Coding’s gone wild with Amazon, Claude, and Cursor

AI-assisted computer coding is having a “ChatGPT moment,” and the world of computer interfaces may never be the same again. It started last week when AI legend Andrej Karpathy tweeted that he was starting to transition to coding his computer projects in English, using an AI code generator named Cursor to generate the actual computer instructions. This hit the arcane world of elite coders like an earthquake, similar to when US cross country skier Bill Koch won silver in the 1976 Winter Olympics with no-wax skis (formerly sneered at), or when Dick Fosbury began winning high jump events with his oddly devastating Fosbury Flop backwards roll. The floodgates opened, and social media for the past week has been filled with videos of coders making startlingly complete apps in minutes, or just hours, instead of the standard days or weeks. My favorite demo video is the one in the link below, where an 8-year-old girl creates a Harry Potter-themed chatbot in just 45 minutes, without knowing how to write a single line of computer code.

8 year old girl creates Harry Potter chatbot in 45 minutes with Cursor

Cursor-mania is building on the recent mini-flurry of excitement about the app-creating ability of Anthropic’s Claude Sonnet AI model, in a feature called “Artifacts.” Artifacts allows noncoding users to create working apps, such as simple games (tic-tac-toe, Mario-type side scrollers) or utilities (to-do apps or calculators) just by asking Claude in English. See the brief video in the link below. Artifacts is available to all Claude users, free, paid, and on the Claude smartphone app.

Explaining Claude Artifacts

Finally, as an example of the real-world impact of AI coding assistance, the CEO of Amazon recently tweeted that Amazon’s AI assistant “Q” had allowed company developers to automate 99% of the work of updating Java on Amazon’s servers, saving 4,500 years of developer time, and hundreds of millions of dollars. The future is here, with fantastically productive software developers, and ordinary people able to ask the computer to build them a custom app in minutes or hours.

China dominates World Robotics Conference

Chinese robotics firms unveiled 27 different humanoid robots in last week’s World Robotics Conference in Beijing. The sole Western entrant was Tesla’s Optimus model. Leading-edge AI robotics companies such as Figure and Boston Dynamics likely stayed away because of the Chinese home field advantage. Nonetheless, the Chines entrants impressed with both their variety and their technical sophistication. Over the past 50 years, China has become the workshop of the world, and their deep engineering talent plus their manufacturing knowhow will make them a fearsome competitor in the race to manufacture humanoid robots cost effectively and at a global scale., as we have seen previously with solar panels and electric vehicles.

China dominates World Robotics conference

WaPo uses AI to detect misinformation in political ads

Large news publishers have expressed a good deal of skepticism about the usefulness and ethics of generative AI models. Famously, the New York Times is currently suing OpenAI for billions of dollars for what it claims is unauthorized use of their copyrighted articles to train ChatGPT. Now we find that the Washington Post has based a major investigative article, on the use of misinformation in political ads, on data acquired and categorized by AI. WaPo used AI to analyze transcripts, images, and on-screen text of over 700 political ads about immigration at the US border. Actually, this is a very good use case for AI, making possible an analysis not feasible economically or within the constraints of time so near the election. Kudos, WaPo, for creative and effective use of AI in investigative journalism. (And please face reality and cut a deal with OpenAI to get paid for your content.)

WaPo uses AI to detect misinformation in political ads

Not just for killing - AI drones detect landmines in Ukraine

The war in Ukraine is rewriting the book on modern warfare tactics, not least by demonstrating the surprising effectiveness of relatively cheap drones, either piloted or autonomous. Now Ukraine is repurposing drones from attack to defense, and are finding that the semi-autonomous vehicles are well-suited to help clear their landscape of unexploded landmines. Ukraine has more mines deployed on its territory than any other nation - it is estimated that landmines cover up to 1/3 of all Ukraine land.

Not just for killing - AI drones detect landmines in Ukraine

Apple to announce new AI iPhones on 9/9: “It’s Glowtime!”

Apple generally announces next-generation iPhones each September, and this week the company scheduled a press conference at their Cupertino headquarters for September 9. Apple is widely expected to launch AI-capable iPhones, powered by the beefy new A18 chip. The “Glowtime” tag line is apparently a pun on “It’s Showtime!” for kickoffs, as well as a veiled reference to an AI-powered Siri which is typically denoted by a glowing orb on the screen.

Apple to announce new AI iPhones on 9/9: ”It’s Glowtime!”

Mars rover uses AI to search for signs of life

NASA designed the Perseverance rover, which landed on Mars on February 18, 2021, to search for signs of life on the red planet semi-autonomously, using an AI-guided Xray spectroscope called PIXL (Planetary Instrument for Xray Lithochemistry.) PIXL uses AI to decide on promising rocks to analyze, and then taps the AI again to plan out the pattern of Xray beams to scan the rock for analysis. All this processing is done locally, on the rover, avoiding the 6 minute round trip by radio wave to mission control on Earth and back. This greatly shortens the time needed for each analysis, enabling a lot more science to be performed in the finite lifetime of the rover.

AI on the Mars rover

Penny-a-minute Voice bots help India’s AI startups compete

India’s AI startups face a number of challenges in trying to outcompete the lavishly funded US AI companies like OpenAI, even in their home market. Making lemonade out of lemons, Indian AI startup firms like Sarvam AI have focused on using rock bottom user fees (as low as a penny a minute), voice technology (which has been slow-rolled in the West due to safety concerns), and training their AI chatbot on the scores of local languages, which US firms have not prioritized. Early signs are that the voice interface in their native language is extremely popular with users, opening up AI to the many millions of potential customers with poor English skills, poor computer skills, or both.

Penny-a-minute Voice bots help India’s AI startups compete

LLMs predict rainfall by training on past weather maps

A team of climate scientists in China have developed AI systems that predict future rainfall as well or better than the current state-of-the-art AI models. The Chinese scientists took the simplest possible strategy to develop the weather predictions. They took an AI image generation system, similar to Midjourney, DALL-E, and other consumer text-to-image systems, trained it on a large set of sequential weather maps, and then prompted the system to predict the next weather map in the current sequence. It turns out that “Hey, draw me the next picture in this sequence” is a surprisingly effective way to produce a highly accurate prediction of future rainfall patterns, despite the fact that the AI model had essentially zero knowledge of climate science or atmospheric physics.

LLMs accurately predict rainfall by training on past weather maps

Epic leans into AI

Epic Systems Corporation, the leading vendor of electronic medical records to large health systems, with over 325 million current patients in its databases, has gone all-in on AI, and they are weaving it into the fabric of many of their core functions. This includes automated patient charting with AI Scribes (live in 186 organizations), chatbots to assist physicians in generating replies to inbox queries from patients (live in 150 organizations), a “Look Alikes” search tool that allows clinicians to identify rare diseases by matching patients with other patients with similar signs and symptoms (65 sites), a “Best Care Choices” tool to identify what treatments have worked for other similar patients (being piloted at NYU Langone and at Parkview Health), systems to streamline and speed up prior authorization requests , and over 100 other uses of AI.

Epic leans into AI

Epic’s webpage on their AI products and future roadmap

ElevenLabs plans to help 1 million people reclaim their voice

AI Audio startup ElevenLabs, famed for its high-fidelity voice cloning technology that has shown up in a number of notorious deepfakes, is now trying to use its AI for good, by allowing patients who have lost their natural voice due to neurodegenerative diseases like ALS to reclaim their voice. Patients use old recordings of their former voice to train the ElevenLabs AI voice cloning system, then can type their message into their computer and have their computer “speak” for them in a simulacrum of their natural voice. We recently featured a viral video of US Representative Jennifer Wexton (D-VA), rendered all but mute by Progressive Supranuclear Palsy, using the text-to-speech system to continue her official duties in the Congress. ElevenLabs is pledging to make their system available at no cost to up to 1 million patients.

ElevenLabs plans to help 1 million people reclaim their voice

Northwell, Mayo invest in robotics startup

Northwell Health, New York’s largest health care system, and the famed Mayo Clinic have each invested in a $36 million Series B-1 funding round for clinical robotics startup Clarapath, who makes automated tissue processing machines for pathology departments. Clarapath’s flagship SectionStar is an example of the type of lab automation with robotics and AI that promises to transform both pathology and scientific research, as we have reported in past issues.

Northwell, Mayo invest in robotics startup

One ER physician’s hopes for AI