Tech Brew Ride Home - Fri. 07/28 – Is The Clock Ticking For Threads?

Episode Date: July 28, 2023

The whole Threads saga has been a whirlwind. At the beginning of the month I asked if Threads had already won. At the end of the month, I’m wondering if the clock is ticking in terms of their chances of survival. Generative AI but for robots. Again. Real robots. Are VCs pulling back from the crypto space? And, of course, the Weekend Longreads suggestions. Sponsors: Nutrafol.com/men code ridehome Crashplan.com code techmeme Links: Meta plans retention 'hooks' for Threads as more than half of users leave app (Reuters) Aided by A.I. Language Models, Google’s Robots Are Getting Smart (NYTimes) App Store to require developers to describe why their apps use certain APIs (9to5Mac) Sequoia Capital Slashes Crypto Fund as It Downsizes Amid Startup Crunch (WSJ) Weekend Longreads Suggestions: Large language models, explained with a minimum of math and jargon (UnderstandingAI.org) The making of ‘Acquired,’ the No. 1 tech podcast sensation (Fast Company) Internet cafes introduced Uganda to the internet (Rest of World) 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 time. Tech me right home for Friday, July 28th, 2020. I'm Brian McCullough today. The whole Threads saga has been a whirlwind. At the beginning of the month, I asked if Threads had already won. And now at the end of the month,

I'm wondering if the clock is ticking in terms of their chances of survival. Generative AI. But for robots, again, real robots. Are VCs pulling back from the crypto space and, of course, the weekend long-read suggestions. Here's what you missed today in the world of tech. How's Threads doing? Part 11? Yeah, it seems even Zuck is aware. that it remains undercooked, and if it remains that way too much longer, it's going to be a problem increasingly. At an internal town hall, meta-exexed said meta is looking to add more retention driving hooks to threads, like showing Instagram users important threads. Though, again, desktop, DMs, anything, hashtags? Maybe that sooner rather than later, quoting Reuters.

Meta Platforms executives are heavily focused on boosting retention for their new Twitter rival threads. After the app lost more than half of its users in the weeks following its buzzy launch, CEO Mark Zuckerberg told employees on Thursday. Retention of users on the text base app was better than executives had expected, though it was not perfect, said Zuckerberg, speaking at an internal company town hall, the audio of which was heard by Reuters. Obviously, if you have more than 100 million people sign up, ideally it would be awesome if all of them or even half of them stuck around were not there yet, he said. Zuckerberg said he considered the drop-off normal and expected retention to grow as the company adds more features to the app,

including a desktop version and search functionality. Meta is looking at adding more retention driving hooks to entice users to return to the app, like, quote, making sure people who are on the Instagram app can see important threads, said chief product officer Chris Cox, end quote. Jackbots are cool, image generating AI is cool, but what about an LLM that enables robots to exist in the real world, in real meatspace. Google has launched RT2 or Robotics Transformer 2, a, quote, vision-language action model trained on text and images from the web that can output robotic actions again in the real world.

Quoting the New York Times. A quiet revolution is underway in robotics, one that piggybacks on recent advances in so-called large language models, the same type of artificial intelligence system that powers chat GPT, bard, and other chatbots. Google has recently begun plugging state-of-the-art language models into its robots, giving them the equivalent of artificial brains. The secretive project has made the robots far smarter and given them new powers of understanding and problem-solving. I got a glimpse of that progress during a private demonstration of Google's latest robotics model

called RT2. The model, which is being unveiled on Friday, amounts to a first step toward what Google executives described as a major leap in the way robots are built and programmed. A one-armed robot stood in the front of a table. On the table sat three plastic figurines, a lion, a whale, and a dinosaur. An engineer gave the robot an instruction, pick up the extinct animal. The robot word for a moment, then its arm extended and its claw opened and descended. It grabbed the dinosaur. Until very recently, this demonstration, which I witnessed during a podcast interview at Google's Robotics Division in Mountain View, California last week, would have been impossible. Robots weren't able to reliably manipulate objects they had never seen before, and they certainly weren't capable of making the logical leap from extinct animal to plastic dinosaur.

To understand the magnitude of this, it helps to know a little bit about how robots have conventionally been built. For years, the way engineers at Google and other companies trained robots to do a mechanical task, say flipping a burger, for example, was by programming them with a specific list of instructions. Lower the spatula, 6.5 inches, slide it forward until it encounters resistance, raise it 4.2 inches, rotate it 180 degrees, and so on. Robots would then practice the task again and again, with engineers tweaking the instructions each time until they got it right. This approach worked for certain limited uses, but training robots this way is slow and labor intensive. It requires collecting lots of data from real-world tests. And if you wanted to teach a robot to do

something new, to flip a pancake instead of a burger, say, you usually had to reprogram it from scratch. In recent years, researchers at Google had an idea. What if, instead of being programmed for specific tasks one by one, robots could use an AI language model, one that had been trained on vast swaths of internet texts to learn new skills for themselves. Google's new robotics model, RT2, can do just that. It's what the company calls a vision language action model or an AI system that has the ability, not just to see and analyze the world around it, but to tell a robot how to move. It does so by translating the robot's movements into a series of numbers, a process called tokenizing and incorporating these tokens into the same training data as the language model.

Eventually, just as chatyBT or Bard learns how to guess what words should come next in a poem or a history essay, RT2 can learn to guess how a robot's arm should move to pick up a ball or throw. an empty soda can into the recycling bin. In other words, this model can learn to speak robot, Mr. Hausman said, end quote. This is either extremely impressive and exciting or extremely worrying. I can never tell anymore. Apple plans to require developers to submit reasons to use certain APIs going forward in their apps, beginning this fall, apparently to prevent the misuse of those APIs for fingerprinting purposes.

Quoting 9 to 5 Mac. As detailed on the Apple developer website, some APIs are now classified as required reason APIs. This means that in order to use them in an app, the developer must describe to Apple the purpose of that API in the app. The company explains that the measure aims to crack down on fingerprinting a technique for tracking users across different apps and websites. Starting this fall with the release of iOS 17, TVOS 17, WatchOS 10, and MacOS Sonoma to the public, developers will be notified about submitting apps using a required reason reason. API without describing the reasons for using it. From Spring 2024, apps that use these APIs without a valid reason will be rejected. To prevent the misuse of certain APIs that can be used

to collect data about users' devices through fingerprinting, you'll need to declare the reasons for using these APIs in your app's privacy manifest. This will help ensure that apps only use these APIs for their intended purpose, Apple explains. While this measure was created with privacy in mind, some developers told 9 to 5 Mac they're concerned about app and up date rejection rates increasing further. For example, Apple says that user defaults is one of the required reason APIs. For those unfamiliar, this is a basic and fairly common API that stores user preferences for an app, which means lots of apps use it. This can result in developers having their apps rejected simply for forgetting to add an explanation for using the API. At the same

time, it's hard to imagine how Apple will control the use of this API, since most developers can simply say they're storing user preferences with it. Apple will let developers appeal a rejection and submit a request to approve a situation that is not covered in the current guidelines. More details can be found on the Apple developer website, end quote. Sources say Sequoia cut the size of its cryptocurrency fund to $200 million from $585 million. They also cut their ecosystem fund, which invests in other funds, to $450 million from $900 million, quoting the Wall Street Journal. Sequoia told fund investors, in March it made the decision to reduce the funds

to better reflect the changed market. The cryptocurrency fund, for example, will focus more on backing young startups after an industry crash wiped out opportunities to back larger companies. By pairing back the fund sizes, Sequoia is lowering the amount of committed capital required from investors, known as limited partners who are already seeing lower returns from venture funds and are bracing for further markdowns. The changes show the difficult cuts venture firms are making during one of the roughest years and recent memory for the startup industry. They are trying to undo the breakneck expansion and liberal spending that characterized a historic startup boom, which no longer makes sense as dealmaking slows and funds struggle to raise more cash.

Sequoia announced the two funds in February 2022 as part of an ambitious firm restructuring after spending months ramping up its investments in crypto. The crypto crash has since wiped out revenue for many blockchain startups, end quote. It is part of my job here to Trendspot for you all. And look, Sequoia is going through major, major generational transformation of a It's very structure at this moment, so take this as a grain of salt anic data, but it is interesting to note that the area they want to cut back on first is crypto. Oh, and fund of funds that support emerging fund managers like myself, too. But also, I'm going to keep an eye open for how

investment money is or maybe is not continuing to flow into the crypto space. Time for the weekend long read suggestions. And first up, this week I stumbled across the best explainer for lay people I've found yet, an explainer of how large language models actually work. Like, seriously, if you are non-technical in the space and want the clearest walkthrough of how this works conceptually, try this piece from understanding AI. Quote, the above diagram depicts a purely hypothetical LLM, so don't take the details too seriously. We'll take a look at research into real language models shortly. Real LLMs tend to have a lot more than two layers. The most powerful version of GPT3, for example, has 96 layers. Research suggests that the first few layers focus on understanding

the syntax of the sentence and resolving ambiguities like we've shown above. Later layers, which we're not showing to keep the diagram a manageable size, work to develop a high-level understanding of the passage as a whole. For example, as an LLM reads through a short story, it appears to keep track of a variety of information about this story's characters. sex and age, relationships with other characters, past and current location, personalities and goals, and so forth. Researchers don't understand exactly how LLMs keep track of this information, but logically speaking, the model must be doing it by modifying the hidden state vectors as they get passed from one layer to the next. It helps that in modern LLMs, these vectors are extremely

large. For example, the most powerful version of GPT3 uses word vectors with 12,288 dimensions. That is, each word is represented by a list of 12,288 numbers. That's 20 times larger than Google's 2013 Word2-Vec scheme. You can think of all of these extra dimensions as a kind of scratch space that GPT3 can use to write notes to itself about the context of each word. Notes made by earlier layers can be read and modified by later layers, allowing the model to gradually sharpen its understanding of the passage as a whole. So suppose we changed our diagram above to depict a 96-layer language model interpreting a thousand-word story. The 60th layer might include a vector for John with a parenthetical comment like main character, male, married to Cheryl, cousin of Donald from Minnesota, currently in

Boise, trying to find his missing wallet. Again, all of these facts and probably a lot more, would somehow be encoded as a list of 12,288 numbers corresponding to the word John. Or perhaps some of this information might be encoded in the 12,288 dimensional vectors for Cheryl Donald Boise Wallet or other words in the story. The goal is for the 96th and final layer of the network to output a hidden state for the final word that includes all of the information necessary to predict the next word, end quote. Then Fast Company has a profile of my friends over at the Acquired podcast. You've heard the Acquired guys. They've been on this show for our World Cup of Entrepreneurs bonus episodes. I've been on their show a couple times. But here's a great behind the scenes of

how and why their pod has gotten so popular. Quote, the research effort to produce an episode of Acquired wasn't always this Herculian. In January 2020, the duo told the Divinations newsletter writer Nathan Batchez that prep typically requires five to ten hours per episode. Two and a half years later, investor and podcaster Logan Bartlett asked about what goes into an episode, and Gilbert and Rosenthal replied that they cumulatively spend a hundred hours. Now, a year later, the host seemed almost consumed by the process. process, logging 60 hours a week for a month. The night before recording the Nike episode,

Rosenthal, says, my wife, Jenny, was like, I get my husband back tomorrow night. Gilbert chimes in, yeah, my wife feels that way, too. When asked if there's an upper limit to the amount of work that goes into an episode, Rosenthal admits, I've literally had this discussion over the past month with my therapist. Let's pause the flag that we haven't yet mentioned Rosenthal's 39-page Nike script that he produced from all that research, or Gilbert's list of insights. He's curated to bring up during taping that runs almost 4,000 words. They both stand during what can now be an eight-hour taping session with Gilbert wearing his cushiest Nike Invincible Two running shoes, each of them powering through on an occasional bite of an energy bar, a kind five grams of sugar

bar for Gilbert, a Met RX, 100 for Rosenthal. An editor helps whittle down the recording to about four and a half hours. Then Gilbert's self-described persnicketyness, along with some bigger picture suggestions from Rosenthal, further refine the episode to four hours, three minutes, 28 seconds, end quote. Finally, from rest of world, a profile of the vanishing world of the internet cafe. History will look back at internet cafes as a key vector for giving millions of people around the world their first access to the online world. But now that digital data is ubiquitous, even in the most remote corners of the world, do internet cafes still have a place? quote, Derek Buchena, now nearly 40 years old, has spent most of the last two decades around computers.

In the early 2000s, computers were very rare in Uganda, as were opportunities to learn how to use them. But the church Buchania attended announced free computer classes in partnership with Zion Internet Cafe, one of the first and most prominent internet cafes in Uganda, located in Kampala's central business district. In 2014, Buchania launched his own internet cafe, BK Internet Cafe, in a low-income Kampala suburb. At the time, the Uganda Communications Commission estimated that Uganda had about 10 million internet users representing roughly a third of the country's population. When I launched my cafe, the business was booming, Buchania said. He launched four more cafes in the neighborhood and remembers them being packed with young people. Facebook was the go-to social network. Young people could spend

hours just chatting and catching up with friends, he said. Then also Google had taken a foothold, so our customers were just searching for anything. YouTube was prohibited because it would consume a lot of data and the internet speeds weren't great anyway, end quote. Bukenia said the industry shifted in 2016. Cheap Chinese smartphones suddenly became widely available across Uganda. All of the sudden, one could get a good smartphone for less than 500,000 shillings or 170, he said. In their wake, telecom providers launched cheap internet and data bundles. Over the next three years, internet cafes in the country started to close down. By the time COVID-19 came around, I was the only internet cafe left standing, said Buchania, who has since diversified his revenue. He now makes the most money

selling movies and offering printing, photocopying, and scanning services. Buchania admits the internet part of his cafe no longer attracts users. Some days, no one uses the computers at all. I don't think it makes sense to have a business that serves very few users, he said. Printing, scanning, and photocopying documents sustain us for now, but I am thinking of pivoting to a co-working space to tap into the remote workers, end quote. No week on bonus episodes for you this weekend, but I think we might have one next week. Talk to you on Monday.