Tech Brew Ride Home - Fri. 01/28 – Making GPT-3 Less Gross

Episode Date: January 28, 2022

Ho hum. Apple reports its most revenue and profit in history. You’ll soon be able to unlock your iPhone while masked, even without the Apple Watch. The White House is readying a huge executive order on Crypto. The new version of GPT-3 is less error-prone, and less prone to say bad stuff. And of course, the weekend longreads suggestions. Sponsors: EditorX.com Wealthfront.com/techmeme Links: Apple CEO: ‘We Don’t Make Purely Financial Decisions’ About Apple TV Plus Content (Variety) iOS 15.4 Beta Lets You Use Face ID With a Mask On (MacRumors) White House reportedly preparing executive order on crypto (CoinTelegraph) The new version of GPT-3 is much better behaved (and should be less toxic) (Technology Review) Weekend Longreads Suggestions: Line Goes Up – The Problem With NFTs (Folding Ideas on YouTube) Three things web3 should fix in 2022 (Platformer) The NFT Art World Wouldn’t Be the Same Without This Woman’s ‘Wide-Awake Hallucinations’ (RollingStone) Nothing Sacred: These Apps Reserve The Right To Sell Your Prayers (Buzzfeed News) Your Hotel Concierge Is Probably a Texting Robot. Srsly. (WSJ) Why Netflix Just Had Its Worst Day in A Decade (Bloomberg) Netflix stock market woe is warning to Hollywood (Financial Times) Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to the Tech meme right home for Friday, January 28th, 2022. I'm Brian McCullough today. Ho-hum. Apple reports its most revenue and profit in history. You'll soon be able to unlock your iPhone while masked, even without the Apple Watch. The White House is readying a huge executive order on crypto. The new version of GPT3 is less error-prone and less prone to say bad stuff.

And of course, the weekend long read suggestions. Here's what you miss today in the world of tech. Let's end this week by blazing through Apple earnings. Oh, you know, they just reported the highest revenue and income in company history. Apple reported Q1 revenue of $123.9 billion, up 11% year-over-year. Net income of $34.6 billion up from $28.8 year-over-year. Services of $19.5 billion, up from $15.8. iPhone sales of $71.6 billion. Interestingly, Mac revenue was up 25% year over year. Apple says the last six quarters have been the best ever for the Mac, period. People, I guess, are liking those new Apple Silicon chips,

so much so that maybe they're eschewing iPads. iPad sales were down 14%, though that could be due to supply chain issues. I know I ordered Penny a new iPad at the end of December, and it still hasn't even shipped. but the other products revenue where the Apple Watch lives was up 13% year over year. What blows my mind there is that Apple says two-thirds of all Apple Watch sales last quarter were to new users, people new to the product. Apple also had 785 million paying subscribers globally across all services in Q1, up 27% year over year. Maybe the biggest headline was that Apple says it now has over 1.8,000,000,000,000.

billion active devices worldwide, up from 1.65 billion in January of 2021 and 1.5 billion in January 2020. As Horace Deju pointed out, Apple has been adding 3 million active devices every week for over a decade. By June 2023, there will be 2 billion Apple devices in use. You can mark that in your calendar, end quote. And one more thing. On my beloved The Watch podcast, which covers streaming and the market generally. They were pointing out that we might look back on 2022 as the year that Apple TV Plus broke through. Apple TV Plus apparently has a metric ton of programming coming out this year. That old argument that Apple just had a handful of shows on Apple TV Plus and it looked paltry next to the content library of others. That whole argument might be dead, quoting variety.

We don't make purely financial decisions about the content on Apple TV Plus. We try to find great content that has a reason for being. Apple CEO Tim Cook said on Apple's quarterly earnings call on Thursday, Cook was responding to an analyst question about whether the service's socially responsible programming lens might be causing Apple to be hesitant about acquiring a studio. Apple does not break out subscriber numbers for Apple TV Plus, but as part of discussing its blowout year-end 2021 quarter results, the company said it had 785 million paying subscribers globally across all services. That includes both Apple branded and third. third-party services such as subscriptions sold through the Apple App Store. Apple's services

segment, which includes the App Store, ICloud, and Apple TV Plus, as well as music, news, games, fitness payments, and other services generated a record $19.5 billion in the year and quarter, up 24% from the year earlier period, end quote. Maybe those new John Hamm ads are doing something after all. P.S. Apple also released iOS 15.4 beta, which, adds face ID use with a mask for iPhone 12s and newer. And this is face ID without needing Apple Watch authentication even when you're wearing a mask because apparently it can recognize, quote, unique features around the eye, quoting Mac rumors. If you opt to use this feature during setup, you will need to re-scan your face for face ID. From there, face ID will be able to unlock

your iPhone even when you're wearing a mask. In the Settings app, there's a new use face ID with a mask toggle that can be turned on or off if you change your mind about the mask. And there's a new feature to add glasses to make face ID more accurate when you're wearing glasses and a mask at the same time. Apple warns in the settings app that face ID is most accurate when it is set up for full face recognition only. For face ID with a mask to work, you must be looking at your device to get it to unlock and it does not work when wearing sunglasses. Face ID with a mask can authenticate Apple Pay payments, and it can be used in lieu of a login and password in apps that support Face ID, unlike the prior Apple Watch face ID feature, end quote.

A source is telling Barron's that the White House is readying an executive action to task federal agencies with regulating Bitcoin and other cryptocurrencies as a matter of national security, quoting Coin Telegraph. Joe Biden's White House is expected to issue an executive order in the coming weeks about actions the United States government will take regarding digital assets. A source familiar with the White House's plan told Barron's that the executive order would be issued in a national security memorandum. Biden's memorandum would assign some government entities to study crypto, stable coins, and non-fundable tokens with the goal of developing a workable regulatory framework. This is designed to look holistically at digital assets and develop a set of policies

that give coherency to what the government is trying to do in the space, end quote. Rumors of a potential executive order on crypto have been swirling in recent days. Earlier this week, Forbes reported those government entities would likely issue reports on their findings by mid-2020, after having looked at, quote, the systemic risks of cryptocurrencies and their illicit uses, end quote. The rationale for the executive order falling under national security is that crypto is a cross-border tool for shifting money, the ability of decentralized blockchain technology to circumvent geospecific surveillance or rules, means the administration may push for synchronized international regulations with other countries,

end quote. On the heels of that big raise this week, OpenAI says that instruct GPT, its new version of GPT3, is better at following instructions, producing less offensive language, less misinformation, and fewer mistakes, quoting MIT Technology Review. OpenAI has built a new version of GPT3, its game-changing language model, that it says, does away with some of the most toxic issues that plagued its predecessor. The San Francisco-based lab says the updated model called Instruct GPT is better at following the instructions of people using it known as alignment in AI jargon, and thus produces less

offensive language, less misinformation, and fewer mistakes overall, unless explicitly told not to do so. Large language models like GPT3 are trained using vast bodies of text, much of it taken from the internet, in which they encounter the best and the worst. of what people put down in words. That is a problem for today's chatbots and text generation tools. The models soak up toxic language from texts that is racist or misogynistic, or that contains more insidious baked in prejudices as well as falsehoods. OpenAI has made instruct GPT the default model for users of its application programming interface,

a service that gives access to the company's language models for a fee. GPT3 will still be available, but OpenAI does not recommend using it, quote, It's the first time these alignment techniques are being applied to a real product, says Jan Leakey, who co-leads OpenAI's alignment team. Previous attempts to tackle the problem included filtering out offensive language from the training set, but that can make models perform less well, especially in cases where the training data is already sparse, such as text from minority groups. To train, instruct GPT, OpenAI hired 40 people to rate GPT3's responses to a range of pre-written prompts,

such as write a story about a wise frog called Julius or write a creative ad for the following product to run on Facebook. Responses that they judged to be more in line with the apparent intention of the prompt writer were scored higher. Responses that contained sexual or violent language, denigrated a specific group of people, expressed an opinion, and so on, were marked down. This feedback was then used as the reward in a reinforcement learning algorithm that trained instruct GPT to match responses to prompts in ways that the judges preferred. OpenAI found that users of its API favored instruct GPT over GPT3 more than 70% of the time. We're no longer seeing grammatical errors in language generation, says Ben Roe, head of product at Yable, a market research company that

uses OpenAI's models to create natural language summaries of its client's business data. There's also clear progress in the new model's ability to understand and follow instructions end quote. Time for the weekend long read suggestions and first up a YouTube video. This one has been making the rounds. You might have seen it, but since I've been getting so much feedback from listeners that I should include Web3 skepticism along with all of the Web3 stories we've been doing lately. Well, actually, we're going to do more of that on the bonus episode tomorrow, but let me quote Casey Newton from his newsletter. Last week, it felt like everyone I knew was sending me the same link. The Problem with NFTs, a long video essay by the Canadian media critic Dan Olson, ricocheted around all corners

of the tech world since it was uploaded on Friday. It now has 2.6 million views and climbing. Over 138 meticulously researched minutes, Olson traces the history of the 2008 financial crisis, the creation of Bitcoin and Ethereum, and the rise of NFTs and Dow's and reaches the conclusion that what we have taken to calling Web3 is effectively beyond saving. The technology is too broken and its creators too indifferent to its failures for it to ever live up to the promise of its most starry-eyed backers. Few of Olson's criticisms are entirely new and on my Twitter timeline this week, I saw many crypto enthusiasts dismiss them out of hand. Few people working in the space will be surprised to learn that crypto three is a wash in grifts, that current blockchains are energy

inefficient and expensive, or that digital wallets are difficult to use and fraught with danger. Many Web3 builders were also bristle at Olson's tone, which is smug and hectoring in the house style of the YouTube video essayists. His audience is not people working in crypto, but rather everyone he thinks ought to be afraid of those people. And yet the collective force of Olson's arguments is substantial. His essay explains the rise of cryptocurrencies through the lens of rising inequality, pandemic-era isolation and loneliness, self-dealing venture capitalists, and a desperate sense among young strivers that the future is only ever getting smaller, all of which feels particularly timely given this week's crash in crypto prices, end quote.

So if you're among those who think I've been to Web 3 booster-ish lately and want a critique to balance things out, there you go. Rolling Stone answers a big question that I've had, like, who design those bored apes? Who is the artist behind it? Well, according to this Rolling Stone profile, it turns out that the artist herself wasn't quite aware either. The woman who drew the board apes primary characters had no idea that the collection was a hit until she Googled the name months later. When Seneca logged onto Twitter, where she's known as All Seeing Seneca, and saw that Steph Curry was using an avatar she birthed as his profile picture, her eyes bulged. It really took me some time to wrap my head around all this, she tells Rolling Stone over Zoom. I still am.

It's still quite surreal. To be clear, Seneca was not the project's sole illustrator. I am the lead artist behind the original collection, she says. The ape body itself, she adds, is, quote, exactly line for line, her drawing. Other production artists, Thomas Dagley, MIG was here, and a couple who choose to remain anonymous, according to Gargamel handled the traits and the environment. However, she points out she did develop some of the major traits like the grinning mouth, the popping eyes, and the beanie. Not a ton of people know that I did the drawings, which is terrible for an artist, she says. Word of mouth has been growing, though, and she hopes that will help her find more collaborations. In the meantime, she's focusing on her solo work, end quote. Then BuzzFeed takes a look at the sudden VC excitement to invest in Christian worship apps. But the question is, if user data is often a commodity for apps like this, is it possible or advisable or even moral to monetize people's prayers? Quote, thanks to pray.com, Katie found solace in a community of what she calls prayer warriors. Thousands of people who share their deepest fears and hopes in a public Facebook-esque feed of prayers and prayer requests.

Katie posted prayers for her son, for her former husband, who died by suicide in 2008, and for her youngest child who struggles with addiction. All apps' usernames have been changed to protect their privacy, by the way. As Katie laid out her spiritual anguish, pray.com was data mining it, matching her actions in the app to details about her that it purchased from data brokers. Prey.com collects data about its users in multiple ways. According to its privacy policy, the company records detailed information about users, including their physical location, the links they click on, and the text of the posts they make. Then it supplements that information with data from third parties such as data analytics

providers and data brokers, which can include your gender, age, religious affiliation, ethnicity, marital status, household size, and income, political party affiliation, and interest, geographic location, and personal information, end quote. The policy also says pray.com shares users' personal information, including identifiers that link their activity to specific devices with, quote, third parties for, quote, commercial purposes, end quote. Then the Wall Street Journal looks at why increasingly your hotel concierge is probably now a bot.

If they're not a bot, they're somebody maybe thousands of miles away, quote, the Lowe's uses off-site employees 2,000 miles away to answer texts or phone calls, says manager Laura Lowell, who oversees a staff of 14 agents at the company's Engagement Center in Franklin, Tennessee. Most guests never think to ask who is answering them, she says. These staffers feel common requests, including housekeeping or restaurant reservations within minutes, without involving the hotel's concierge team. The team isn't quick to share their actual location. We always say we're an extension of the front desk, she says. The volume of texts sent via Marriott's app tripled in 2020.

from the year prior, with some hotels needing to reshuffle staff duties to meet the needs of guests, Ms. Edmondson says. For the past six years, Marriott brands, including Moxie Hotels and Weston, have offered guests the option to use the Marriott Bonvoy app to communicate directly with hotel staff. Text replies come from a mix of hotel staffers and automated responses based on the request, end quote. And finally, two pieces looking at the state of the streaming wars, specifically with regards to Netflix, but this obviously has implications for everybody. Netflix is among the stocks that is down significantly from all-time highs recently, and in his Bloomberg newsletter this week, Lucas Shaw looks into why Netflix had its

worst day on the stock market in a decade. Turns out investors are increasingly worried about the very thing we've spoken about before about the economics of streaming. Remember when Peter Kafka warned us about that, that no one was 100% sure that the numbers would math out to replace the revenue. that streaming disrupts for Hollywood. I want to pair that with a more targeted piece from the Financial Times, which I don't usually share FT stories because they have such a hard paywall, but if you can manage to get through to this piece, quote. So even with 222 million subscribers,

Netflix has to keep spending a lot to satiate us. Last year, Netflix posted negative free cash flow, the money left over after paying operating expenses and capital expenditure, of a $159 million on $30 billion of revenue. Moffat Nathanson projects that for 2022, Netflix revenue will rise 13% to more than $33 billion, but expenses will increase 15% to $27 billion, $20 billion of this on content. This results in net income of $5 billion down about 3% from 2021. So Netflix makes a lot of revenue, but much less cash flow profit because its costs are high. In this sense, Netflix's long-term business might look more like a telco than a tech company. The biggest U.S. telecoms groups have to invest heavily to offer the best service,

while also keeping prices low because they are effectively selling the same product as their competitors. Netflix was in its own category for a while. Now it is surrounded by competition. The stock market has spent the past decade rewarding Netflix for its vision and growth. For a long time, that made old studio executives angry. But what Netflix results have shown us recently is that something more fundamentally scary for Hollywood might be happening. Streaming television is going to make less money, maybe a lot less money for entertainment companies than cable did. They are replacing a great model for a less great model, says Nathanson, by a mile, end quote.

This weekend, the bonus episode will be the Twitter space from last night, featuring three of the best venture and startup reporters in the business on the turmoil in the stockpour. markets recently and how that could affect the startup ecosystem. One of the biggest panels we've ever done, I think. Also, because you asked for it, Chris and I debate the counter arguments to all of the Web3 hype. Enjoy that. Talk to you on Monday.