Google steals OpenAI’s “Ship-mas” thunder

For most of the past 2 ½ years, Google has been on its back foot competing against the scrappy OpenAI and its wildly popular ChatGPT chatbot (393 million monthly users in October.) But this month there are signs that Google may be getting its groove back. OpenAI has been releasing a flood of new products,l and upgrades to old ones, this past 2 weeks, in an event they have been calling the “12 Days of Ship-mas” (because they are “shipping” products, get it?). Yet in midst of the frenzy from OpenAI, Google has dropped some eye-popping releases and upgrades of their own, releasing some of the air from the OpenAI hype balloon. To wit:

1. Gemini 2.0, Google’s most advanced AI model yet, with multimodal (text, image, audio, and video) input and output, plus ability to use tools like Search and Maps as an autonomous agent.

2. Agents - Project Astra is developing a universal AI assistant, and Project Mariner is working on AI task automation

3. Veo, a text to video generator that may be even better than OpenAI’s latest version of Sora, which has had so much demand that the site is down more than it is up.

4. Imagen - a text to image generator that is just as capable as, and much easier to use than, rival apps from OpenAI and Midjourney.

5. Whisk - an app to remix an existing image, or a set of images, in order to make something new in ways that you can specify.

It is clear that Google is not giving up without a fight, and with a depth of AI talent unmatched by any company (their head of AI research just won a Nobel Prize, just for starters) plus metric tons of cash to burn, no one should count Google out any time soon.

Google steals OpenAI’s “Ship-mas” thunder

Zuck joins Elon to try to block OpenAI’s switch to for-profit

I swear, the jockeying among the tech-bros in the AI world is starting to seem more and more like scenes from the movie “Mean Girls”, with everyone scrambling to become Queen Bee. Now Facebook CEO Mark Zuckerberg is joining up with former arch-rival Elon Musk to try to block OpenAI from changing from a nonprofit organization to a more conventional business. Remember just last year, when Elon challenged Zuck to a cage match (and Elon’s mother had to shut it down)? Good times. But the enemy of my enemy is my friend, as the ancient Indian Brahmin Kautilya wrote (in Sanskrit, just to show off.) And both Elon and Zuck really, really want OpenAI to fail, because they each have a competing AI chatbot they want to monetize. Elon is suing OpenAI on dubious legal grounds to block the transition, and now Zuck publicly allies with Musk to call on the California state government to intervene to quash the move. Amazing how a few billion dollars gives socially awkward geeks the illusion that they should run the world.

Zuck joins Elon to try to block OpenAI’s switch to for-profit

OpenAI cofounder says we are at “peak data”, so the future is agents

Revered AI researcher Ilya Sutskever, a co-founder of OpenAI who is now heading up a competitor startup known as Safe Superintelligence, says that the days when we can make AI models smarter by training them on more data are over, because AI companies have already hoovered up all the data on the internet, “and there is only one internet.” Instead, he predicts that future AI models will be designed to be more autonomous learners, able to learn from smaller datasets by actively engaging with the material in order to detect the underlying structure with fewer examples - more like humans do.

OpenAI co-founder says we are already at “peak data”, so the future is agents

How Claude is used in the real world

Anthropic’s chatbot Claude has become a fan favorite, second only to OpenAI’s ChatGPT. Anthropic decided to dig below the buzz, and developed some AI agents to monitor and categorize millions of user interactions with the chatbot to discover patterns of usage. The results held some surprises. The top category of use, representing over 10% of all interactions, was help with coding a website or mobile app. This is likely because Claude has become the major AI model behind Cursor, a popular coding assistant. Other major uses included content creation, academic research, career coaching, business advice, and language translation. Now that chatbots are developing voice interfaces, expect ever more use cases to develop, as people of all types develop a chatty, conversational mode of interacting with AI.

Goldman Sachs: only 6.1% of US companies using AI in the workplace

A recent survey by investment bank Goldman Sachs found that official adoption of AI by US companies lags well behind the hype - only 6.1% of the companies in its sample are using AI to produce their products or services. Reasons cited by firms for the slow pace of adoption include skepticism over the return on investment of scarce time and talent to AI projects. Until there are more success stories, many companies are taking a wait and see approach to AI. (In contrast, other surveys show that at least 25% of US workers are using AI in their jobs - they’re just not telling management about it, for fear of getting into trouble, or having management wise up to the productivity gains and instituting mass layoffs.)

Goldman Sachs: only 6.1% of US companies are using AI in the workplace

Time unveils new chatbot

Alongside its announcement of Trump as its Man of the Year, Time Magazine unveiled a new AI chatbot that has been trained to answer readers’ questions about the story, as well as about the three previous winners: Taylor Swift, Volodymyr Zelinsky, and Elon Musk. The chatbot can translate the story into German, Spanish, French, Russian, Ukrainian, and Mandarin. It can also summarize the story, and play an audio version of it.

Time unveils new chatbot

Waymo goes to Tokyo

Google-backed robotaxi company Waymo has been expanding operations to ever more cities in the US, and now is poised to make its first international leap, to Japan. The company recently announced that it is preparing to begin testing its driverless cabs in Tokyo, partnering with Japan’s largest taxi operator, Nihon Kotsu, and ride-hailing app GO, in order to start testing Waymo’s Jaguar I-PACE electric SUVs on Tokyo streets. Initially, human drivers will manually operate the Waymo vehicles in targeted areas of Tokyo, to train Waymo’s AI system. In keeping with its cautious approach to expansion that has allowed it to avoid the missteps of competitors in the US, Waymo projects an extended testing period before beginning operations in Tokyo, in order to learn not just the physical environment, but also the local customs and regulations. 

Waymo goes to Tokyo

AI designs perfect eye surgery to restore legally blind woman’s sight

A legally blind woman in the UK underwent an AI-designed procedure to resculpt her cornea and restore her sight. Physicians first created a detailed 3-D digital model of the 31-year-old patient’s eyes based on scans, then used AI to determine the exact procedures needed to resculpt her corneas with lasers, in order to restore normal sight without glasses. The procedure was so successful that the patient, previously legally blind, now has 20/16 eyesight, better than the normal 20/20.

AI designs perfect eye surgery to restore legally blind woman’s sight

Chinese researchers train AI doctors in a virtual hospital

Researchers at China’s Tsinghua University have developed a highly detailed simulation of a working hospital, populated by swarms of autonomous agents within an AI model. These agents take on the roles of doctors, nurses, and patients. The goal is to train the AI agents within the model to accurately diagnose patients. After treating a total of 10,000 virtual patients, the AI doctors gained sufficient skill that they scored 93.06% on the section of the MedQA benchmark that tests for knowledge of respiratory diseases. MedQA is a medical question dataset that mimics the questions in the US Medical License Exam, which is accepted by all US states for qualification for licensure as a physician.

Chinese researchers train AI doctors in a virtual hospital

Microsoft AI is developing a consumer health division

Microsoft’s head of AI, Mustafa Suleyman, has begun recruiting former colleagues at Google’s DeepMind AI research lab to join a budding consumer health division. Suleyman was a co-founder of DeepMind, but was forced out after Google acquired the company in 2014. (His co-founder Demis Hassabis stayed with Google, and was recently awarded the Nobel Prize in Chemistry for his work on medical AI there.) Suleyman landed on his feet, ultimately being hired by Microsoft to head up its consumer AI division. Now Suleyman has recruited Dominic King, a surgeon who was head of DeepMind’s health unit, as Vice President of Microsoft’s new AI health division, based in London. Several other DeepMind colleagues have been reported to be joining the new AI health team at Microsoft. Suleyman appears to harbor a deep grudge against his former company, and stealing talent from Google’s DeepMind satisfies both business purposes and a personal agenda.

Microsoft AI is developing a consumer health division

AI-designed inhibitor treats ulcerative colitis

More effective treatments for inflammatory bowel disease (IBD) are urgently needed, because some 40% of patients with moderate-to-severe IBD do not respond adequately to currently available treatments. Now a commercial drug discovery company has used AI to design a drug that targets a pro-inflammatory enzyme in the gut called prolyl hydroxylase (PHD). In animal testing, it restored the gut lining and reduced pro-inflammatory cytokines. It is now in Phase 1 clinical trials in humans.

AI-designed inhibitor treats ulcerative colitis

Nature article