Loading summary
Transcript1 lines
- [00:05]
Brian McCullough
Welcome to the Techmeme ride home for Thursday, November 21st, 2024. I'm Brian McCullough. Today the DOJ has filed its remedy for Google, but what would it mean for end users if their recommendations actually come to pass? Nvidia's earnings continue to be historic, but are they worried about current AI models hitting a wall? How AI might help make quantum computing become reality? And did a major new AI player just release its first product? Here's what you missed today in the world of tech. It's officially official the DOJ has asked the federal judge to force Google to sell Chrome, restrict Android from favoring Google's search engine ban. Default search deals on iOS and other devices require Google to allow websites more ability to opt out of its AI products and provide more ad placement controls to advertisers and syndicate its search results to rival search engines for at least a decade. Google says the DOJ's quote, wildly overbroad proposal goes miles beyond the court's decision and would hurt US Consumers and jeopardize US Global tech leadership. Quoting the Associated Press Although regulators stopped short of demanding Google sell Android 2, they asserted the judge should make it clear the company could still be required to divest its smartphone operating system if its oversight committee continues to see evidence of misconduct. A sale of Chrome, quote will permanently stop Google's control of this critical search access point and allow rival search engines the ability to access the browser that for many users is a gateway to the Internet, Justice Department lawyers argued in their filing. The Justice Department decision makers, who will inherit the case after president elected Donald Trump takes office next year, might not be as strident. The Washington, D.C. court hearings on Google's punishment are scheduled to begin in April, and Judge Mehta is aiming to issue his final decision before Labor Day. If Meta embraces the government's recommendations, Google would be forced to sell its 16 year old Chrome browser within six months of the final ruling. But the company certainly would appeal any punishment, potentially prolonging a legal tussle that has dragged on for more than four years. Besides seeking a Chrome spinoff and corralling of the Android software, the Justice Department wants the judge to ban Google from forging multibillion dollar deals to lock in its dominant search engine as the default option on Apple's iPhone and other devices. It would also ban Google from favoring its own services, such as YouTube or its recently launched artificial intelligence platform Gemini. Regulators also want Google to license the search index data it collects from people's queries to its rivals, giving them a better chance at competing with the tech giant on the commercial side of its search engine. Google would be required to provide more transparency into how it sets the prices that advertisers pay to be listed near the top of some targeted search results. Instead of going into further details about this remedy, I looked around for a piece describing what this might mean for users if it were to actually come to pass. Here you go, quoting I News. If it were to go ahead, and it remains a big if, as Google could still make the case against it, users may well see more friction in how easily they can access services like Gmail, Google's email service through a web browser, or Google Drive. Any potential change in ownership could impact how Google tracks web browsers and collects data that is the lifeblood of its business. This balance between data collection and privacy is something a new owner would have to navigate, with unknown impacts on how it would operate. A new owner could also change a host of product features, from the look and feel of the browser to the way results are displayed or the way businesses advertise on it. For advertisers, the ability to see in as much detail what users are interested in would be affected by the unbundling of Chrome from Google if it is ultimately sold to a smaller company. It also raises the question of whether new owners would have the resources to continue to invest in the product and whether the user experience might degrade or be overtaken by rivals. The US Government department stopped short of suggesting Google should divest its Android operating system, which would have huge effects on everyday users. But it's likely that is being held back as a stick to use should Google not comply with the current decision. Google has said it will be filing its own counter proposal to the judge's decision by year end and will be making its case against the decision in 2025. The big unanswered question from the judge's blockbuster decision this week is who exactly would be able to buy the Chrome browser and what would happen as a result. Amazon and OpenAI, the makers of ChatGPT, have both been touted as potential suitors with enough money to buy the browser, which has been valued by some as worth $20 billion. But neither seems likely. Amazon is facing its own antitrust investigation with the purchase of the leading web browser unlikely to help its own case that it does not hold an unfair market dominance, while OpenAI is the leader in generative AI and as such could see it argued that they too would wield too much power post purchase, end quote. And this is another thing that we've spoken about recently. I think even earlier this week, the CFPB will supervise tech companies offering digital wallets the likes of Apple Pay, Google Pay and Venmo. They'll treat companies with more than 50 million annual transactions now as banks Quoting Bloomberg, the original proposal set the supervision threshold at 5 million transactions a year. While the financial regulator can already take action against companies that break the law, the new rule would allow the CFPB to regularly supervise the large digital wallet and payments firms and their practices. More consumers are turning to digital wallets and payment apps to complete everyday transactions, and competition in the area has intensified, with Apple Pay leading the pack. Digital wallet use among U.S. consumers jumped to 62% last year from around 47% in 2022, according to federal Reserve surveys. Since the CFPB proposed the rule last year, Apple opened up use of its near Field Communication payment chip, changing its long held practice of limiting banks or other payment firms use of the technology. The strategy shift followed a deal with European Union financial regulators that required the Cupertino, California based company provide free access to its wallet technology for a decade. PayPal holdings recently disclosed that it's working with the CFPB to answer questions about backup payment options in its own digital wallet. The final rule will take effect 30 days after it's officially published in the Federal Register. End quote. Nvidia earnings continue to be historic Q3 revenue was up 94% year over year to $35.1 billion, above estimates by about $2 billion. Data center revenue was up 112% to $30.8 billion. And crucially, Nvidia forecasts Q4 revenue above estimates for our purposes. What I found interesting was Jensen Huang saying, don't worry, the current AI paradigm isn't broken. Quoting TechCrunch on its earnings call, Analyst prodded CEO Jensen Huang about how Nvidia would fare if tech companies started using new methods to improve their AI models. The method that underpins OpenAI's O1 model, or time test scaling, came up quite a lot. It's the idea that AI models will give better answers if you give them more time and computing power to think through questions. Specifically, it adds more compute to the AI inference phase, which is everything that happens after a user hits enter on their prompt. Nvidia's CEO was asked whether he was seeing AI model developers shift over to these new methods and how Nvidia's older chips would work for AI inference. Huang indicated that and test time scaling more broadly could play a larger role in Nvidia's business moving forward, calling it, quote, one of the most exciting developments and a new scaling law. Huang did his best to assure investors that Nvidia is well positioned for the change. The Nvidia CEO's remarks aligned with what Microsoft CEO Satya Nadella said on stage at a Microsoft event on Tuesday, 01 represents a new way for the AI industry to improve its models. This is a big deal for the chip industry because it places a greater emphasis on AI inference. While Nvidia's chips are the gold ST standard for training AI models, there's a broad set of well funded startups crafting lightning fast AI inference chips such as Grok and Cerebrus. It could be a more competitive space for Nvidia to operate in. Despite recent reports that improvements in generative models are slowing, Huang told analysts that AI model developers are still improving their models by adding more compute and data during the pre training phase. Anthropic CEO Dario Amodai also said on Wednesday during an on stage interview at the Cerebral Valley Summit in San Francisco that he is now not seeing a slowdown in model development foundation model pre training scaling is intact and it's continuing, said Huang on Wednesday. As you know, this is an empirical law, not a fundamental physical law, but the evidence is that it continues to scale. What we're learning, however, is that it's not enough. That's certainly what Nvidia investors wanted to hear since the chipmaker stock has soared more than 180% in 2024 by selling the AI chips that OpenAI, Google and Meta train their models on. However, Andreessen Horowitz Partners and several other AI executives previously said that these models are already starting to show diminishing returns. Huang noted that most of Nvidia's computing workloads today are around the pre training of AI models, not inference. But he attributed that more to where the AI world is today. He said that one day there will simply be more people running AI models, meaning more AI inference will happen. Hoang noted that Nvidia is the largest inference platform in the world today and the company's scale and reliability gives it a huge advantage compared to startups. Our hopes and dreams are that someday the world does a ton ton of inference and that's when AI has really succeeded, said Huang. Everyone knows that if they innovate on top of CUDA and Nvidia's architecture, they can innovate more quickly and they know that everything should work. Lumen is the world's first handheld metabolic coach. It's a device that measures your metabolism through your breath and on the app, it lets you know if you're burning fat or carbs and gives you tailored guidance to improve your nutrition, workouts, sleep, even stress management. Because your metabolism is at the center of everything your body does, optimal metabolic health translates to a bunch of benefits, including easier weight management, improved energy levels, better fitness results, better sleep, etc. All you have to do is breathe into your lumen first thing in the morning and you'll know what's going on with your metabolism. Then. Lumen gives you a personalized nutrition plan for that day based on your measurements. You can also breathe into it before and after workouts and meals so you know exactly what's going on in your body in real time. And Lumen will give you tips to keep you on top of your health game. So if you want to stay on track with your health this holiday season, go to Lum Lumen meride and get 15% off your lumen. That's L u m e N me ride for 15% off your purchase. Lumen makes a great gift too. Thank you Lumen for sponsoring this episode. This episode is brought to you by Incogni. Incogni is a service that helps protect your personal data from data brokers who collect, aggregate and sell it. Incogni reaches out to data brokers on your behalf and requests your personal data removal with the Family and Friends plan. You can also add up to four members to your subscription. And since many data brokers continue collection of your personal information even after they've removed it, Incogni makes sure your data stays off the market by conducting repeated removal requests. Incogni will handle any objections from data brokers and keep you updated on their progress every step of the way. Take control of your data Privacy today. Visit incogni.com, sign up and enjoy a 30 day money back guarantee to ensure you're completely satisfied with their service. Take your personal data back with Incogni. Use Ride Home at the link below and get 60% off an annual plan. Incogni.com ridehome that's incogni I n-c o g n I.com ridehome and use code ridehome to get 60% off their annual plan. Google researchers have introduced Alpha Qubit, a machine learning decoder that surpasses existing methods for identifying and correcting quantum computer errors. This could be a big deal. Quoting Quantum Insider Quantum computers, which leverage principles like superposition and entanglement, are poised to solve specific problems exponentially faster than classical machines, according to the post. However, qubits, the building blocks of quantum computers, are highly susceptible to noise, leading to freak errors. Overcoming this vulnerability is critical to scaling quantum devices for practical applications. To counteract this, quantum error correction uses redundancy. Multiple physical qubits are grouped into a single logical qubit, and consistency checks are performed to detect and correct errors. The challenge lies in decoding these checks efficiently and accurately, especially as quantum processors scale up. Current hardware typically exhibits error rates of 1 to 10% per operation, far too high for reliable computations. Future systems algorithms will require Error rates below 0.000000001% for practical applications like drug discovery, materials design, and cryptographic tests. Alpha qubit is built on the transformer architecture. Transformer refers to a type of neural network architecture designed to process sequential data efficiently by, for example, focusing on the most important parts of the data it analyzes. This helps alphaqubit to decode quantum errors accurately. As the name suggests, neural networks are meant to the human brain's neurons. Generally speaking, just like people have to learn before they master a new skill and continually hone that skill, neural networks have to learn and practice too. Alpha Qubit employs a two stage training process, pre training and fine tuning. In the pre training phase, the model is first exposed to synthetic examples generated by a quantum simulator. This enables it to learn general error patterns under various noise conditions. Then the system goes through the fine tuning. Here, the model is further trained on real world error data from Google Sycamore processor, tailoring it to the specific noise characteristics of the hardware. The decoder adapts to complex error types, including crosstalk, unwanted qubit interactions and leakage qubits drifting into non computational states. It also utilizes soft readouts, probabilistic measurements that provide richer information about qubit states. The team suggests that their success with AlphaCubits performance represents a significant step forward in the integration of machine learning and quantum computing. By automating the decoding process, the model reduces the reliance on handcraft algorithms, which often struggle with the complexity of real world noise. End quote in other words, AI to make quantum computing possible. Possibly. Sources are telling the journal that XAI has told investors it raised $5 billion in a funding round, valuing it at $50 billion, and that its revenue has reached $100 million on an annualized basis. Quote XAI was valued at $24 billion when it raised $6 billion in the spring. XAI's primary product is its Grok chatbot, available to premium subscribers of Musk's social network X. The company also recently made Grok available to business customers. Grok was launched in November 2023, making it late to a race with competitors including OpenAI, Anthropic and Alphabet's Google XAI spent this past summer constructing a data center in Memphis, Tennessee that houses 100,000 Nvidia chips for building its AI models. Musk has said the Memphis data center contains the most powerful AI cluster in the world and that he is planning to double its size. Musk is particularly focused on beating OpenAI, the ChatGPT creator he co founded in 2015. He has sued the startup and its chief executive, Sam Altman, for alleged fraud and antitrust violations, claims OpenAI have called baseless. Xai is set to debut the third version of its Grok language model in December. Musk has said it will be the world's most powerful AI by every metric and finally today, an interesting raise because it might signal a significant new player in the AI space. French startup H has announced its first product, Runner H, an agentic AI model available in private beta. H raised a lot of eyebrows when it raised a $220 million seed round back in May. You heard that right, $220 million seed round. Quoting TechCrunch, RunnerH is built atop the startup's own proprietary compact LLM based on just 2 billion parameters. H has set up a waitlist for Runner H on its site. CEO Charles Cantor said that it will be releasing APIs to those on the list over the coming days to use agents off the shelf that have been pre built by H, as well as for developers to create their own. Access to the API will also come along with access to something called H Studio to test and manage how these services work. Work initially using those APIs will be free and later there will be a payment model introduced. Even using compact LLMs, building and running AI is not cheap, especially as competition continues to raise money to develop their own products. TechCrunch has also confirmed that H is raising a Series A to build what Cantor describes as part of the second era of AI. With LLM companies like OpenAI being part of the first era, Runner H will initially focus on three specific use cases robotic process automation or RPA, quality assurance and business process outsourcing. RPA is an area that has existed for years, using basic scripts to automate the most repetitive tasks that humans have had to perform, such as reading forms, checking boxes and sending files from one place to another. In fact, a lot of RPA has never been built with AI baked in, even after AI started to develop advanced skills. The idea with Runner H is that it will be able to run RPA across forums, sites and other templates even when they have been modified, something that might have broken previous scripts and across a much wider range of sources. Quality assurance can cover a wide range of applications, but Cantor said that one of the most popular so far has been reducing the maintenance burdens around website testing, validating page availability, simulating real user actions, or ensuring compatibility across payment methods, in particular when modifications have been made. BPO is a catch all area that will cover not just fixing and improving billing processes, but also speeding up how an agent can use and access data from different sources and more. There has been a race among foundational AI companies around how many parameters are going into LLMs. GPT4, for example, has 175 billion parameters, but runner H is taking a very different approach with just 2 billion parameters, both for its LLM and for its computer vision based vlm. Cantor's argument is that this makes them significantly more efficient in terms of costs and operations, key when working on winning and keeping business deals and H's own operational costs. We are specialists, he said. We are building, building for the agentic era. The company also claims that it works. It says that its compact model outperforms Anthropic's computer use by 29% based on web Voyager benchmarks as well as models from Mistral and Meta. Nothing more for you today. Talk to you tomorrow.