OpenAI’s “12 Days of Ship-mas” releases a bagful of goodies

OpenAI loves to release (or “ship”) new products and upgrades, to keep the buzz going that keeps the company alive while it burns billions in VC cash. Riffing off of the 12 days of Christmas, OpenAI has committed to releasing new products or updates every weekday for 12 days, in an event they call “The 12 Days of Ship-mas.” Here’s the roundup so far:

Day 1: OpenAI’s new “reasoning” model that is touted to think more deeply and systematically about questions, making it much better at complex math problems and computer coding. This release comes in 2 flavors: o1, which is available to all paying users; and o1 Pro, supposedly for PhD-level problems, for an eye-popping $200 monthly subscription fee.

Day 2: Reinforcement Fine Tuning, a way to train models on your own data, to make them better at answering queries in a defined subject area, whether it’s a scientific sub-field or the HR policies in your large company.

Day 3: Sora, the advanced text-to-video app that has been eagerly anticipated since it was first previewed on February 15 of this year, an eon ago in AI-time. The app is supposedly available to all of OpenAI’s paying customers, but the demand has been so high that so far almost no one can get through to try it out.

Day 4: Canvas, a side-by-side editor of documents or code (your prompts on one side, the resulting code or text on the other side to enhance real time interactivity) that looks and acts suspiciously like Claude’s Artifacts.

Day 5 (Today): Integration with Apple Intelligence. Apple has contracted with OpenAI to provide the industrial strength AI that some applications will demand. Meanwhile, Apple is moving full steam ahead to develop its own small, capable models that can run on your Apple device, be it an iMac, iPad, iPhone, or Apple Watch.

With 7 more days of announcements to go, expect at least one more major release that will keep the buzz going that OpenAI needs to keep attracting more VC cash to burn.

OpenAI’s 12 Days of Ship-mas product releases

Trump taps tech millionaire David Sacks as Czar for AI and Crypto

The incoming Trump presidency is shaping up to be one of the most tech-friendly administrations on record, with Elon Musk being a key advisor, and now Silicon Valley royalty David Sacks has been tapped to be Czar for both AI and Cryptocurrencies. Sacks, like Musk, is a South African entrepreneur and part of the legendary “PayPal mafia” that have spawned so many subsequent technology startups. Sacks is expected to favor a lighter regulatory touch on both AI and cryptocurrencies, so there is much rejoicing in both camps.

Trump taps tech-bro Davis Sacks as Czar for AI and Crypto

Microsoft previews CoPilot Vision, an app that can see your screen and take voice commands

AI agents are all the rage these days, as some of the best minds in tech try to hook up AI models with tools that let them do, not just talk. Apple is working on a smarter Siri, Claude has the clunkily named Computer Use function, Google is embedding its Gemini model into Maps, and now Microsoft is previewing CoPilot Vision. CoPilot is Microsoft’s brand for AI embedded into its productivity tools, and Vision allows CoPilot to see your screen, and lets you talk conversationally with it to find information or make plans. Expect to see more and more apps like this as AI companies race to finally get chatbots able to serve as personal assistants.

Microsoft previews CoPilot Vision to allow voice-driven AI browsing

Meta/Facebook proves less can be more by releasing tiny Llama 3.3 7B, which is as powerful as its older 405B model, but 100x smaller

Meta/Facebook keeps releasing highly capable open-source AI models for free, as part of its strategy of driving AI prices down to weaken its AI competitors. Meta makes so much money from its social media sites that it can afford to give away its AI models, while using these models to enhance the experience for its 3.6 billion users.

Now Meta has released a small AI model (“only” 7 billion parameters, which can easily run on the average laptop) whose performance rivals that of its current flagship Llama 405B (for 405 billion parameters, over 100 times larger, which needs to run on advanced GPUs.) Apple is also working to develop highly capable small AI models that can run on your iPhone, to avoid having to be dependent on rivals Microsoft or Google for artificial intelligence services. Expect to see more and more of these high-performing midget models in the future.

Meta proves less can be more with tiny Llama 3.3B AI model

Google releases ultra-accurate AI weather forecaster 

Until now, the best weather prediction models have used highly complex equations that required supercomputers to run. Now Google’s DeepMind AI research arm has released GenCast, an AI model trained on decades of past weather data, which uses the same diffusion technology that is behind text-to-image generators that create images on demand. GenCast is more accurate than standard models, much better at predicting severe weather events such as hurricanes, and runs quickly on standard computing hardware. As good as GenCast is now, its breakthrough use of standard AI image technology points the way to ever better weather prediction on ordinary computers.

Nature article

Tenant uses ChatGPT to successfully sue landlord for security deposit

A 23-year-old New York man moved out of a rundown apartment owned by a slumlord. The landlord kept the tenant’s security deposit on bogus grounds, a common practice among slumlords whose tenants usually cannot afford attorneys. Being Gen Z, the tenant ran his lease agreement through ChatGPT, which identified that under New York law, a landlord cannot withhold more than 1 month’s rent. The tenant then had ChatGPT walk him through the process of suing in small claims court, and represented himself in the proceeding. In the end, the court awarded the 23-year-old tenant full damages.

Tenant uses ChatGPT to successfully sue landlord for security deposit

48 hours in Tokyo with an AI travel companion

Tokyo is known as a fascinating tourist destination, but one that requires a local travel guide to make the most of, given the scarcity of English speakers and the complexities of local customs. Now, one intrepid traveler has chronicled his first trip to Tokyo, going without a human guide, but assisted by an AI travel buddy via an app on his smartphone, known as PalUp. Although there were some initial glitches, as the traveler learned how to query the AI in ways that kept it on track, the overall experience was positive. The AI acted as a pocket travel guide, on-call translator of signs and menus, and could even identify local sights or even restaurant meals from just a picture. As good as a knowledgeable human guide? Probably not, but much cheaper and more accessible.

48 hours in Tokyo with an AI travel companion

AI is twice as accurate at determining time of stroke onset

A multinational team of researchers has developed an AI system that can more precisely determine the time of onset and the potential reversibility of strokes from unenhanced CT scans. These judgements are crucial to treatment decisions and outcomes, since timely administration of clot-busting drugs can completely reverse the damage of a stroke if administered in the first two hours after onset. Studies show that the AI system is twice as accurate as the current state-of-the-art methodology for determining time of onset and reversibility from CT scan.

AI 2x as accurate as prior models in determining onset time of stroke

Nature article

Studies show AI chatbots can deliver effective psychotherapy; and young adults are signing up in droves

Hundreds of millions of people around the world are using AI chatbots to help with homework, assist with writing, create and summarize transcripts of meetings, and more. There is now growing interest in using chatbots for psychotherapy, since the demand and need for such services wildly exceeds the supply of trained human therapists. Two recent studies (see links below) find that 1) transcripts of chatbot interactions with human subjects seeking psychotherapy were judged as equal in quality to transcripts of patients and human psychotherapists; and 2) fine-tuned small AI models (which should be cheaper and potentially more secure than larger models) can effectively deliver Cognitive Behavioral Therapy to patients. (CBT is a very structured method of psychotherapy which focuses on identifying and combating self-limiting false beliefs in the patient, and setting concrete achievable goals for improvement in functioning.) Finally, an article by the BBC notes that young people who use an “AI companion” website called Character AI are flocking to use an AI chatbot persona known as Psychologist, which offers psychotherapeutic advice.

Article 1: Human-Human versus Human-AI therapy (a tie)

Article 2: Fine-tuned small AI models deliver effective Cognitive Behavioral Therapy

Article 3: Young people are turning to AI therapist bots

AI-designed protein fights cancer while reducing inflammation

A team of biochemists at the University of Washington, including this year’s Nobel laureate in Chemistry David Baker, created an AI system that designs special-purpose proteins for medical applications. In a recent paper, the system designed a protein that binds to Tumor Necrosis Factor receptors (which help kill cancer), selectively blocking the receptors that increase inflammation. This opens the possibility of devising future cancer therapies with fewer toxic side effects.

AI-designed protein fights cancer while reducing inflammation

Science article

Mount Sinai opens $100 million AI research center in Manhattan

Mount Sinai Health System, one of New York City’s five premier hospital systems, has just opened a $100 million research center for AI in Medicine. The new center will initially house 40 principal investigators, plus 250 graduate students, postdoctoral fellows, computer scientists, and support staff. The project was seeded by a major contribution form Hamilton James, executive vice president of Manhattan-based investment fund giant Blackrock, for whom, apparently a few tens of millions is pocket change.

Mount Sinai opens $100 million AI research center in Manhattan