Rajesh Jain

NYTimes: “One hypothesis for how large language models such as o1 think is that they use what logicians call abduction, or abductive reasoning. Deduction is reasoning from general laws to specific conclusions. Induction is the opposite, reasoning from the specific to the general. Abduction isn’t as well known, but it’s common in daily life, not to mention possibly inside A.I. It’s inferring the most likely explanation for a given observation. Unlike deduction, which is a straightforward procedure, and induction, which can be purely statistical, abduction requires creativity…Large language models generate sentences one word at a time based on their estimates of probability. Their designers can make the models more creative by having them choose not the most probable next word but, say, the fifth- or 10th-most probable next word. That’s called raising the temperature of the model. One hypothesis for why the models sometimes hallucinate is that their temperature is set too high.”

FT: “Once known as a producer of everything from washing machines to chips, Hitachi has slimmed down, with a primary focus on digitising infrastructure and power grids…From the outside, Hitachi still looks like a sprawling conglomerate spread across train infrastructure, power grids and factory automation. But investors are convinced it has successfully broken conglomerate silos, applying IT and data science to become something like a management consultant to utilities, manufacturers and railway operators.”

Arm CEO Rene Haas: “At our core, we are computer architecture. That’s what we do. We have great products. Our CPUs are wonderful, our GPUs are wonderful, but our products are nothing without software. The software is what makes our engine go. If you are defining a computer architecture and you’re building the future of computing, one of the things you need to be very mindful of is that link between hardware and software. You need to understand where the trade-offs are being made, where the optimizations are being made, and what are the ultimate benefits to consumers from a chip that has that type of integration. That is easier to do if you’re building something than if you’re licensing IP. This is from the standpoint where if you’re building something, you’re much closer to that interlock and you have a much better perspective in terms of the design trade-offs to make.”

Sajith Pai: “I have been looking for a term, an acronym or a phrase that describes these families who speak English predominantly at home. These constitute an influential demographic, or rather a psychographic, in India – affluent, urban, highly educated, usually in intercaste or inter-religious unions. I propose to call them Indo-Anglians. Unlike Anglo-Indians, the original English-speaking community in India, who were Christians, Indo-Anglians comprise all religions, though Hindus dominate. Indo-Anglians are also a highly urban lot; concentrated in the top 7 large cities of India (Mumbai, Delhi, Bangalore, Chennai, Pune, Hyderabad and Kolkata) with a smattering across the smaller towns in the hills and in Goa.”

WSJ: “The outsize success of America’s talented entrepreneurs doesn’t stem from their superior intelligence. It comes from working at companies such as Google and Microsoft, which mine the technological frontier and expose employees to valuable knowledge, insights and opportunities. Apple is worth more than the 30 largest German companies combined. Apple’s employees and its alumni use their knowledge and training to create more value than their counterparts in Europe. Unlike Europe, the enormous success of American entrepreneurs motivated an army of talented Americans to get valuable on-the-job training, work longer hours, take risks and succeed. A small amount of success bubbles up from a large pool of failure.”

Debashis Basu: “While India has wasted three decades in muddling along, even after the so-called economic liberalisation of 1991, under the Modi government, there is a faint element of economic nationalism in schemes such as production-linked incentives (PLI) and Make in India. But for these schemes to be effective, it has to use the playbook of export champions. The incentive has to be linked to export, not just import substitution or higher production. Initially it will be hard, which will automatically reveal what needs to be done to make each of the sectors export-competitive. In each of the four countries that have recorded extraordinary growth, the government worked with the manufacturers to help them import technology, arranged cheap finance, culled the weaker players, and relentlessly imposed export discipline. India should learn from this and adapt.”

FT: “[Brain-computer interface (BCI)] devices use a variety of methods to collect signals from the brain, which are then interpreted using artificial intelligence and used to control computers. Neuralink, whose electrodes have been implanted in two people, says its devices have been used to play video games and manipulate computer-aided design software. The first brain implants in humans date back two decades, but recent advances in the electronics used to collect and transmit brain signals, as well as the machine learning needed to analyse and make sense of the data, have raised hopes that the devices could soon be medically useful.”

Cass Sunstein: “Could AI predict the outcome of a coin flip? Could AI have predicted in (say) 2006 that Barack Hussein Obama would be elected president of the United States in 2008? Could AI have predicted in (say) 2014 that Donald Trump would be elected president of the United States in both 2016 and 2024? Could AI have predicted in (say) 2005 that Taylor Swift would become a worldwide sensation? The answer to all of these questions is “No.” AI could not have predicted those things (and no human being could have predicted those things, either). There are some prediction problems on which AI will not do well; the reason lies not in randomness, but in an absence of adequate data. There are disparate challenges here, but all of them are closely connected to the knowledge problem, and in particular to the unfathomably large number of factors that account for some kinds of outcomes and the critical importance of social interactions. In important respects, the Socialist Calculation Debate and the AI Calculation Debate are the same thing.”

Andy Kessler: “How do you debunk conspiracy theories? It’s hard. First, they must pass the smell test. Most don’t. Then ask if someone can hold a secret for that long. Don’t believe movies, podcasters or even politicians. Find some real science. Most important, figure out who benefits from spreading the story. The trick is not to let your emotions get the better of you. Question authority.”

WSJ reviews Tae Kim’s book on Nvidia: “Artificial intelligence without Nvidia is impossible to imagine. Its chips are the building blocks for the AI infrastructure being developed by both entrenched technology giants and well-funded newcomers. Its software, called Compute Unified Device Architecture, or CUDA, lets developers take full advantage of Nvidia’s hardware: It’s the paraffin you toss on the dry tinder of Nvidia’s ever more potent and multiplying GPUs (graphics processing units).”

TIME on AMD’s Lisa Su: “When she became CEO a decade ago, AMD stock was languishing around $3, its share of the data-center chip market had fallen so far that executives rounded it down to zero, and the question on everybody’s lips was how long the company had left. An engineer by training, Su spearheaded a bottom-up redesign of AMD’s products, ­repaired ­relationships with customers, and rode the AI boom to new heights. In 2022 the company’s overall value surpassed its historical rival ­Intel’s for the first time. AMD stock now trades at around $140, a nearly 50-fold increase since Su took over. This fall, Harvard Business School began teaching Su’s stewardship of AMD as a case study. “It really is one of the great turnaround stories of modern American business history,” says Chris Miller, a historian of the semiconductor industry and the author of Chip War.”

WSJ: “Dell specializes in everything sandwiched in between those two ends of the technology stack—between chips and software. It turns out that in the age of AI, there’s a tremendous demand for racks of servers and huge arrays of storage. Each rack of servers is a stack of computers about the size of a bookshelf. These racks are crammed together inside the vast data centers where the internet actually resides, and the most power-hungry ones, for training AI, can consume as much power as 100 average American homes. They generate so much excess heat that they have to be liquid-cooled. Each one costs hundreds of thousands of dollars—[Michael] Dell won’t say exactly how much. In the past two years, his company has sold storage arrays capable of holding a total of 120,000 petabytes, says Dell. For perspective, OpenAI’s latest chatbot, GPT-4o, was trained on about a petabyte of data, which represents all the text on the open internet, the transcripts of over a million hours of YouTube videos, plus countless images.”

The Verge: “Next-generation models, [Ilya Sutskever] predicted, are going to “be agentic in a real ways.” Agents have become a real buzzword in the AI field. While Sutskever didn’t define them during his talk, they are commonly understood to be an autonomous AI system that performs tasks, makes decisions, and interacts with software on its own. Along with being “agentic,” he said future systems will also be able to reason. Unlike today’s AI, which mostly pattern-matches based on what a model has seen before, future AI systems will be able to work things out step-by-step in a way that is more comparable to thinking. The more a system reasons, “the more unpredictable it becomes,” according to Sutskever. He compared the unpredictability of “truly reasoning systems” to how advanced AIs that play chess “are unpredictable to the best human chess players.” “They will understand things from limited data,” he said. “They will not get confused.””

CNN Business: “As opportunities arise in streaming, Sony is trying to transition from being a legacy consumer electronics company to an original content and entertainment company. The strategy is working: In the past three years, Sony’s stock has started to break out of a decades-long slump. Sony’s stock price in Japan recently closed at the first record high since March 2000, signifying confidence in the company’s ability to evolve its game offerings and steer itself toward entertainment, Damian Thong, a research equity analyst at Macquarie, told CNN. “If you went back 30 years ago, it was an electronics company, so best known as a seller of hardware,” Thong said. “But today, the company is primarily generating profits off of entertainment, which is games, music and (TV and movies).””

Fei-Fei Li: “I think spatial intelligence is where visual intelligence is going. If we are serious about cracking the problem of vision and also connecting it to doing, there’s an extremely simple, laid-out-in-the-daylight fact: The world is 3D. We don’t live in a flat world. Our physical agents, whether they’re robots or devices, will live in the 3D world. Even the virtual world is becoming more and more 3D. If you talk to artists, game developers, designers, architects, doctors, even when they are working in a virtual world, much of this is 3D. If you just take a moment and recognize this simple but profound fact, there is no question that cracking the problem of 3D intelligence is fundamental.”

Semi Analysis: “The reality is that there are more dimensions for scaling beyond simply focusing on pre-training, which has been the sole focus of most of the part-time prognosticators. OpenAI’s o1 release has proved the utility and potential of reasoning models, opening a new unexplored dimension for scaling. This is not the only technique, however, that delivers meaningful improvements in model performance as compute is scaled up. Other areas that deliver model improvements with more compute include Synthetic Data Generation, Proximal Policy Optimization (PPO), Functional Verifiers, and other training infrastructure for reasoning. The sands of scaling are still shifting and evolving, and, with it, the entire AI development process has continued to accelerate. Shifting from faulty benchmarks to more challenging ones will enable better measures of progress. In this report we will outline the old pre-training scaling trend as well as the new scaling trends for post-training and inference time. This includes how new methods will push the frontier – and will require even more training time compute scaling then thought before.”

WSJ: ““Long thinking” didn’t make it into the zeitgeist when OpenAI’s ChatGPT first stunned the world two years ago with rapid replies to questions about almost anything. But it has the potential to reduce or eliminate the errors that frequently peppered those responses. The idea is just what it sounds like, at least at the highest level: Long-thinking AI models are designed to take more time to “think over” the results they generate for us. They will be intelligent enough to give us updates on their progress and ask us for feedback along the way. That can mean spending a few more seconds on a problem—or much, much longer, as Huang indicated in another telling remark last June…As the models’ reasoning ability develops, AI is expected to evolve far beyond the current tech that works on our behalf in customer service or automation, or the even more sophisticated agents that are just beginning to appear.”

NYTimes: “Claude, a creation of the artificial intelligence company Anthropic, is not the best-known A.I. chatbot on the market. (That would be OpenAI’s ChatGPT, which has more than 300 million weekly users and a spot in the bookmark bar of every high school student in America.) It’s also not designed to draw users into relationships with lifelike A.I. companions, the way apps like Character.AI and Replika are. But Claude has become the chatbot of choice for a crowd of savvy tech insiders who say it’s helping them with everything from legal advice to health coaching to makeshift therapy sessions. “Some mix of raw intellectual horsepower and willingness to express opinions makes Claude feel much closer to a thing than a tool,” said Aidan McLaughlin, the chief executive of Topology Research, an A.I. start-up. “I, and many other users, find that magical.””

Tae Kim, author of “The Nvidia Way”: “Two important lessons from the book are: mission is the boss and speed of light. Mission is the boss means making the right decision for the company and the customer—not what’s good for your boss’s boss. In corporate America, a huge percentage of time—30 to 50 percent—is spent doing things that don’t help the end customer but instead make your boss look good to their boss. Minimizing that and focusing on the mission, not someone’s bonus, is critical. Speed of light means working with extreme speed and velocity. At most companies, when you do a project, you have KPIs, and you might say, “I did 10 percent better than last time” or “We’re 10 percent better than the competitor.” If you said that at NVIDIA, you’d get dressed down. They don’t care about what you did last time or how you compare to competitors. They care about what’s physically possible. If everything were perfect—no queues, no lag—what’s the absolute limit of physics?”

Walter Lippmann (1937): “Thus it has come about that under gradual collectivism the struggle for power has become ever more intense. As men learn that their fortunes depend increasingly upon their political position, the control of the authority of the state becomes a prize of infinite value.” [via Cafe Hayek]

FT: “Scientists have raised the alarm about the potentially existential threat of “mirror life” — manufactured bacteria that are structural reflections of natural microbes and could overwhelm the defences of people, other animals and plants. An international group of almost 40 researchers, including two Nobel laureates, warned on Thursday that such synthetic organisms might “pose unprecedented and largely overlooked risks to much of existing life”. The stark message highlights how advances in synthetic biology that have helped drive big health breakthroughs could one day have the capacity to generate deadly new organisms by accident or design.”

WSJ: “Companies that succeed with modern generative AI tools…or offerings from a raft of new startups…are discovering that in order to get real value from AI, they have to organize their data in ways they might not have before. And this isn’t a one-and-done effort. To keep their shiny new AIs up-to-date, the information they feed them must be kept constantly updated—creating more work for humans…Every company I talked with mentioned that to get real value out of their shiny new generative AI systems—no matter the application—they needed to overhaul or double down on their strategy for feeding it the kind of data that today’s AI excels at processing—“unstructured” data. About 90% of the data most companies have is this kind of data—not numerical, not in a spreadsheet, but in the form of documents, emails, manuals, customer-service chats, contracts and the like. And the real value of today’s generative AI for companies is in unlocking it. Think of it as centralizing all the know-how that is typically spread out across the brains and storage accounts of all the humans in an organization. Making all of that information and knowledge available to everyone else in an organization has long been the dream of corporate IT—and generative AI can get companies one step closer to it.”

Jaspreet Bindra: “The pre-internet customer, who I call the industrial customer, gave way to the digital customer as Instagram, Google and other apps started dominating their lives. Now the digital customer will give way to the AI customer, as ChatGPT and other AI tools and agents inveigle themselves into our lives. The industrial customer had limited and standardized choice, while the digital one reckoned with the abundant variety that Amazon threw open. The AI customer’s choice will be infinite and hyper-personalized, as agents scour the internet to find what she wants based on her innate preferences. Interaction with products was transactional and local for the industrial customer, while it’s social and omnichannel for the digital one; for the AI customer, chatbots will make this interaction conversational (as with another human) and digitally immersive as companies like Meta infuse AI into our visual and tactile environments.”

WSJ: “Madison Avenue has been rapidly changing for over a decade, but the threats to an industry once centered on creatives have never been so great. Tech giants control more than half of the $1 trillion ad market, and quants armed with reams of data direct ad buying. Now, generative artificial intelligence is sending shock waves through the marketing world, promising to create and personalize ads cheaper and faster than ever.”

Business Standard has a review of Kevin Rudd’s book “On Xi Jinping”: “The major theme of the book is that Xi has abandoned Deng Xiaoping’s dictum to “hide the strength, and bide the time”. Under Xi, China is ready to flex its muscles. The primary argument is that China under Xi is moving left on the economy and right on foreign policy. The author calls this “Xi’s Marxist Nationalism”. The clampdown on the private sector has been the norm under Xi, who perceives its existence as an aberration to CPC tenets. The trend is towards the “contraction of the private sphere and expansion of the public sphere controlled by the party”. When it comes to domestic policy, the posturing is closely dependent on nationalism. This is apparent with the rise of “wolf warrior” diplomacy. These two trends are crucial because Xi does not want the Chinese state to become a challenger to the “political and operational primacy of a Leninist Party”. The Leninist Party is the soul of China and being “red’ is mandatory. No surprises that Xi has executed the longest anti-corruption drive in the CPC’s history, with the projected goal of preserving the party’s sanctity.”