Tech Brew Ride Home - Thu. 06/23 – What If My Voice Was Your Alexa?

Episode Date: June 23, 2022

Would you want my voice to be the voice of your Alexa? An interesting raise that can change people’s vocal accents in real time. Instagram rolls out a comprehensive system to check ages on their platform. And is that new Brave search engine becoming a contender? Sponsors: HubSpot.com WorkCheck Podcast Links: Alexa will soon be able to read stories as your dead grandma (TechCrunch) That agent who sounds like they’re from Paris, Texas? Try Paris, France (TechCrunch) Cerebras Slays GPUs, Breaks Record for Largest AI Models Trained on a Single Device (Tom's Hardware) Instagram Tests Child Age Verification, but Falls Short of Full ID Checks (WSJ) Privacy-focused Brave Search grew by 5,000% in a year (Bleeping Computer) NBCUniversal, Google Compete to Help Netflix Develop Ad-Backed Tier (WSJ) 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 Thursday, June 23rd, 2022. I'm Brian McCullough today. Would you want my voice to be the voice of your Alexa, an interesting raise that can change people's vocal accents in real time? Instagram rolls out a comprehensive system to check ages on their platform, and is that new brave search engine becoming a contender? Here's what you miss today in the world of tech. Amazon is working on letting Alexa mimic any voice after hearing less than a minute of audio. They say that they want to do this as a way to, quote, make the memories last.

Memories of deceased family members, presumably, which sounds cool, I suppose, but also maybe a little creepy, quoting TechCrunch. At its annual Remar's conference today in Las Vegas, Amazon Senior Vice President and Head Scientist for Alexa, Rahit Prasad, announced a spate of new and upcoming features for the company's smart assistant. The most head-turning of the bunch was a potential new feature that can synthesize, short audio clips into longer speech. In the scenario presented at the event, the voice of a deceased loved one, a grandmother in this case, is used to record a grandson, a bedtime story. Prasad notes that using the new technology the company is able to accomplish some very impressive audio output using just one minute of speech. This required inventions where we had to learn to produce a

high-quality voice with less than a minute of recording versus hours of recording in the studio. The executive notes. The way we made it happen is, is by framing the problem as a voice conversion task and not a speech generation path. We are unquestionably living in the golden era of AI, where our dreams and science fictions are becoming a reality, and quote. So I'm wondering, if you played back this podcast, a minute of it, a minute of me talking to Alexa, could you make me the voice of your Alexa? Might there be some copyright issues therein? And while I feel like it could be creepy, I don't know, Longtime listeners of this show will recall when my lifelong mentor died a couple years ago,

I still have voicemails from him that I've saved on my phone and I listen to those now and then. Just did that last week, in fact. So, I don't know, maybe this is a thing people want. Somewhat related, interesting rays. I guess we really are in the golden age of AI, at least for voice stuff. Mountain View-based Sanas, S-A-S-N-A-S, which develops a call center voice AI that changes users' accents in real-time, has raised a $32 million series A led by Insight Partners. Get this, quoting TechCrunch. Our daily lives are more linked into a globalized grid than ever before. Products are sourced and shipped from afar, traveling to a place 3,000 miles away can be easier than getting across a big city in traffic.

and the information disseminates to anyone and everyone at the tap of a finger. A startup called Sanna's has developed some AI voice technology that aims to make one critical component of that grid work more smoothly. How people speaking the same language but with different accents can understand each other better by filtering accented voices and turning those accents in real time into other ones. Today, the startup is announcing $32 million in funding on the heels of strong momentum for its tools as it comes out of stealth and launches more widely. This series A is one of the biggest for a voice AI startup, and from what we understand, it's coming after Sanis turned down an acquisition offer from Big Tech.

If you can't buy them, invest in them, I guess. As you might have surmised from its list of investors, Sanis's technology is already being deployed in call centers. Specifically, it's found a lot of traction with far-flung customer service providers, which have become a hotbed of abuse against agents who might speak the same language as a customer, but heavily accented. There is a certain unease around the very concept of what Sanis is tackling and doing here. It raises a lot of questions of potential abuse, and apart from that, some might find it distasteful and passe for technology to be developed specifically to obscure a person's true identity. Shouldn't the people who are judging over accent, the ones who should be learning to be more open-minded and accepting rather than people forever accommodating prejudices by hiding anything that marks them

out as others or different? I had a demo of the service during my interview where Sanis rang up one of their customers' agents in India and got him to chat with me first with his own accent and then turning on his Midwestern neutral tone. It was a little creepy knowing what was happening in the background, but on the surface I was very surprised at how natural it all seemed. Well, natural enough, at least. His voice was clear, but maybe a little too clear and almost a tiny bit robotic and emotion-free. Apparently, that too, is somewhat intentional for now and might evolve if that's what customers and other users want, end quote. I guess this is somewhat related to Cerebrus says its wafor scale chip has set a record for the largest AI model trained on a single device with up to 20 billion parameters, quoting Tom's hardware. Cerebrus, the company behind the world's largest accelerator chip in existence, the CS2 Waferscale engine, has just announced a milestone. the training of the world's largest NLP natural language processing AI model in a single device.

While that in itself could mean many things, it wouldn't be much of a record to break if the previous largest model was trained in a smartwatch, for instance, the AI model trained by Cerebrus ascended towards a staggering and unprecedented 20 billion parameters, all without the workload having to be scaled across multiple accelerators. That's enough to fit the Internet's latest sensation. the image from text generator OpenAI's 12 billion parameter doll E. The most important bit in Cerebrus's achievement is the reduction in infrastructure and software complexity requirements. Granted, a single CS2 system is akin to a supercomputer all on its own.

The Wafers Scale Engine 2, which, like the name implies, is etched in a single 7 nanometer wafer, usually enough 400s of mainstream chips features a staggering 2.6 trillion 7 nanometer transistors, 850,000 cores and 40 gigabytes of integrated cash in a package consuming around 15 kilowatts. Keeping up to 20 billion parameter NLP models in a single chip significantly reduces the overhead in training costs across thousands of GPUs and associated hardware and scaling requirements while doing away with the technical difficulties of partitioning models across them. Cerebra says this is one of the most painful aspects of NLP workloads,

sometimes taking months to complete. It's a bespoke problem that's unique not only to each neural network being processed, the specs of each GPU and the network that ties it all together, elements that must be worked out in advance before the first training is ever started, and it can't be ported across systems. Pure numbers may make Cerebrus's achievement look underwhelming. OpenAIs, GPT3, an NLP model that can write entire articles that may sometimes fool human readers, features a staggering 175 billion parameters. Yeah, just like, clock speeds in the world's best CPUs, the number of parameters is but a single possible indicator of performance. Recently, work has been done in achieving better results with fewer

parameters. Chinchilla, for instance, routinely outperforms both GPT3 and Gofer with just 70 billion of them. The aim is to work smarter, not harder. As such, Cerebrus's achievement is more important than might first meet the eye. Researchers are bound to be able to fit increasingly complex models, even if the company does say that its systems have the potential to support models with hundreds of billions, even trillions of parameters, end quote. Instagram has begun testing tools to verify ages, starting with 17-year-olds or under, using photo IDs, software, or adults vouching for users, quoting the Wall Street Journal. Meta platforms said Thursday that if users of its photocentric social network try to edit their

dates of birth so that they are deemed to be over 18, it will now require additional verification. Instagram is testing three options, requiring users to upload IDs, ask adult Instagram friends to vouch for their ages, or submit videos of their faces that will be analyzed to estimate age. Instagram is adding these extra steps as part of its efforts to ensure age-appropriate experiences for minors. While children under 13 are prohibited by the network's terms of service, those who say they are ages 13 to 17 can use it with some limitations. Yet Instagram doesn't verify the age a user declares when creating an account, and Instagram said, these new tools won't change that. Instagram is testing the tools in the U.S. before deciding

whether to roll them out more broadly. For teen users, Instagram offers parental controls, automatically makes the account private and filters out contact from adults they don't follow. Those protections go away when the user's registered age is 18. Since 2019, Instagram has prompted new users to disclose their age and in 2021 made it a requirement for everyone signing up for an account. But it is an honor system and one that isn't always followed, according to a 2021 survey of more than 3,000 high school students across the country conducted by the Social Institute, an organization that works to help students develop healthy relationships with social media. 47% said they created social media accounts before age 13.

For now, under Instagram's test system, Meta is gathering new information that goes to its servers. The company will store IDs offered as proof of age for 30 days before deleting them. For the video selfie age verification option, Instagram will send files to the London-based digital identity company Yoti. That file will be used to estimate a user's age, then deleted immediately, says Julie Dawson, Yote's chief policy and regulatory officer, who added that it won't be used to train Yote's algorithm, end quote. Jeez, this has been a day of AI news, hasn't it? Didn't plan it that way. These were just the stories that we were dealt this morning. Remember how Brave recently launched a search engine of its own? Brave, the web browser,

the privacy-focused browser. Well, they now say that that new search engine had 411.7 million queries in May up around 5,000 percent since its June 2021 launch. Also, it is launching goggles, which lets users customize searches, which is very interesting. Quoting bleeping computer. Brave Search, the browser developer's privacy-centric internet search engine, is celebrating its first anniversary after surpassing 2.5 billion queries and seeing almost 5,000% growth in a year. To celebrate this success, Brave Software announced that Brave Search

is finally exiting its beta phase and will become the default search engine for all users of the Brave browser. Brave says it grew its current query volumes four times quicker than Duck.com. likely assisted by its large community of Brave browser users. Brave says that independence has remained the epicenter of the company's focus, with Brave search users receiving 92% of their queries directly from Brave's independent search index rather than through Bing and Google indexes. Goggles is a feature that allows Brave search users to customize how search results are ranked, setting custom preferences and priorities.

For example, users may favor results from small news blogs instead of large media outlets. So instead of looking through multiple search result pages, they can create a goggle for it and have these results rank higher. Brave says it will populate goggles with a few practical examples to help users understand how the feature works and what it can do, but the community is expected to build many more. A white paper gives more details about goggles, including examples of excluding the top 1,000 most popular domains for any search term, and excluding product reviews with commercial backing. With goggles, users can get highly curated search results that would be otherwise impossible in the context of a search engine that doesn't log queries for user profiling and tracking, end quote. Finally, today's sources are telling the journal that Comcast and Google have emerged as the top contenders to help Netflix develop its new ad-supported tier,

though Netflix has discussed ad partnerships with Roku, as we've discussed recently, quoting the journal. Netflix, which is hoping to boost revenue by selling ads around its programming, is still in the early stages of developing the strategy and has explored a range of tie-ups in recent weeks. A partnership with NBC Universal would likely be exclusive, the people familiar with the matter said. Comcast Video Ad Unit Freewheel would supply technology to help serve up ads, while NBC Universal's ad sales team would help sell ads in the U.S. and Europe, the people said. A partnership with NBC Universal would likely involve revenue sharing, and one issue might be whether Netflix would be guaranteed. a certain amount of revenue, they said. Linda Yacarino, Chairman of Global Advertising and Partnerships for NBC Universal would be a major player in such a partnership. Google brings to the table its own ad-serving technology and experience in video through YouTube and its online channel bundle, YouTube TV,

people close to the discussions said. Google already has a commercial relationship with Netflix, which is a customer of its ad-buying tools, they said. It is likely Google would also pursue an exclusive arrangement. Roku has also had early talks with Netflix about ad partnerships the people familiar with the matter said. The information previously reported that Netflix talked to Comcast and Roku about getting help on technical infrastructure or ad sales. In April, when Netflix announced its first quarterly subscriber loss in more than a decade, the company said it would move toward putting ads into its service, something co-chief executive Reed Hastings has long resisted. Forming partnerships with big competitors such as Google or NBC Universal and outsourcing technical

infrastructure and ad sales could help Netflix move faster to bring ad-supported versions of its service to market. Several ad industry executives estimated that it would take at least a year for Netflix to roll out ads across the service globally, though others said Netflix could begin testing ads in some markets much sooner. The major competitors, Netflix is considering each have experience in supporting other companies' ad sales efforts. NBCU has been the exclusive reseller of ads for Apple News and Apple stocks in the U.S. since 2017, and recently said it expanded its relationship to include the UK. NBC Universal would give Netflix open access to its ad tech partners, one of the people familiar with the matter said. Google began helping Walt

Disney, which had been a free will customer serve ads across video, smartphones, and desktops in late 2018, end quote. Reminder that we are doing a space today in a few hours, 3 p.m. Eastern, noon Pacific. And also, I want to note this. Ad free feed subscribers. Some of you didn't get an episode yesterday, and yes, I'm aware of it. Look, all I can tell you at this point is that the ad-free feeds have been a never-ending source of headaches for me. You might be aware that we have two different companies serving the ad-free feeds. The first one, Glow, got bought by Libson a while ago, which is a notoriously flaky company. Last month, I think it was. They didn't serve episodes for half a week. I couldn't even log into their service for a couple of days, and as recently as

this week for several days, their CSS was broken. So the page was all janky without formatting and such, and though I was able to figure out how to post, you know, your CSS breaking is something that should be fixed in a matter of hours, not days. Our second most recent vendor for the ad-free feed is Supercast, which you sign up for at tech.combe.comcast.com. Now, they've generally been better than glow in terms of reliability, but it's Supercast that didn't post yesterday's episode.

I emailed them last night, but haven't heard back yet at the time of this recording. And the way it works is I post the ad-free shows from our main podcast host feed, and then these other guys are supposed to, you know, notice that our RSS feed has been updated and update and serve the new episodes accordingly. When they don't, there's literally nothing I can do. The episode is there. It's available. But if they're asleep at the switch, it is entirely frustratingly out of my hands.

I log into Supercast. I click the link that says, hey, assholes, update the feed, but nothing happens. I cannot tell you how frustrating it is for me when this happens, especially because I can't do anything to make it better for you. I do all this work to write and record and put out a show, give it to you, the paying subscribers, my best paying fans. But you don't get it. It's maddening. I don't blame you if you unsubscribe and stop paying for this. Believe me, I wish it would just work, and it's such a source of constant hassle that I'm always tempted to just stop doing the ad-free shows entirely. The only thing that stops me from doing that is knowing how much some of you

value it. This is me apologizing again, but also throwing my hands in the air and being like, I don't know what to do at this point. Talk to you tomorrow.