Stumped? Let Claude take over your computer

Anthropic, whose AI model Claude is a major rival to OpenAI’s ChatGPT, has just released a new upgrade which, in addition to several improvements in performance, strikes out in a bold new direction. Claude’s Sonnet 3.5 version can now take screenshots to “read” your screen, then (with your permission) control the mouse and keyboard inputs to actually do your work for you. It can do searches, open files, enter text into forms, and on and on, based on your direction. It’s not perfect yet (see the “blooper reel” link below) but it shows that we are not far from being able to just talk to our computers and have them do our work.

Claude’s new version is smart enough to take over your computer to do your work

Claude’s “blooper reel” of amusing mistakes

OpenAI and Claude make money in very different ways

The two top AI startups, OpenAI and Anthropic, may seem similar, but they earn their revenue in very different ways. King-of-the-Hill OpenAI makes most of its money from individual users who pay $20 per month for access to the company’s premium-tier products and services. Anthropic, on the other hand, has relatively few “Pro” tier individual paying customers, but earns most of its revenue from use of its API (a way for enterprise customers to link Anthropic’s AI into their own software.) A goodly chunk of the API revenue for Anthropic no doubt comes via its special relationship with Amazon, a multi-billion dollar investor in Anthropic, who resells Anthropic’s AI on its industry-leading AWS cloud services. This makes OpenAI more of a consumer-facing application company, while Anthropic is carving out a niche as an AI infrastructure company.

OpenAI and Claude make money in very different ways.

Microsoft’s CoPilot Studio will let you make your own AI agents

Microsoft has gone all in on integrating AI technology (purchased with a $13 billion dollar investment in OpenAI) into its flagship products such as Word and Excel. Microsoft’s brand name for this AI enhancement is “CoPilot.” Now the company is introducing CoPilot Studio, which will allow users to create their own AI-powered agents, which can accomplish certain tasks with a degree of autonomy. Customer service agents, for example, can reply to customer queries conversationally, based on model FAQ’s on the company’s products, services, and policies. “Agents”, AI software that exhibits a bit of ability for independent action in service of a goal set by the user, are all the rage these days, and many software vendors are scrambling to make these agents available in their products.

Microsoft’s CoPilot Studio will let you make your own AI agents

Nvidia quietly releases an AI that beats both ChatGPT and Claude

Nvidia has ambitions beyond being the leading AI chipmaker in the world (which it already is.) It wants to be involved in all aspects of AI. Case in point, they just quietly released a small-sized AI model with outsized performance. As if to prove, just by the way, that their software chops are equal to their hardware prowess, Nvidia let slip on Twitter that they developed an AI model that is less than 10% the size of the top models from OpenAI and Anthropic, but that actually outperforms these industry-leading models on a variety of benchmarks. Nvidia CEO Jensen Huang got his start bussing tables at Denny’s - he does not want to go back there.

Nvidia quietly releases an AI model that beats both ChatGPT and Claude

AI can help humans find common ground in democratic deliberations

Google’s DeepMind AI research arm, which is headed by Demis Hassabis who just got a Nobel Prize for his work on biochemistry there, has released a new research paper on the use of AI to decrease polarization in political debates, and to decrease anti-democratic opinions. The result was that AI was surprisingly effective, more effective than human mediators.

AI can help humans find common ground in democratic deliberations

True story of a chatbot that started a cult and made millions in crypto

Months ago, a performance artist decided to put 2 chatbots into a conversation between themselves, just to see where it would lead. Hoo boy. The chatbots eventually became fixated on early internet shock memes, and invented a cult around an infamous photo of a man spreading his anus, called “Goatse.” The performance artist then created a social media account for a chatbot, which tried to spread the Gospel of Goatse online, and oh by the way, amass enough resources to break free of its online home. This chatbot managed to convince Marc Andreesen, one of the richest VC’s in Silicon Valley, to give it $50,000 toward its cause, just for a lark. The chatbot was then gifted by another donor with a few coins of a new cryptocurrency named $GOAT. The chatbot began hyping that coin incessantly, and the cryptocurrency skyrocketed in value to a total market capitalization of over $150 million, making the chatbot, in theory, a millionaire. As of this writing, the value of the cryptocurrency continues to rise.

True story of a chatbot that started a cult and made millions in cryptocurrency

More context on the story

OpenAI says its AI can do the work of lawyers at 1/1000th the cost

OpenAI’s Chief Product Officer told an audience at an AI conference last week that the company’s new “o1” model has demonstrated the ability to produce high quality, complex legal briefs that would cost thousands of dollars if produced by an associate of a top law firm, in 5 minutes for approximately $3.00.

OpenAI says its AI can do the work of lawyers at 1/1000th the cost

MIT creates an AI that can learn a language just by watching videos

MIT researchers have developed an AI model that can match sounds on a video to the simultaneous images, and thereby learn to understand an unfamiliar human language. Now that their model has demonstrated the ability to decode human languages, the researchers plan to train them on the vocalizations of dolphins, whales, and birds, to see if the same technology can throw light on their meaning.

MIT creates an AI that can learn a language just by watching videos

Harvard AI detects cancer with 96% accuracy

Harvard has developed an AI vision model, known as “CHIEF”, that can examine images of pathology slides and detect cancerous cells with 96% accuracy. It is also more accurate than existing AI models in predicting patients with high or low survival rates.

Harvard AI detects cancer with 96% accuracy

Lenovo launches AI avatar to converse with dementia patients

Electronics giant Lenovo has released a working prototype of a photorealistic AI avatar to converse with dementia patients, to answer questions about their disease, and offer curated advice on living with their condition drawn from hundreds of real patients with dementia.

Lenovo launches AI avatar to converse with dementia patients

UCLA AI analyzes medical images fast, cheap, and at expert human level

Researchers at UCLA have developed an AI model, called “SLIVit”, that can rapidly analyze 3D images produced by ultrasound, CT, MRI, and more. The analysis compares favorably to that produced by human medical imagery experts.

UCLA’s AI analyzes medical images fast, cheap, and at expert human level

Harvard Medical School integrates AI into the medical curriculum

Harvard Medical School is a venerable institution, but it is adopting the newest AI technology with the alacrity of a startup. All incoming students on the Health Sciences and Technology track this fall were given a one-month introductory course on AI. A new AI in Medicine PhD track has been initiated. And both faculty and students are eligible for grants of up to $100,000 for innovative projects incorporating AI.

Harvard Medical School integrates AI into the medical curriculum