, then people can, two, compare their hashes to make sure the server is being consistent.

This means that I can verify that the server is linking the expected public key for my identity, and now it will be very hard for the server to say something different, i.e. offer a different public key for me to anyone else. End quote. He goes on to explain how effective it is, but how substantial the engineering lift is. Quoting again, proposals like key transparency don't address every one of these attack vectors, but they signal very loudly that developing further attacks in this direction,

probably is not a great investment. So the major signal we're getting from WhatsApp and from Apple a few months ago is that they intend to make secure messaging as safe and private as they possibly can. These systems are important and they will invest in blocking even sophisticated attacks, end quote. Twitter officially announced last night that they have upped the character limit to 10,000 characters if you are using Twitter Blue. It will also include, bold and italic text formatting. No more all caps or asterisks to indicate emphasis. Quoting the verge, the announcement comes just weeks after the character limit for Twitter Blue was first bumped up from 280 characters to 4,000 back in February, with the latest increase and

additional formatting features seemingly designed to help Twitter compete with newsletter platforms like Substack, which has recently found itself in Elon Musk's crosshairs. This isn't the first time Twitter has introduced long-form writing features. The company, under previous leadership, was testing a new notes feature for writers and had previously purchased the review newsletter firm in 2021, which served as a rival to Substack. Both of those programs were later scrapped by Musk. Twitter isn't being subtle about its intentions to rival services like Substack. The company recently rebranded its Superfollows feature, which allows users to subscribe to individual accounts to access exclusive content to subscriptions, advertising it alongside the new character limit

and formatting options as a means for creators to earn income directly through Twitter. Musk also promised that Twitter wouldn't be taking any money creators make through subscriptions for the next 12 months. Monetization through subscriptions is currently only available to users in the U.S. Twitter has an uphill battle to fight if it wants to shirk its legacy as a short-form blogging platform. While some creators have expressed an interest in the new long-form features, many everyday Twitter users appear reluctant to read beyond the platform's text snippets. Data reported by the information also indicates that Twitter Blue is struggling to attract subscribers, with the service estimated to have around 290,000 global subscribers. Just 0.1% of the roughly 250 million daily active users reported by Twitter last year.

end quote. And just to give you a sense of scale on the character limits here, a 15-minute episode of this show, read by me, is about 2,300 words. Brian goes faster, so he packs in more like 2,500 words. In a recent script, those 2,300 words equated to about 12,400 characters. So you can think of 10,000 characters as text that would be almost as long as this show to read. If you you did text to speech on a tweet that made use of the full character limit, it would be almost as long as an episode of this show. That is a long tweet. I'll also add here that the character limit on substack notes, while not publicized, their FAQ just dares you to hit it, is just over 4,000 characters. At least that's how many I was able to cram into a post. Did substack settle on just

over 4,000 characters for their notes, because at the time, that was the maximum anyone could tweet even with Twitter Blue. Did Twitter up the limit in response to substack notes? Maybe trying to make tweets themselves replacements for full newsletters? After all, 10,000 characters is more than enough for most people's newsletters. Will any of it actually have an impact on one company over the other? I guess we'll see. And now it is time for your weekend.

long read suggestions. First up from Nate Rogers at the Ringer, a profile on weird Twitter patron Saint Drill. Real name, as exposed six years ago, Paul Dockney. To many people, he's still pretty anonymous, so much so that I couldn't find a video with his last name being said aloud, so apologies if I am mispronouncing it. If you are a fan of drill, this is a fascinating pulling back of the curtain, and if you have no idea who or what I am talking about, strap yourself in. Quoting the ringer, to most people, he is nothing, show the unaffiliated some of his posts, and they will likely just generate confusion and possibly anguish. But to a large sect of the very online, he is king, the undisputed poet laureate of shit posting, the architect of a satire

so effective that it has become impossible to tell when Drill stopped mocking the way people speak online and when we instead started speaking like Drill online. For almost 10 years, he was entirely anonymous. Like a decent number of the people in the so-called weird Twitter scene that Drill is still vaguely a part of, he doesn't use his real name on the account. But as time has gone on and his popularity has grown, it's become nothing short of miraculous that he's kept up the mystery. Still, over the years, some of his digital curtain has begun to part, largely spurred

by his being doxed in 2017, when his identity was revealed to supposedly be that of a man named Paul. The character of Drill is fluid, but taken as a whole, the blurry image starts to come into focus. It's that of an easily agitated, overly confident, wildly crass, IBS-ridden middle-aged man thrashing away on a computer, probably a PC. He speaks in outlandish non-sequiters and engages with brands with unreasonable love and hate in equal measure. He is the dark democratic promise of the internet, that anyone can use it to broadcast their opinions at any time, fulfilled. I just go back to how specific and unique it is, comedian and actor David Cross told me, there's nobody quite like him, and it perfectly encapsulates

that Twitter dialogue. When I asked Dockney whether the type of person he's satirizing is more common now or merely more visible, he said, I think there's just so many of those minds out of there that we can only see because of the internet. In the 1920s or whatever, there were just as many dumb, crazy people who only met like four people in their entire life and just died in obscurity. He noted that he appreciates that the site records this interesting snapshot of all the insane people who exist in the background and just post. It was one of the only moments when he had anything remotely complimentary to say about Twitter. End quote. A chunk of the article is about Doc Nade trying to find his place in the arts and entertainment realm. He's self-published books and

had an adult swim show. The article talks about how people are still figuring out how social media works in the artistic realm. I would argue that a lot of people like Drill and younger folks have figured out how art works on social media. The professional entertainment and marketing worlds just haven't figured it out yet. So you get a lot of real weird artistic geniuses, kind of languishing away in obscure corners online. You know, I like Dockney's comparison to how there were kind of out there folks like his drill character in every era before social media, because I think that applies to Dockney himself.

A lot of the artists making some absolutely bizarre stuff online right now might have applied that genius to more traditional media in the past. Like, you want to tell me Da Vinci wouldn't have made the most off-the-wall memes and animations. We just have different tools now. Anyways, next long-read suggestion, dovetailing a bit from the AWS news today, from Sharon Goldman at VentureBeat. With a wave of new LLMs, open source AI is having a moment and a red-hot debate. Now first, this commentary from Brian, quote, open source is having a moment with the rise of the LLMs. It's literally like the dream of the 90s is alive in AI. But there's a larger question here, are LLMs going to turn this AI stuff into commodities and will open source

get us to that place? And if we get there, is that good or bad? End quote. So quoting from VentureBeat, the latest wave of open source LLMs are much smaller and not as cutting edge as chat GPT, but they get the job done, said Simon Willison, an open source developer and co-creator of Django, a free and open source Python-based web framework. Before Lama came along, I think lots of people thought that in order to run a language model that was of any use at all, you needed $16,000 worth of video cards and a stack of 100 GPUs, he told Venture Beat. So the only way to access these models was through OpenAI or other organizations. But now, he explained, open source LLMs can run on a laptop.

It turns out maybe we don't need the cutting edge for a lot of things, he said. However, the ethical implications of using these open source LLMs are complicated and difficult to navigate. said Willison. Open AI, for example, has extra filters and rules in place to prevent writing things like a Hitler manifesto, he explained. And it's concerning because it opens the door to harmful content creation at scale. Willison pointed to romance scams as an example. Now with language models, scammers could potentially use them to convince people to fall in love and steal their money on a massive scale, he said. Currently, Willison said he leans towards open source AI, but he still expressed concern. What if I'm wrong, he said. What if the risks of misuse

outweigh the benefits of openness? It's difficult to balance the pros and cons, end quote. And finally today from Vulture, is the economist making the best news podcasts right now? No, of course not. Ride Home makes the best news podcast, at least in the world of tech. From the article, quote, the economist's growing audio portfolio primarily consists of several regularly publishing news shows that span the publication's range of state interests. None are particularly groundbreaking on paper. In fact, they're almost throwbacks, somewhat devoid of the affable presentation and emotionality of modern American news podcasts. They are incredibly British, which is to say genteel, studious, a little awkward, and dry as sand. They also seem to be catching on. A spokesperson for

the economist tells me that its podcasts averaged around 25 million downloads in March. That's still a fraction of what you'd expect from a heavyweight like, say, the New York Times, $125 million, according to the industry measurement firm pod track. But the larger story here is one of growth. The network has more than doubled its unique listenership, another salient audience metric over the three years, from around $2 million in 2019 to about $5 million today. The momentum is inspiring further investment. This is our fastest growing platform, said Bob Cohn, the economist's president, over email. We're thinking hard about how to keep adding to the portfolio, likely in the back half of this year, end quote. This makes for a striking contrast with other parts of the industry,

which has been facing cuts and show cancellations in the midst of navigating a difficult economic environment. The Economist's history with podcasts extends far beyond this contemporary iteration. The August British magazine was actually part of the original wave of hype surrounding the medium's inception in the 2000s, back when many observers thought that all this podcast stuff was going to be the next major frontier in media. That prognostication wouldn't really be validated until another few hype cycles, of course. Podcasts are always the next big thing, just as podcasting has always been dying. The Economist's modern podcasting era earnestly began in 2019 with the launch of the intelligence.

Like the publication as a whole, its entry into the daily news genre is heavy on analysis and proudly esoteric. There's something distinctly appealing about the somewhat unmodern nature of the Economist's podcasts. This might simply have to do with how it breaks from much of the contemporary news podcast environment, which tends to interpret the podcast medium's off-touted intimacy by collapsing the gap between the show and the listener, you're conditioned to develop a feeling of closeness with a daily news podcast and its hosts. In style, there's often a lot of handholding, informationally and emotionally. This framework of news communication is understandable within the modern American context, reflecting a public service-driven desire to meet as many listeners as possible on their level.

But there are ways in which it can sometimes be off-putting, even condescending. In contrast, the Economist's podcasts are pleasantly aloof, even indifferent. Rather than broadly calibrating for whatever common-denominating level of knowledge listeners might have with the subject matter, they often project a sense that it's less about them meeting people where they're at than it is about people meeting them at theirs, end quote. As a fan generally of the Economist and particularly their podcasts as a free. alternative to subscribing to their text articles, I think Nicholas Qua at Vulture gets it exactly right. I started reading The Economist in high school because it felt challenging to me. I had to look things up as I read. Of course there is something to be said for meeting people where they're at, but not all media available to us needs to do that. And I'm grateful for the ones that push us a little further. Well, that is it from me for this week. Thanks to all of you for tuning in and those of you who sent some words of support this week. That was very nice to see.

I've had a great time filling in for Brian. He will be back in your feeds this Monday, April 17th. Thanks for having me and have a great weekend.