OpenAI wants to be everyone’s AI platform

At its recent annual Developer’s Day (DevDay) on Monday, OpenAI released a bevy of new features and offerings, which all together make clear that the company wants to be the defining platform of the AI era, in the way that Google was the defining platform of the internet era, and Apple was the defining platform of the smartphone era.

The individual pieces of this strategy were:

• Apps integration - OpenAI released a development kit that allows external apps like Canva or Zapier to port themselves into ChatGPT, so that the user never has to leave the ChatGPT environment, and the app gains AI ease of use.

• Agent Builder - the company also released a no-code drag-and-drop tool for building autonomous AI agents, that can independently perform tasks for the user. So you never again have to use n8n or Make for your AI automation tasks.

• Codex - OpenAI has made this “vibe coding” tool much more capable, at least equal to and possibly better than the former developer favorite, Claude Code. This is aimed at developer lock-in for this pivotal sector of AI automation.

• Cheaper access to better AI models - the company is making its best models extremely affordable for enterprise-level use.

OpenAI’s DevDay announcements - their plan for Total World Domination

Nvidia’s market cap exceeds all of Big Pharma

Premiere AI chipmaker Nvidia has rapidly become the most valuable company in the world, with a market capitalization of $4.6 trillion. It has eclipsed Apple ($3.8 trillion), Microsoft ($3.9 trillion) and Google ($2.9 trillion.) Even more startling, Nvidia’s market capitalization is larger than many entire industries. Case in point, Nvidia is worth more than the combined market capitalization of all major pharmaceutical companies in the world (collectively known as Big Pharma.)

Nvidia’s market cap exceeds all of Big Pharma combined

OpenAI signs mega-deal with AMD, sending stock soaring

OpenAI CEO continues to make headlines with gigantic deals for AI infrastructure, such as chips and datacenters. This week, OpenAI and chipmaker AMD ( a distant rival of industry-leading Nvidia) announced an agreement in which OpenAI will purchase 6 gigawatts (GW) of computing power from AMD, and if all goes well, will acquire 10% of AMD at a penny a share.

This is a landmark deal on a number of counts. First, 6 GW is a LOT of computing power. It is approximately equal to the electrical use of New York City. And this is on top of the recent landmark deal OpenAI struck with Oracle, for up to 10 GW of compute. And similar to the results of the Oracle announcement, AMD stock launched into the stratosphere on the news, rising over 30% in a single day.

Finally, this deal is historic because it allows OpenAI to purchase up to 10% of AMD at a penny a share if certain deployment goals are met. We see OpenAI CEO Altman leveraging his ability to raise insane amounts of VC cash to make massive deals with suppliers, who are then arm-twisted into giving OpenAI equity in the supplier company. This allows OpenAI to vertically integrate his supply chain just by making massive equipment purchases that they need to do anyway.

OpenAI signs mega-deal with AMD, sending stock soaring

Mira Murati’s Thinking Machine Lab releases Tinker, AI customizer

Former Chief Technical Officer of OpenAI Mira Murati recently raised a staggering $2 billion in her new startup’s seed round (the very first, exploratory funding round for most companies, generally for less than a million dollars, up to a few million.) $2 billion for a seed round is unprecedented in the entire history of venture capital.

Now, a mere 2 months later, the company has announced its first product, Tinker, which makes customizing AI models faster, easier, and lots, lots cheaper. This type of AI model customization, called fine-tuning, is used to make a generic open-source model behave like a model that was specially-built for a single purpose, such as medical diagnosis or financial analysis. Fine-tuning applies primarily to open-source, freely available models, because owners of large proprietary models like OpenAI’s ChatGPT or Google’s Gemini are not going to let users muck about with their settings. Tinker will appeal to researchers seeking to invent new uses for AI, and to medium-sized companies who want custom AI models for their business without the overhead of large cloud-based models like ChatGPT and Gemini.

Murati herself is an interesting figure who embodies many of the contradictions of our present moment. She was born in a small city (Vlore) in a small country (Albania) that was then under Communist rule. With fierce drive and intelligence, she has remade herself into one of the most influential figures in the central technology of the age, an immigrant with American citizenship, a woman at the top of a male-dominated profession, and a child of communism who is now a billionaire.

Mira Murati’s Thinking Machines Lab releases Tinker

OpenAI’s Sora 2 video-creation app is a viral hit

Just prior to this week’s major reveals at OpenAI’s DevDay (see Top Story above), the company had released a crowd-pleasing video creation app, Sora 2. Similar to the viral popularity of Google’s image editing app Nano Banana the week before, OpenAI’s Sora 2 took the internet by storm. The ease with which users could generate realistic video clips just from a text description had immense appeal, and the fact that you could use a selfie photo to insert yourself as a character in the action turbocharged the adoption. OpenAI is hinting at ways that copyright owners like Disney or Pokemon might get revenue for allowing the masses to create videos containing their proprietary images.

OpenAI’s Sora 2 video-creation app is a viral hit

NBA uses AI to give fans real-time stats

The NBA has partnered with AWS, Amazon’s industry-leading cloud computing service, to launch Inside the Game, a “basketball intelligence platform” that will provide fans with real-time advanced analytics on the games, teams, and players they follow. The AI system will “watch” games in real time with camera feeds, and analyze the movements of each player and of the ball to measure the performance of each player and of both teams. Users will be able to access the performance metrics of the players instantly and continuously. The NBA projects that this will increase fan engagement, and it no doubt will for some, those who are especially stat-happy. Regardless of this, the AI system offers a gold mine of real-time stats for the coaching staff, both for in-the-game suggestions to players, and for later analysis to derive lessons learned.

NBA uses AI to give fans real-time stats

Deloitte chooses Claude, just hours after it has an AI whoopsie

Global accounting and business consulting firm Deloitte has recently announced that it will be giving all of its 470,000 employees access to the AI model Claude, the flagship product of AI startup Anthropic. In addition, the firm will be creating an internal Claude Center of Excellence to develop best practices and industry-specific solutions for clients.

Mere hours prior to this announcement, Deloitte agreed to repay the government of Australia part of a $440,000 fee for a report that turned out to be AI-generated, and riddled with errors and hallucinations, such as citing sources that didn’t exist.

Presumably, the Claude Center of Excellence will do better.

 Deloitte chooses Claude, just hours after it has an AI whoopsie

AI discovers new antibiotic to treat flares of Crohn’s Disease

Researchers at MIT and McMaster University have used AI to discover a new, narrow-spectrum antibiotic that is targeted at a bacterial species that is associated with flares of Crohn’s Disease. In the past, flare-ups of this debilitating inflammatory bowel disease have been treated with conventional broad-spectrum antibiotics, which decimate beneficial gut bacteria along with the usual culprit, adherent-invasive E. Coli, or AIEC. Since beneficial gut bacteria tend to suppress growth of the pathogenic AIEC , this can leave the patient more susceptible to subsequent flares. The MIT-McMaster team used an AI system to discover and validate a new narrow-spectrum antibiotic that targets AIEC but doesn’t suppress the beneficial bacteria.

AI discovers new antibiotic to treat flares of Crohn’s Disease

DoorDash unveils Dot, the adorable delivery robot

Door-to-door delivery giant DoorDash has introduced Dot, a compact autonomous robot for deliveries within a small local area. Approximately the size and weight of a riding lawnmower, Dot has an internal storage compartment that can hold 6 pizzas, or other items of similar size. It can travel at speeds up to 20 miles per hour with its all-electric power train, and is equipped with a variety of sensors to help it navigate and avoid obstacles, people, and pets. From its beginning, DoorDash has relied on gig work from contracted part-time drivers using their own cars. Dot allows the company to eliminate much of that overhead for hyperlocal trips within the same neighborhood.

DoorDash unveils Dot, the adorable delivery robot

Flock Safety’s AI drones may chase shoplifters soon

Flock Safety, an AI surveillance-tech company that supplies police departments with drones and license plate readers, has developed a system to protect big box stores from shoplifting. Drones will be docked on the roof of the store, and when they receive an activation signal, they launch and pursue shoplifters who are on foot or in a vehicle, all the while streaming a live camera feed to the store and to local law enforcement. The aim is to assist apprehension of offenders, and deter future attempts at theft. Maybe it’s just me, but this feels a bit too much like a dystopian “Robocop” or “Minority Report” future.

Flock Safety’s AI drones may chase shoplifters soon

Studies show that AI is not ready to replace doctors yet

In the wake of numerous reports of AI systems equaling or exceeding the performance of human physicians, a new crop of studies show that despite its impressive achievements, AI is not yet capable of fully replacing human physicians. As one example, reports of AI systems achieving superhuman accuracy in diagnosing medical conditions from X-rays or scans are legion. On closer study, it turns out that all of these systems are one-trick ponies, superhuman at diagnosing one disease, like pneumonia or heart failure, and vastly inferior for all others.  

As another example, AI startup Mercor has developed benchmarks for testing AI on real world tasks of many professions, including generalist physicians, such as PCPs. In its latest round of tests, Anthropic’s Claude Sonnet 4.5 topped the field of AI models by completing 62.1% of the tasks performed by a generalist physician, barely edging out OpenAI’s GPT 5 by 0.1%. These scores were significantly below the performance of experienced human physicians. AI capabilities will continue to grow, and healthcare systems will need to reimagine patient care to reap the benefits of AI’s developing strengths while maintaining a humane experience for all involved.

Why AI isn’t replacing radiologists - yet

Mercor’s General Practitioner Benchmark for AI models

AI Scribes reduce burnout

A recent study of 236 ambulatory care physicians and advance practice practitioners (PAs and NPs) in 6 different health care systems, found that within 30 days of implementing an AI Scribe system, indicators of physician burnout dropped from an average of 51.9% to an average of 38.8%, a 25% improvement. This was accompanied by significant improvements in reported cognitive task load, ability to give patients undivided attention, ability to add patients with urgent conditions to an already full schedule, and time spent after hours in documentation.

AI Scribes reduce burnout