Tech Brew Ride Home - Wed. 03/15 – GPT-4 Is Here

Starting point is 00:00:00 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 Wednesday, March 15th, 2023. I'm Brian McCullough today. GPT4 is unveiled and it's multimodal and man is everybody rushing to make use of this stuff. Alleged renders of Google's Pixel 8 Pro appear to show a mysterious new sensor,

Starting point is 00:00:50 the state of the music industry in the streaming era, and Deadpool sells his phone company to another phone company. Here's what you miss today in the world of tech. It's here, everyone. OpenAI debuted GPT4 yesterday. claiming the new model, quote, surpasses chat GPT and its advanced reasoning capabilities. It's available in chat GPT plus right now and as an API that has a wait list. So what's new here? Multimodal, baby, quoting Windows Central. One of the big new additions with GPT4 is its ability to accept images in addition to text as an input method.

Starting point is 00:01:33 Now the AI will be able to analyze images and output answers via text. ChatGPT Plus users will be able to access these new GPT4 capabilities starting today, and it's also available as an API for developers. GPT4 leverages more data and more computation that results in a more sophisticated and capable language model. It can now handle over 25,000 words of text, which means it can handle longer context queries and conversations without getting lost or confused. OpenAI says GPT4 was trained on Microsoft Azure AI supercomputers with Azure's AI

Starting point is 00:02:07 optimize infrastructure, which allows the company to deliver GPT4 to users around the world. It is also supposedly safer, as it's 82% less likely to respond to requests for disallowed content, and 40% more likely to produce factual responses, according to OpenAI. Of course, as was the case with GPT3.5, GPT4 also has limitations that OpenAI is still working to address. Issues such as social biases, hallucinations, and adversarial prompts are still occurrences. that may be presented by the AI win-in-use. OpenAI also says GPT-4 scores in higher approximate percentiles among test takers versus GPT3.5. Chat GPT-plus users can try GPT4 now, and developers can join a waitlist to begin implementing this technology into their own applications and services, end quote.

Starting point is 00:02:57 More on that multi-modality from the New York Times, quote, GPT4 has a new ability to respond to images as well as text. Greg Brockman, OpenAI's president and co-founder, demonstrated how the system could describe an image from the Hubble Space Telescope in painstaking detail. The description went on for paragraphs. It can also answer questions about an image. If given a photograph of the inside of a fridge, it can suggest a few meals to make from what's on hand, end quote. And this is interesting. OpenAI also open-sourced e-vals. Its framework for automatically evaluating the performance of its AI models, letting users report shortcomings and guide improvements. Quoting TechCrunch, Open AI created Evales to develop and run benchmarks for evaluating models like GPT4 while inspecting

Starting point is 00:03:45 their performance. With Evales, developers can use datasets to generate prompts, measure the quality of completions provided by an Open AI model, and compare performance across different data sets and models. Evals, which is compatible with several popular AI benchmarks, also supports writing new classes to implement custom evaluation logic. As an example to follow, OpenAI created a logic puzzle's evaluation that contains 10 prompts where GPT4 fails. It's all unpaid work, very unfortunately, but to incentivize Eval's usage, OpenAI plans to grant GPT4 access to those who contribute high-quality benchmarks. We believe that Eval's will be an integral part of the process for using and building on top of our models, and we welcome direct contributions, questions, and feedback, the company

Starting point is 00:04:30 wrote, end quote. Now, that's nowhere near all the news there is about this. As many folks speculated, Microsoft confirmed that its new Bing has been, quote, running on GPT4, which we've customized for search, their words, and was using an early version of the model over the past five weeks. But it's also amazing how fast people beyond Microsoft are rushing to apply this stuff. For example, Duo Lingo debuted a max subscription tier for $30 per month or $168 per year, which offers GPT4 features for English speakers taking Spanish and French courses on iOS, quoting TechCrunch. This unlocks two new AI-powered features called RollPlay and Explain My Answer.

Starting point is 00:05:21 The roleplay feature unlocks an AI chatbot that allows users to practice real-world conversation skills with characters in the app. Users will be chatting with an AI that is both responsive. and interactive thanks to GPT4. The feature guides users through different scenarios. For instance, you can pretend to order coffee at a cafe in Paris or discuss future vacation plans with a certain character in the app. After users are done with the conversation, they will get AI-powered feedback on the accuracy and complexity of their responses, as well as tips for future conversations. Duolingo says human experts write the scenarios that users see in role play and that they make sure

Starting point is 00:05:57 the initial prompt is aligned with where the user is in their course. The experts also write the initial message in the chat and tell the AI model where to take the conversation. The Explain My Answer feature gives users the chance to learn more about their response in a lesson, whether their answer was correct or incorrect. Users can tap a button after certain exercises to enter a chat with Duo to get an explanation on why their answer was right or wrong and ask for examples or further clarification. The company says the feature is designed to help you when you make a mistake and are unsure what the right answer would be. The feature can also help in instances where you keep making the same mistake again and again, and you're not sure why.

Starting point is 00:06:36 The only courses that can utilize these two new features are Spanish and French for English speakers on iOS. The company plans to roll out the new features to more courses and platforms soon, end quote. Also, PWC has announced a partnership with AI startup Harvey to launch a chatbot for its 4,000 lawyers, built using OpenAI's GPT4 as law firms rush to find ways to integrate AI into what they do. And then listen to this. The Be My Eyes app for the visually impaired has integrated GPT4, which can inspect and describe images through a virtual volunteer feature in closed beta. TechCrunch again.

Starting point is 00:07:15 We've written about Be My Eyes plenty of times since it was started in 2015, and of course the rise of computer vision and other tools has figured prominently in its story of helping the visually impaired more easily navigate everyday life. But the app itself can only do so much, and a core feature was always being able to get a helping hand from a volunteer who could look through your phone's camera view and give detailed descriptions or instructions. The new version of the app is the first to integrate GPT4's multimodal capability, which is to say its ability to not just chat intelligibly, but to inspect and understand images it's given. Users can send images via the app to an AI-powered virtual volunteer, which will answer any questions

Starting point is 00:07:54 about the image and provide instantaneous visual assistance for a wide variety of tasks. For example, if a user sends a picture of the inside of their refrigerator, the virtual volunteer will not only be able to correctly identify what's in it, but also extrapolate and analyze what can be prepared with those ingredients. The tool can also then offer a number of recipes for those ingredients and send a step-by-step guide on how to make them. But the video accompanying the description is more illuminating, by the way, click on the link because the video is in the piece. In it, be my eyes user Lucy shows the app helping her with a bunch of things live. If you're not familiar with the rapid

Starting point is 00:08:30 fire patois of a screen reader, you may miss some of the dialogue, but she has it described the look of a dress, identify a plant, read a map, translate a label, direct her to an unoccupied machine treadmill at the gym, and tell her which buttons to put. at a vending machine, end quote. And if that's not enough, also sort of lost in the hype around GPT4, Anthropic, which was co-founded by former OpenAI employees, debuted Claude yesterday, which is their chat GPT rival, which the startup says is less likely to hallucinate. Companies can request access to that right now, so let the generative AI wars begin, I guess. I'm still hearing lots of rumblings that something might happen with TikTok this year. And sources have told Bloomberg that

Starting point is 00:09:25 in light of that, TikTok's leadership is considering divesting itself from ByteDance as a sort of last resort if Sefius rejects the existing proposal on the table to distance the company from its Chinese owners, which would possibly lead to a TikTok IPO. Quote, a divestiture which could result in a sale or initial public offering is considered a last resort to be pursued only if the company's existing proposal with national security officials doesn't get approved, according to people familiar with the matter, who declined to be named discussing non-public information. Even then, the Chinese government would have to agree to such a transaction the people said. Ticktox U.S. business could be valued at $40 to $50 billion based on social media multiples and other factors, according to

Starting point is 00:10:10 Bloomberg intelligence analysts Mandip Singh and Damien Remertz. The company which is undergoing a National Security Review by the Committee on Foreign Investment in the United States, agreed last year to implement a number of changes in a plan it calls Project Texas. The proposal includes bringing in American Tech Giant Oracle to host U.S. user data and review its software and appointing a three-person government-approved oversight board. Many of the moves are already underway. But Seifius, which is a panel of multiple agencies involved in national security, has stalled in its review process, leaving TikTok unsure of whether its plans will be sufficient to continue operating in the U.S., the people said. Members of the committee from the Justice Department have been unwilling to accept TikTok's proposal,

Starting point is 00:10:53 according to people familiar with the matter. Neither a ban of TikTok nor a divestiture of TikTok from Bytance does anything to address national security concerns about data transfers, said Brooke Oberwetter, a spokesperson for TikTok. Under Project Texas, TikTok data for our U.S. users would be held to a significantly higher security standard than any comparable American company, end quote. Alleged renders of Google's forthcoming Pixel 8 Pro smartphone have surfaced, which appear to show a flat 6.52-inch display, and a new sensor on the camera bar, after also, by the way, a pixel fold and 7A retail listing also leaked, quoting 9 to 5 Google.

Starting point is 00:11:41 Courtesy of at-on leaks and SmartPree, this first look at the Google, Pixel 8 Pro in CAD-based leaks shows off a design that's quite similar to the Pixel 7 Pro, but with some obvious and welcome changes. For one, the phone appears to have finally ditched its curved display, at least mostly. The display lacks the signature curve we've seen on Google's past two pro releases, but there's certainly still some room for slight curves along the edges. Either way, it's a major improvement in our book, given the frustration that comes with curved displays. Notably, too, the display appears to have a much slimmer bezel all around compared to past Google phones, though with CAD-based renders, this is a detail that's easy to get wrong. We'll have to

Starting point is 00:12:21 see the final product to know for sure. But what's most intriguing here is the addition of a new sensor on the Google Pixel 8 Pro's camera bar. The phone has the same triple camera array as the Pixel 6 Pro and Pixel 7 Pro, though this time with all three sensors grouped into the same sizable cutout on the camera bar. But off to the right side, there's a fourth sensor in place. What this sensor does is unclear at the moment. It could be as simple as a better auto-focus system like what Samsung offers on the Galaxy S23 Ultra, a macro or depth sensor as many other phones offered today, or something more advanced. One possibility could be LiDAR as our Max Weinbach has heard in the past. The Pixel 8 Pro is expected to debut later this year, almost surely with a Tensor G3

Starting point is 00:13:04 processor along with it. It's unclear if the standard Pixel 8 would also include this mysterious new sensor, but the phone is expected to be smaller, end quote. You might have seen the headlines recently that sales of vinyl records have outpaced CD sales for the first time since 1987, but lost in those headlines is some additional data that I thought was useful to note in terms of the health of the music industry in the era of streaming, quoting 9 to 5 Mac. In total, the recording industry of America says that music revenue in the United States grew by 6% in 2022 to a new record high of $15.9 billion. This marks the six year in a row of growth for the industry. In total, streaming accounted for 84% of that revenue, followed by physical sales at 11%, digital downloads at 3%, and synchronization royalties at 2%. According to the RIA, the average number of streaming subscriptions for the year

Starting point is 00:14:03 grew 10% to 92 million compared with an average of 84 million for 2021. These figures exclude limited tier services and count multi-user plans as a single subscription, the RIA. A.A. Ads. Revenue from digitally downloaded music continued to decline in 2022, down 20% to just $495 million. Both digital album sales and individual track sales were down 20% to 242 and 214 million, respectively. Downloads accounted for just 3% of U.S. recorded music revenues in 2022 down from a peak of 43% of revenues in 2012. Revenues from physical music formats continue to grow after their remarkable resurgence in 2021. Total physical revenues of $1.7 billion were up 4% versus the prior year. Revenues from vinyl records

Starting point is 00:14:51 grew 17% to $1.2 billion, the 16th consecutive year of growth and accounted for 71% of physical format revenues. For the first time since 1987, vinyl albums outsold CDs and units, 41 million versus 33 million. And after a 2021 rebound versus the COVID-impacted 2020, revenues from CDs fell 18% to just $483 million in 2022, end quote. And finally today, something, something insert Deadpool joke here, although I'm not versed enough in Deadpool to insert one competently here. So anyway, T-Mobile apparently plans to acquire the Ryan Reynolds-backed budget wireless provider MintMobile for as much as $1.35 billion seeking to expand its pay-as-you-go market share,

Starting point is 00:15:44 quoting Bloomberg. The second largest U.S. wireless provider is acquiring Mint's closely held parent company, Keina Corp, with a combination of 39% cash and 61% stock, according to a statement Wednesday. The ultimate purchase price will be based on Mint reaching certain performance goals, both before and after the transaction closes. The sale talks were first reported by Bloomberg News in January. Reynolds, who owns an undisclosed but, quote, significant stake in Mint, will continue to make commercial appearances on the company's behalf, co-founder David Glickman said in an interview, adding the actor has incentives to

Starting point is 00:16:17 continue for years. Glickman and his partner Riswan Kasim will join T-Mobile and manage the business, which includes ultra-mobile in international phone service. Mint offers some of the country's lowest-price mobile plans starting at $15 a month for four gigabytes of wireless data. The companies don't disclose Mint's subscriber count. Its annual subscriber growth over the past four years has been 50 percent, and revenue growth has been 70 percent or more a year, Glickman said. The business has no stores and sells phones and mobile plans entirely online. The service is provided by T-Mobile already as part of a wholesale network sharing agreement. The Pays You Go market is seen as a source of overall subscriber growth as credit-challenged customers are eventually pulled into regular monthly

Starting point is 00:17:00 billings. Mint competes with other prepaid phone brands including Cricket from AT&T, total from Verizon, and Boost Mobile from Dish Network. Reynolds, a pitchman and co-owner of Aviation gin sold that brand to DeAjohn. GEO PLC in 2020 for $610 million, with nearly half that potential payment based on sales performance over 10 years. The actor, known for his work in the superhero series Deadpool, met Glickman through their joint work with the Michael J. Fox Foundation for Parkinson's Research. Reynolds also co-owns the Welsh soccer team Rexum AFC, which has been featured in a documentary series on the FX Network, end quote. I have a friend I've never seen. He hides his head

Starting point is 00:17:51 inside a dream. Someone should call him and see if he can come out. Try to lose the down that he's found. Talk to you tomorrow.

Tech Brew Ride Home - Wed. 03/15 – GPT-4 Is Here

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.