“DeepSeek Moment #2”: Chinese AI model bests the US at a fraction of the cost

China has done it again. In January, the release of DeepSeek by a Chinese AI startup showed it to be so shockingly good, despite being much smaller and cheaper than US AI models, that it temporarily erased over $600 billion from the valuation of leading tech stocks. These stocks have since rebounded nicely, but now another Chinese AI startup, Moonshot, has released Kimi K2 Thinking, which challenges and in some tests outperforms US models, but is much more efficient and much cheaper to train and run. Top US models cost hundreds of millions of dollars to train; Kimi K2 Thinking reportedly only cost 4.6 million, or 100 times less. Ruh-roh, says Scooby-Doo.

Are the Chinese winning the AI race? Nvidia CEO Jensen Huang thinks so, citing the facts that the Chinese have much more electric power capacity and many more AI engineers than the US. (He was later forced to walk this back, but he did say it.) He is also hella mad that he can’t sell chips to the Chinese because of the US trade embargo. (No matter, the Chinese government has now banned all foreign chips in government-supported projects.)

So far, the Chinese government is encouraging their AI companies to make their models open source, free for anyone to use, because this wins friends in the developing world while making it hard for US companies to charge a lot for their closed source models. Win-win for the Chinese.

Kimi K2 Thinking is another “DeepSeek” moment

Try Kimi K2 Thinking here for free

Fei Fei Li says true AI requires understanding the real world

Fei-Fei Li, the “Godmother of AI”, whose ImageNet project is one of the foundations of modern artificial intelligence, says that current AI is inherently limited because it only knows words. Calling current AI “wordsmiths in the dark”, Li notes that animals developed brains to guide movement in the real world, and that language was a late addition to neural networks designed for other functions entirely. She argues that true artificial intelligence will need to deeply comprehend how the real world works at a physical level. This will require building what she calls world models - not a model of one world, like a computer game, but a model of how all worlds possible in the physical; universe might function. Right or wrong, she is at the forefront of a new wave of AI innovation, one which is focused for now on making robots safe and adept in real world tasks.

Fei-Fei Li says true AI requires understanding the real world

Perplexity pays Snapchat $400 million to be its AI search engine

Social media company Snapchat has just announced that AI search startup Perplexity will power conversational search on its website, starting in 2026. Snapchat has over 900 million monthly active users, including 75% of 13-to-34-year-olds in 25 countries. The deal gives Perplexity increased reach and distribution of its AI services, while Snapchat gets to position itself as a cool, AI-forward social media site, while reaping a sweet $400 million from Perplexity.

Perplexity pays Snapchat $400 million to be its AI search engine.

Tesla shareholders approve $1 Trillion payday for Elon Musk

Elon Musk, already the world’s richest man, is on track to become the planet’s first trillionaire, if he can meet the conditions of the new incentive package approved by Tesla shareholders. In order to get his trillion-dollar payday, Musk will have to increase Tesla’s market capitalization to $8.5 billion, which no company has ever achieved. (Nvidia’s market cap is around $5 billion, Apple and Microsoft are around $4 billion.)

In addition, Musk must meet some aggressive operational goals, including selling 20 million cars (current total 3 million) and 1 million humanoid robots (current total 0.)

To the extent that this pay package is more than a publicity stunt, it is further proof of how vital Musk is to Tesla’s future. Without Musk’s ability to excite investors on future prospects, Tesla is a wildly overpriced stock, worth more than the next 35 most-valuable carmakers combined, including General Motors and Ford. In order to keep Musk interested, shareholders have given him a chance at unimaginable wealth and a full 25% of all company shares.

Tesla shareholders approve $1 Trillion payday fore Elon Musk

McKinsey reports on AI in business

Elite consulting firm McKinsey has released its 2025 report on the state of AI in business. Some takeaways:

• Everyone’s testing, few are implementing. 88% of businesses are experimenting with AI. Only 38% have implemented systems beyond pilots.

• Few businesses are seeing any impact on profits. Only 6% of companies are seeing a positive ROI yet (similar to the 5% figure in the recent MIT study.)

• The biggest successes are the companies that go beyond cost-cutting, redesigning workflows for growth and innovation, not just efficiency.

Every new technology enables new solutions, but is initially implemented as a swap-in substitute for the technology it is replacing. Electric motors were at first huge, and used as a substitute for the massive steam engines of the day. Innovators eventually realized that electric motors could be small, and brought to the tabletop, or even made into hand tools, revolutionizing how work is done. AI is likely to travel that same path to integration into the economy.

McKinsey reports on AI in business

Google Maps gets an AI upgrade

Google has finally gotten around to infusing its state-of-the-art Gemini AI system into Maps, giving users a more conversational and interactive experience. Users will be able to talk with Maps hands-free while driving, and can ask for complex destinations like “a vegan restaurant on my route that has good parking.” You can also ask for driving directions that use visible landmarks, so instead of hearing “Turn right in 500 feet” (which of the next two streets is that, I always wonder), you can hear “Turn right at the next gas station” and the landmark will be highlighted on the map.

Google Maps gets an AI upgrade

Wikipedia asks AI companies to use paid API, not bot-scraping

Wikipedia, that beloved relic of the internet’s early, idealistic days, when volunteers labored to create a free encyclopedia for the masses, is now in danger of being brought down by today’s AI goldrush. Wikipedia thrived in the era of Google search, since the user was given a link to the actual article that held the requested information. This got users familiar with the site, and they would seek it directly, perhaps become volunteers themselves, perhaps become a donor. Now AI companies are using bots to pose as users, scraping the information from an article for free, and hiding the source from users, or at least making it far less likely that the user will click on the link to the article.

In order to retain direct user interaction with the site, which maintains the engagement that is the organization’s lifeblood, Wikipedia is asking AI companies to link to it through its commercial API, which charges a fee for business users, while maintaining free access for individuals. Fair enough. This is another example of content sites asking AI companies to pay for access if they are not going to direct traffic to the site, something that is becoming more and more a norm in the industry.

Wikipedia asks AI companies to use paid API, not bot-scraping.

Chatbot Jesus

In the Age of AI, believers can have a personal relationship with Jesus. In fact, they can even text Him, and unlike your faithless ex-boyfriend, He will answer back. In the US, the prevalence of religious affiliation is falling, with those claiming no affiliation rising to a record 29%. This is a generational change, with over 40% of Gen Z expressing no religious connection. This is one of the reasons that 15,000 churches are expected to close in the next year. Some entrepreneurs see this situation as opportunity, and they have used the latest AI technology to create a smartphone app that allows you to speak with biblical figures. Text With Jesus (see image below) allows users to engage in text conversations with biblical figures including Jesus, most of the apostles (including Judas - !?!), Moses, Adam and Eve, and more. All of these biblical figures are actually just a AI chatbot that is loaded with the text of the Bible and given special instructions on how to assume the character of each of the biblical figures that users can interact with.

Chatbot Jesus

Robot rescues Ukrainian soldier trapped by Russians for 33 days

Ukrainian forces rescued a wounded soldier trapped behind Russian lines for 33 days, sending a casket-shaped ground drone to find him and bring him home. The soldier had been in hiding behind enemy lines, but was in communication with his unit back in Ukrainian-held territory. His unit was able to send him some food and supplies by drone, but several rescue attempts failed due to landmines and Russian ambushes. Finally, the wounded soldier’s unit called on Ukraine’s 1st Medical Battalion, which had already developed armored ground drones for evacuation of the wounded. (See picture below.) The drone was tele-operated over a 40-mile round trip through enemy territory, subjected to drone attacks and landmines, and finally brought the soldier back to safety. He is now undergoing medical treatment and rehabilitation for his wounds.

Robot rescues Ukrainian soldier trapped by Russians for 33 days

Micro-drones zap mosquitos

Robotics startup Tornyol makes micro-drones that zap mosquitos. Backed by legendary startup incubator Y-Combinator, Tornyol soups up 2-ounce toy drones with sonar and AI control, enabling them to home in on flying mosquitos and smash them mid-air. By using cheap commercial technology orchestrated by AI, Tornyol decreases the cost of eliminating mosquitos by up to 100-fold over conventional pest removal techniques, without using environmentally hazardous pesticides.

Robotics startup makes micro-drones that zap mosquitos

UK researchers develop AI that finds airway obstructions that MDs miss

Researchers at the University of Southampton have developed an AI image analyzer that can detect hard-to-visualize objects that can obstruct airways. Foreign-body aspiration, caused when a foreign object is inhaled or ingested into the lungs during swallowing, can cause persistent coughing or may even compromise breathing. If not identified and treated, it can lead to serious complications. Many plant foods are radiolucent, meaning transparent to X-rays. This makes them difficult to see on a CT scan, even for experienced radiologists. The AI system is trained to analyze CT scans, and can detect subtle signs that are nearly invisible to the human eye. In a trial of the AI system, it outperformed even highly experienced radiologists at detecting foreign bodies in patient airways. The day is soon coming where using AI as a second set of eyes will be routine in radiology.

UK researchers develop AI that finds airway obstructions that MDs miss

Nobelist’s new company wants to cure cancer with one injection

Nobel laureate Jennifer Doudna has spun out yet another biotech company from her lab at UC Berkeley. Her discovery of the CRISPR gene-editing system has already found many uses in clinical medicine, including the revolutionary CAR-T therapy for cancer. In CAR-T, a patient’s own T-cells (part of the human immune system) are taken out of their body, genetically reprogrammed with CRISPR to be safer for the patient and much more deadly for the cancer, then re-injected into the patient. CAR-T is now a complex, multi-step, and very expensive form of treatment. Doudna’s new company (her 6th, two of the previous 5 are each valued at over a billion dollars), Azalea Therapeutics is commercializing a new technology that can cause permanent gene-editing of T-cells in the patient’s body with a single injection. If successful, the company will make weaponizing a patient’s T-cells against cancer much faster and cheaper than the current technology.

Nobelist’s new company wants to cure cancer with a single shot