Example Summaries

Original
Characters

The best way to have complex discussions? 512 points by anandbaburajan on May 6, 2024 | hide | past | favorite | 322 comments --- layer8 on May 6, 2024 | next [–] > we started searching for a tool specifically built for complex discussions. We found none This was basically solved in Usenet, more specifically, in news reader software. You had a clearly arranged threaded view (you could see the thread structure of as many as 50 postings on a single screen), with unread threads and unread postings highlighted, and pressing Tab jumped to the next unread posting. Unread status was per posting/comment, not by time. Many more conveniences for quick navigation, filtering, and so on. All newer discussion platforms have been a step back in terms of efficiency of use and ability for deep, long running discussions. Initially due to web browser limitations (though nowadays that shouldn't be much of a problem anymore), and later due to mobile touch interfaces (still poses some difficulties). --- tptacek on May 7, 2024 | parent | next [–] I have a very hard time connecting this to my lived experience of Usenet, which was full of top- and bottom- quoted posts, broken threads, and routine missing posts that sometimes left people talking past each other for days. I dearly loved Usenet; it started my career. But I don't miss it. Even before you get to basic affordances like posts, moderation and sorting, and following specific people, the basic nuts-and-bolts experience of discussing topics on Usenet was worse than it is today on Reddit, Usenet's rightful heir. --- fifticon on May 7, 2024 | root | parent | next [–] I disagree with this; I lived through this switch. The problem might be that it was a paradigm shift; I realize what reddit does is not something USENET could be scaled to(?). Using USENET across multiple decades, I was used to following certain groups meticulously, tracking and reading whatever new items appeared. In particular, the news reader application - NN - had very solid tools for browsing and tracking discussions. When set up correctly, I could work through updates by pressing the space-bar to page through them. Reddit, and the even worse lesser forums, loses pretty much all of that. Browsing reddit, for me, feels more like watching a mix of a river with flotsam drifting by, a busy traffic street, uncoordinated fireworks, and a tornado ripping through a midwestern city. There are no tools to track what you have and haven't read already, or what new comments have appeared. You cannot sort and filter the posts properly, the best you get is a "do you feel lucky?" search, which often shows that "no, you weren't lucky today". On low-traffic subreddits, it IS possible to track new stuff, but you have to do so manually. I offer no solutions, I don't know how to effectively do high-quality discussions for 6 or 7 billion people. --- dredmorbius on May 7, 2024 | root | parent | next [–] What struck me early on with Reddit is that it's where conversations go to die. That's both a matter of design and scale. Forums larger than ~10^2 -- 10^3 participants aren't really discussions so much as a compilation of hit-and-run pieces. For very large subs (10^5 -- 10^6), discussion is effectively over within a day, if not hours. A strong contrast are the now-defunct Google+ and (at certain points in its evolution) Ell o, and the not-quite-dead-yet Diaspora*, all of which had or have a "notifications" pane in which recent discussions are presented in full, and to which all or most prior participants (so long as they've not muted the discussion) see not only direct responses but new comments. I've seen specific conversations continued for days, weeks, months, and even years, productively, and it's a really good way to noodle at an idea (particularly with a good post moderator) over time. --- onemoresoop on May 7, 2024 | root | parent | prev | next [–] I find this part as the general trend of enshitification. Sometimes it's incompetence in UI/UX, other times it's other perverted reasons such as to keep users confused causing them to linger and be server more ads or to confuse users with billing, etc. --- barrkel on May 7, 2024 | root | parent | prev | next [–] Good moderation can increase the signal to noise ratio immensely, but also increases the latency. I vehemently disagree that Reddit is in any way even approaching equivalent, much less better, than Usenet was. There's something about the text area, vs an email-like posting box, which encourages low-thought short replies. --- skrebbel on May 7, 2024 | root | parent | prev | next [–] This must've differed a lot between usenet groups then; the ones I was on strictly followed a bottom (or infix) posting pattern and would chastise people who did it differently. --- navane on May 7, 2024 | root | parent | next [–] So the culture enforced the discussion format, not the technology. Technically we could be all writing in a giant Notepad file, and adhere to any discussion format. --- layer8 on May 7, 2024 | root | parent | next [–] The technology supported and encouraged it, for example by editors placing the cursor and the signature at the bottom of the quoted post by default, and auto-removing the quoted post's signature, so you could immediately start typing your reply at the "right" location. Furthermore, when reading postings, the viewer would automatically jump to the first nonquoted part. In email, top-posting began when Microsoft's first email client placed cursor and signature at the top of the quoted email instead, and didn't provide commands to reflow partial quotes, or any of the features mentioned above. It also had no threaded view, which is what makes it practical to only partially quote instead of fully. Culture is important, but technology can influence it heavily. --- saurik on May 7, 2024 | parent | prev | next [–] FWIW, many email clients -- before Gmail dumbed down email forever and no one seems to understand that email replies form a tree anymore -- also had (and maybe some still do?) that way of viewing the world. --- dredmorbius on May 7, 2024 | root | parent | next [–] There are numerous email clients for Linux-like platforms that provide this, both text-based and GUI. Text: Mutt, Alpine, and emacs's mailer off the top of my head. GUI: Sylpheed, Thunderbird, KMail (KDE's Kontact suite still strikes me as one of the best I've encountered), Evolution, and Clawsmail. My own strong preference remains mutt, and the ability to process huge amounts of complex email reasonably well is still utterly unmatched. --- Ringz on May 7, 2024 | root | parent | next [–] I like mutt, but aerc [0] is imho much better. But no matter which solution you prefer, editing emails in the terminal is so much more efficient. If the majority would switch to pure text emails instead of HTML... --- dredmorbius on May 8, 2024 | root | parent | next [–] Thanks! --- fifticon on May 7, 2024 | root | parent | prev | next [–] Outlook killed it, gmail arrived at a scene with an already dead body :-(

Original
Characters

This transcript is provided for the convenience of investors only, for a full recording please see the Q4 2025 Earnings Call webcast. Operator: Welcome, everyone. Thank you for standing by for the Alphabet Fourth Quarter 2025 Earnings Conference Call. At this time, all participants are in a listen-only mode. After the speaker presentations, there will be a question-and-answer session. To ask a question during the session, you will need to press *1 on your telephone. I would now like to hand the conference over to your speaker today, Jim Friedland, Head of Investor Relations. Jim Friedland, Head of Investor Relations: Thank you. Good afternoon, everyone, and welcome to Alphabet's Fourth Quarter 2025 Earnings Conference Call. With us today are Sundar Pichai, Philipp Schindler and Anat Ashkenazi. Now, I'll quickly cover the Safe Harbor. Some of the statements that we make today regarding our business, operations, and financial performance may be considered forward-looking. Such statements are based on current expectations and assumptions that are subject to a number of risks and uncertainties. Actual results could differ materially. Please refer to our Forms 10-K and 10-Q, including the risk factors. We undertake no obligation to update any forward-looking statement. During this call, we will present both GAAP and non-GAAP financial measures. A reconciliation of non-GAAP to GAAP measures is included in today's earnings press release, which is distributed and available to the public through our Investor Relations website located at abc.xyz/investor. Our comments will be on year-over-year comparisons unless we state otherwise. And now, I'll turn the call over to Sundar. Sundar Pichai, CEO, Alphabet and Google: Thanks, Jim. Hi, everyone. Thanks for joining us. It was a tremendous quarter for Alphabet. The launch of Gemini 3 was a major milestone and we have great momentum. Alphabet annual revenues exceeded $400 billion for the first time. This quarter, Search continued to accelerate with revenues growing 17%. YouTube's annual revenues surpassed $60 billion across Ads and Subscriptions. Cloud significantly accelerated with revenues growing 48%, now on an annual run rate of over $70 billion. Backlog grew by 55% quarter-over-quarter to $240 billion, representing a wide breadth of customers, driven by demand for AI products. We have over 325 million paid subscriptions across consumer services, with strong adoption for Google One and YouTube Premium. In addition, we have sold more than 8 million paid seats of Gemini Enterprise, which we launched just four months ago. And our Gemini App now has over 750 million monthly active users. We're also seeing significantly higher engagement per user, especially since the launch of Gemini 3 in December. Overall, we're seeing our AI investments and infrastructure drive revenue and growth across the board. To meet customer demand and capitalize on the growing opportunities ahead of us, our 2026 CapEx investments are anticipated to be in the range of $175 billion to $185 billion. Today, I'll provide an update on our AI progress and then share highlights from Search, Cloud, YouTube, and Waymo. First, AI progress across the full stack. Our unrivaled infrastructure serves as the bedrock of our AI stack. We have the industry's widest variety of compute options. That includes GPUs from our partner NVIDIA, who announced at CES that we'll be among the first to offer their latest Vera Rubin GPU platform. Plus, our own TPUs that we've been developing for a decade. In December, we announced our intent to acquire Intersect, which provides data center and energy infrastructure solutions. As we scale, we're getting dramatically more efficient. We were able to lower Gemini serving unit costs by 78% over 2025 through model optimizations, efficiency and utilization improvements. Next, world-class AI research, including models and tooling. We offer the most extensive model portfolio in the world and lead across Text, Vision, and Image-to-Video LMArena leaderboards. Gemini 3 Pro drives the state of the art in reasoning and multimodal understanding. It has seen the fastest adoption of any model in our history. Since launch, Gemini 3 Pro has consistently processed 3x as many daily tokens on average as 2.5 Pro. Our latest model powers Google Antigravity, our new development platform where agents can autonomously plan and execute complex software tasks. It already has more than 1.5 million weekly active users after launching just over two months ago. Our first-party models, like Gemini, now process over 10 billion tokens per minute via direct API use by our customers, up from 7 billion last quarter. Third, bringing AI to our products and platforms. We're shipping innovation at scale to bring helpful AI features to people everywhere. In January alone, we have launched Personal Intelligence in AI Mode in Search and the Gemini App; introduced new AI features to Gmail and updated Veo; reimagined Chrome as an AI-first, agentic browser through features like Chrome Auto Browse; announced Project Genie, which lets users create and explore interactive worlds generated in real-time using Genie 3, our general-purpose world model. And we laid the groundwork for shopping in the AI era by introducing a new open standard for agentic commerce, the Universal Commerce Protocol, built alongside many retail industry leaders. Finally, from Android to Pixel, we're getting our best AI capabilities into people's hands. At CES, a range of partners, including Samsung, showcased how they're bringing Gemini to more devices, from XR to the living room and beyond. And to confirm the rumors, we'll be introducing our Pixel 10a to our best-ever rated Pixel 10 series very soon. Turning now to key highlights from the quarter, starting with Search. Search saw more usage in Q4 than ever before, as AI continues to drive an expansionary moment. We've executed with incredible speed. We shipped over 250 product launches within AI Mode and AI Overviews just last quarter. We've integrated Gemini 3 directly into AI Mode in Search. Now Search can better understand your query, dive deeper on the web, and generate interactive UI experiences. And last week, we upgraded AI Overviews to Gemini 3, giving users a best-in-class AI response at the top of the Search results page. We've also made the Search experience more cohesive, ensuring the transition from an AI Overview to a conversation in AI Mode is completely seamless. These new experiences are proving to be more helpful and are driving greater usage. A few highlights: first, once people start using these new experiences, they use them more. In the U.S., we saw daily AI Mode queries per user double since launch and AI Overviews continue to perform very well. Second, people are engaging in longer, more complex sessions. Queries in AI Mode are 3x longer than traditional searches. We are also seeing sessions become more conversational, with a significant portion of queries in AI Mode now leading to a follow-up question. Third, people are searching in new ways beyond text. Nearly one in six AI Mode queries are now non-text, using voice or images. And Circle to Search is now available on over 580 million Android devices. Next, Google Cloud. Our growth in revenue, operating margin and backlog highlights the strength of our entire portfolio. One, we are winning more new customers faster. We exited the year with double the new customer velocity compared to Q1. Two, we are also signing larger customer commitments. The number of deals in 2025 over $1 billion surpassed the previous three years combined. And three, we continue to deepen our relationships with existing customers, who are outpacing their initial commitments by over 30%. Nearly 75% of Google Cloud customers have used our vertically optimized AI, from chips, to models, to AI platforms, and enterprise AI agents which offer superior performance, quality, security, and cost-efficiency. These AI customers use 1.8x as many products as those who do not, enabling us to diversify our product portfolio, deepen customer relationships and accelerate revenue growth. Our product line has multiple monetization levers, spanning infrastructure, platform, and high-margin, AI-powered products and services with 14 product lines each exceeding $1 billion in annual revenue. We offer leading infrastructure for AI training and inference to our Cloud customers, with the industry's widest variety of compute options: from our own seventh-generation Ironwood TPU to the latest NVIDIA GPUs. Our ten-year track record in building our own accelerators—with expertise in chips, systems, networking, and software—translates to leading power and performance efficiency for large-scale inference and training. Our Cloud AI accelerators serve the leading frontier AI labs, capital markets firms like Citadel Securities, enterprises like Mercedes-Benz, and governments for high-performance computing applications. We also offer our leading generative AI models, including Gemini, Imagen, Veo, Chirp, and Lyria to Cloud customers. In December alone, nearly 350 customers each processed more than 100 billion tokens. In Q4, revenue from products built on our generative AI models grew nearly 400% year-over-year, significantly accelerating from the prior quarter. Today, more than 120,000 enterprises use Gemini, including AI unicorns like Lovable and OpenEvidence, and global enterprises like Airbus and Honeywell. 95% of the top 20, and over 80% of the top hundred SaaS companies use Gemini, including Salesforce and Shopify. Gemini is becoming the AI engine for the world's most successful software companies. Leading enterprises are also driving strong demand for our enterprise AI agents. We have sold more than 8 million paid seats of Gemini Enterprise, our enterprise AI platform, to more than 2,800 companies, including BNY and Virgin Voyages, to streamline knowledge management and automate processes. Gemini Enterprise managed over five billion customer interactions in Q4, growing 65% year-over-year, for customers including Wendy's, Kroger, and Woolworths Group. Our integration of Gemini in Google Workspace is driving wins with global brands like Schwarz Group and public sector organizations like the U.S. Department of Transportation. We are also seeing momentum with independent software vendors. Revenue from AI solutions built by our partners increased nearly 300% year-over-year, and commitments from our top 15 software partners grew more than 16x year-over-year. Before moving on, I'm pleased that we are collaborating with Apple as their preferred Cloud provider and to develop the next generation of Apple Foundation Models, based on Gemini technology. Up next, YouTube. I want to highlight four points. First, streaming. In the living room, YouTube continues to be the number one streamer in the U.S. for nearly three years, according to Nielsen. From the NFL to Coachella, YouTube is where people watch today's biggest popular culture moments unfold. Second, Subscriptions. We continue to see strong subscription revenue growth across YouTube, particularly YouTube Music Premium. We'll soon launch new YouTube TV plans, bringing more choice and flexibility to subscribers with over ten genre-specific packages. And the NFL has seen strong NFL Sunday Ticket subscriber growth with YouTube, with the highest paid subscriber number ever in the history of the product. Third, podcasts. To illustrate YouTube's popularity, in October 2025, viewers watched over 700 million hours of podcasts on Living Room devices, up 75% from just a year prior. And fourth, AI is transforming the YouTube experience for both creators and viewers. On average, every day in December, over 1 million channels used our new AI creation tools to supercharge their creativity. During that same month, more than 20 million viewers used our new Ask tool, powered by Gemini, to learn more about the content they watched. And finally, Waymo. This week, Waymo raised its largest investment round to date, and is well positioned to continue its momentum, with safety at the core. In December, we surpassed 20 million fully autonomous trips and are now providing more than 400,000 rides every week. Waymo continues to expand its service territory. Its sixth market, Miami, launched two weeks ago; and Waymo will soon expand its service to multiple cities across the U.S., and in the U.K., and Japan. The team has made incredible progress on important capabilities, including opening up public service to airports and freeways. In closing, 2025 was a fantastic year for the company. A big thanks to our employees and partners worldwide. We're really well positioned going into 2026. Now, over to Philipp. [Transcript continues with CFO and Q&A sections...]

Original
Characters

# Decoding complexity: how machine learning is redefining scientific discovery ABSTRACT As modern scientific instruments generate vast amounts of data and the volume of information in the scientific literature continues to grow, machine learning (ML) has become an essential tool for organising, analysing, and interpreting these complex datasets. This paper explores the transformative role of ML in accelerating breakthroughs across a range of scientific disciplines. By presenting key examples – such as brain mapping and exoplanet detection – we demonstrate how ML is reshaping scientific research. We also explore different scenarios where different levels of knowledge of the underlying phenomenon are available, identifying strategies to overcome limitations and unlock the full potential of ML. Despite its advances, the growing reliance on ML poses challenges for research applications and rigorous validation of discoveries. We argue that even with these challenges, ML is poised to disrupt traditional methodologies and advance the boundaries of knowledge by enabling researchers to tackle increasingly complex problems. Thus, the scientific community can move beyond the necessary traditional oversimplifications to embrace the full complexity of natural systems, ultimately paving the way for interdisciplinary breakthroughs and innovative solutions to humanity's most pressing challenges. Keywords: machine learning (ML); deep learning (DL); artificial intelligence (AI); scientific discovery; complexity; physics; life sciences; computer science --- INTRODUCTION Machines have played a critical role in scientific discovery by providing the tools to observe, measure, and analyze natural phenomena. Scientific instruments, such as telescopes and microscopes, have historically enabled groundbreaking discoveries by revealing details invisible to the naked eye. With the advent of modern scientific instruments, including DNA sequencers, astronomical observatories, and high-resolution imaging devices, research facilities are producing terabytes or even petabytes of information. However, even with the most advanced computers, the sheer volume of data generated by large-scale projects such as the Large Hadron Collider (LHC) and the Square Kilometre Array (SKA) makes traditional analysis methods impractical. Complex problems such as weather forecasting, drug discovery, and genomic analysis often involve highly complex data sets and processes that cannot be efficiently managed without the assistance of machine learning (ML), which can help sift through massive data streams, identify patterns, and extract valuable insights. In this work, we explore the potential of ML and artificial intelligence (AI) in three types of scientific problems: (i) those where all governing equations are known, (ii) those with partial knowledge and (iii) those where little is known. We illustrate this with examples from the physical and life sciences, including turbulent flows, dark matter, drug discovery, and brain research. A good example of a complex system with vast amounts of data that traditional tools cannot process efficiently is brain research. ML enables the reconstruction of countless brain slices into highly accurate three-dimensional (3D) maps. In a recent study, Google researchers used AI to process 300 million brain images from Harvard, creating the largest-ever interactive 3D brain tissue model, now available online. One of the most widespread applications of ML is in drug discovery, addressing the time and cost challenges of traditional methods. The drug-development process can take over a decade and billions of dollars due to the complexity of identifying viable candidates. ML is transforming this by rapidly analyzing vast biological and chemical data, uncovering patterns that might remain hidden, and streamlining the identification of promising candidates. Artificial Intelligence is playing a transformative role in physics by improving data analysis, model development, and experimental interpretation. In astronomy, ML is improving the search for exoplanets by boosting the accuracy and efficiency of data analysis. AI-powered algorithms, particularly convolutional neural networks, can process massive data sets from telescopes to detect Earth-like exoplanets in noisy signals more precisely than traditional methods. ML is also crucial in areas like the Standard Model of particle physics. Automated algorithms were central to the discovery of the Higgs boson, and future experiments will need to operate at higher energies and intensities, generating data volumes too large for traditional methods to handle. Another scientific area in which ML is increasingly assisting research is mathematics, by improving the process of theorem proving, mathematical method development and discovery. ML systems have already demonstrated their ability to automate aspects of theorem proving; for example, Meta AI's neural theorem prover successfully solved 10 International Math Olympiad (IMO) problems, far exceeding the performance of previous ML systems. A collaboration between mathematicians and DeepMind demonstrated that ML can complement human intuition in proving or suggesting complex theorems. The team used ML to investigate long-standing conjectures, including the Kazhdan–Lusztig polynomials, and discovered new connections in knot theory. --- EMBRACING COMPLEXITY Machine learning comprises a growing set of algorithms, enabled by high-performance computing and increasingly vast data, that show incredible promise for handling complexity. Neural networks, despite being governed by simple rules, can perform complex tasks for which no traditional algorithms exist. Although they are fully observable and deterministic, we often cannot explain their decisions. However, they have led to groundbreaking discoveries, such as a new class of antibiotics. This creates challenges, such as the need for explainable AI (XAI). Symbolic approaches, such as gene-expression programming, sparse regression, and sparse Bayesian learning, have been successful, but their complexity grows exponentially with the search-space size. Recent efforts to combine symbolic and deep-learning approaches have enabled advances, such as discovering new materials. This raises fundamental questions about the limits of ML in scientific discovery; for example, can a complex system understand its own complexity? And how much can AI discover beyond its training data? The emergence of scientific foundation models (SFMs) and large language models (LLMs) is further pushing the boundaries of ML methods. Foundation models are large machine-learning or deep-learning generative models trained on vast amounts of data so they can be applied on a wide range of cases. Despite risks like bias and transparency issues, foundation models have revolutionized AI. LLMs now assist with tasks from writing and coding to guiding scientific experiments and generating ideas. Foundation models also show promise in non-text-focused areas, such as protein-structure prediction, protein design and climate simulations. An intriguing feature of foundation models, and particularly LLMs, is that they demonstrate so-called "emergent abilities." The term refers to unexpected, not explicitly programmed, capabilities that arise as model scale increases. Examples of LLM emergent abilities include in-context learning, complex reasoning, and multi-step problem-solving, which are highly valuable for scientific research. Thanks to this, LLMs are found to generalize to new tasks without prior examples (zero-shot learning) or with minimal data (few-shot learning). --- [Paper continues with additional sections on ML methods, case studies, and conclusions]