Thinks 959

Mint on Vande Bharat trains: “Trains running on time have been overtaken by speed as a talking point…Although these trains hurtle along swiftly in comparison with the slow coaches that carry most of our railway passengers, they can only move as fast as our ageing rails allow. Unlike China’s maglev, which uses costly magnetic tracks and has ended up as a showpiece, Vande Bharat has the advantage of low-cost scalability. This means it could eventually save a lot more passenger travel hours. Technically, it represents only a modest advancement over what we had earlier. In a key sense, it’s what the auto industry calls a ‘re-skin’ job—a spiffy new exterior, that is, to mask much the same under the hood. As it boasts of slightly better traction and speed than other Indian trains, it qualifies as an upgrade. But it doesn’t rid the ride of shudders and jerks. Among its most notable features is its use of ‘regenerative braking’, by which the energy generated in this process is captured for re-use. As for cabin comfort, with on-board wi-fi and accessible plug-in points for devices, Vande Bharat offers just about the minimum that a traveller would expect, even if that’s a lot more than what the country’s railway users have been accustomed to. All taken into account, spruced up railway services mark a welcome break from the past. But had it not been for a markedly more contemporary facade, little else about Vande Bharat would impress anyone who has taken trains in well-off countries.”

Jamin Ball: “Frank Slootman said in his opening keynote: “In order to have an AI strategy, you need a data strategy.” I couldn’t agree more with this. Foundation models and their outputs are only as good as the data that feeds / trains them. And as we move from general purpose LLM engines (think ChatGPT) to enterprise specific engines (models trained on enterprise specific data), companies need to get their data in order. The idea that LLMs will be able to scour every system within an enterprise and grab data where it lies (both unstructured and structured) is not something I believe will happen. It ignores things like cost and performance. To truly take advantage of AI, enterprises need organized data in the cloud…Data + Compute + AI.”

Bastiat: “Do what you will, gentlemen; you cannot give money to some without taking it away from others. If you absolutely insist on draining the taxpayer dry, well and good; but at least do not treat him like a fool. Do not tell him: “I am taking this money from you to repay you for what I have already taken from you.”” [via CafeHayek]

Pranay Ahlawat: “Foundational models or the problem of natural language conversation isn’t new. Natural Language Processing, chatbot platforms and out-of-box text APIs from cloud vendors have been around for a decade today. Foundational models like Resnet-50 have been around since 2015. There are two things that are different about modern-day Generative AI.  First, modern language models or Large Language Models (LLMs) are architecturally different and have a significant performance advantage over traditional approaches like Recurrent Neural Networks and LSTM (Long Short-Term Memory). You will often hear the word transformers and “attention”, which simply put is the ability of the model to remember the context of the conversation more effectively. The quality of comprehension and ability to generate longer free-form text is unlike what we have seen in the past. Second, these models have a killer app unlike any other and is immediately consumable by non-technical users. We have had transformative technology breakthroughs in the past – internet, mobile, virtualization and cloud, but nothing has come close to the astonishing rise of Chat GPT, which reached a hundred million users in about two months. This tangibility has added to the hype and despite the huge potential, a lot of the claims about Generative AI are unrealistic.”