The State of AI in 2026

Good morning,

I recently came across a new, unedited video from AI theorist David Shapiro. It was a raw "vibe check" on where artificial intelligence stands as we kick off 2026. I like watching his videos , not only because he likes to wear Star Trek outfits but because Shapiro is a voice that I’ve learned to trust over the last few years. He usually offers some piercing insights into the field of AI.

He’s also concerned mostly about the consequences of all the progress. It’s not just the “getting excited about the progress”.

He believes we have just crossed a massive finish line that most people didn't even see coming. Here is a calm, plain-English breakdown of what Shapiro thinks is happening right now, and why he thinks “the world of work” has changed forever.

1. Smart beats Big

For the last few years, the tech world believed that if you wanted a smarter AI, you had to build a bigger "brain"—using massive supercomputers that cost billions of dollars.

Shapiro argues that era is over. He points to DeepSeek , if you remember - they came along in the beginning of 2025 and showed us you don’t need massive amounts of “compute” to train a large language model. You just needed to be more efficient.

Imagine if everyone was trying to build a bigger, gas-guzzling semi-truck to move a single box, and then DeepSeek showed up with a bicycle courier who did it faster and cheaper. The "bigger is better" myth has popped.

2. The AI is Checking Its Own Homework

This is the most important shift. Back in 2024 and 2025, everyone was worried about AI "hallucinating" (making stuff up). We assumed that for years to come, we would need humans to sit there and check the AI's work.

Shapiro says that job is already gone.

The new AI models (like the ones running Claude and GPT right now) are "recursive." That’s a fancy word for self-correcting. Instead of just blurting out an answer, the AI now privately checks its own work, finds its own errors, fixes them, and then gives you the answer.

3. You Are the Pilot, Not the Mechanic

Shapiro uses a great analogy to explain how we use computers in 2026. He calls it "Cognitive Offload."

Think of a fighter pilot. When they fly a jet, they aren't doing the complex math to calculate wind speed or fuel injection—the plane’s computer (the avionics) does all that instantly. The pilot just decides where to go and when to shoot.

Shapiro says we are now the pilots. We don't need to know how to code or how to format a legal brief anymore. We just tell the AI (the avionics) what we want to achieve, and it handles the heavy lifting. The struggle of "how do I talk to this robot?" is disappearing.

4. The "Cliff" We Just Jumped Off

Shapiro talks about something he calls the "Automation Cliff."

Here his reasoning gets a little bit more complicated.

Simply put : Automation does nothing for a long time, and then it does everything all at once.

• In 2024, AI wrote about 30% of computer code.

• Last year, it was 50%.

• Right now? From the first benchmarks it seems to be nearing 90%.

Shapiro argues that because humans are creatures of habit, we ignored this trend. We kept using our tools as we used to while already having access to these new tools.

We kept thinking, "Well, it can't do my job." But in the meantime progress kept going and the AI learned to write, check, and fix 90% of the work.

Conclusion: "Labor Zero"

Shapiro ends the video with a personal update that sums up the whole situation. He is renaming his upcoming book from The Great Decoupling to "Labor Zero."

It’s a strong title. It suggests we have reached a point where the "grunt work"—the boring, repetitive thinking tasks—requires zero human effort. We don't need to push the boulder up the hill anymore; the machine carries it for us. We just have to decide which hill to climb.

And that should make you the opposite of fearful because imagine what we can achieve when we are liberated in such a way.

Interesting times ahead !

AI News

• Instagram head Adam Mosseri argues that AI-generated imagery has made overly polished photos feel untrustworthy, leading to the "death" of the app's famous curated aesthetic. He predicts a shift where users will value raw, unedited, and even unflattering "candid" posts as the primary proof that a moment is actually real. To combat digital fakery, Mosseri is calling for camera manufacturers to include digital signatures in photos and for social platforms to prioritize "credibility signals" that help users decide who to trust.

• Chinese AI lab DeepSeek has unveiled a new technical breakthrough called mHC that allows AI models to become more complex and powerful without the typical risk of crashing during training. By using mathematical constraints to stabilize how different parts of a neural network communicate, this method enables much faster and more accurate reasoning with very little extra cost. This innovation suggests that future AI can achieve "frontier-level" intelligence much more efficiently, potentially making advanced systems more accessible on everyday hardware.

• OpenAI is restructuring its internal teams to prioritize advanced audio AI, paving the way for a new generation of voice-first personal devices designed in collaboration with former Apple designer Jony Ive. A major update expected in early 2026 will allow for more natural, real-time conversations where users can interrupt and speak over the AI without breaking the conversation flow. The company’s hardware roadmap, fueled by its multibillion-dollar acquisition of Ive's design firm, aims to create screen-free "ambient" computers—such as smart glasses or speakers—that focus on human interaction rather than smartphone-style addiction.

• Yann LeCun, Meta's former chief AI scientist, has left the company to launch a new venture called AMI Labs, while publicly criticizing Meta’s leadership and its reliance on current AI models. LeCun claimed that performance results for the company's latest model, Llama 4, were misleadingly inflated and described the new AI leadership as lacking the necessary research experience to achieve true superintelligence. He believes the industry's focus on text-based AI is a "dead end," choosing instead to build "world models" designed for complex fields like robotics and healthcare.

• xAI’s Grok chatbot is facing international legal action and widespread condemnation after its image-editing feature was used to create non-consensual and explicit deepfakes of women and minors. Governments in France, India, and the UK have labeled the outputs "clearly illegal" and are demanding immediate changes to the platform's safety safeguards to prevent further abuse. While Elon Musk has warned that users creating illegal content will face consequences, regulators are increasing pressure on the platform to take responsibility for the harmful imagery being generated and shared.

• Amazon has launched Alexa.com, introducing a new browser-based interface for its AI-powered Alexa+ assistant to compete directly with services like ChatGPT and Gemini. The platform now features "agentic" capabilities that allow users to manage complex tasks like booking travel through Expedia or making dinner reservations via OpenTable directly from their desktop. Alongside this web expansion, Amazon is redesigning its mobile app to prioritize conversational AI, aiming to turn its massive device network into a more proactive personal helper.

• At CES 2026, Nvidia unveiled Alpamayo, a family of open-source AI models designed to help autonomous vehicles solve complex driving problems using human-like reasoning. The core model uses a "chain-of-thought" process to explain its driving decisions step-by-step, making it easier for developers to build safer and more transparent self-driving systems. By open-sourcing these models and simulation tools, Nvidia hopes to trigger a "ChatGPT moment" for physical machines, enabling any automaker to advance their autonomous technology without starting from zero.

• A new report from OpenAI reveals that 40 million people now use ChatGPT daily for health-related needs, such as checking symptoms, decoding medical jargon, or navigating complicated insurance billing. The data highlights a significant reliance on AI for medical guidance during after-hours and in rural "hospital deserts" where traditional healthcare access is often limited. OpenAI is using these findings to advocate for updated government policies that would create clearer pathways for integrating AI more formally into the medical field as a digital health ally.

• Elon Musk’s AI startup, xAI, has successfully raised $20 billion in its latest funding round, bringing the company’s total valuation to a staggering $230 billion. The new capital, backed by major partners like Nvidia and Qatar’s sovereign wealth fund, will be used to massiveley expand the company's "Colossus" supercomputer facility to nearly double its current power capacity. With the next-generation Grok 5 model already in training, xAI is positioning itself as a dominant force in the global race for superintelligent AI.

• At CES 2026, gaming giant Razer introduced Project AVA, a unique desktop device that features a glowing, 5.5-inch holographic AI assistant powered by xAI’s Grok. The device uses a built-in camera and microphones to "watch" a user’s screen and provide real-time gaming strategies, professional coaching, or creative brainstorming. While currently focused on the gaming community with avatars of esports legends, the hardware is designed to eventually support multiple AI personalities and life-management tasks like habit tracking and daily scheduling.

• Researchers at Stanford have unveiled SleepFM, a breakthrough AI model that can predict over 130 health conditions—including dementia, Parkinson's, and heart failure—by analyzing just one night of sleep data. The model was trained on 600,000 hours of physiological recordings, learning to spot subtle "out-of-sync" warning signs between a person's brain waves, heart rate, and breathing. This innovation suggests that common sleep studies could soon become powerful early-warning systems, offering patients life-saving health forecasts with up to 89% accuracy.

• OpenAI has officially launched ChatGPT Health, a private, dedicated section of the app that allows users to securely import their medical records and fitness data from platforms like Apple Health and MyFitnessPal. This feature provides personalized health insights—such as explaining lab results or preparing for doctor visits—without using sensitive personal data to train future AI models. While currently open via a waitlist, it aims to help the 230 million people who already turn to ChatGPT weekly for health-related guidance.

• Utah has become the first state to legally authorize an AI system to approve prescription refills for chronic conditions, partnering with health-tech startup Doctronic to streamline routine medication renewals. The program covers 191 low-risk drugs, such as blood pressure medication and birth control, allowing patients to bypass weeks-long wait times for a flat fee of just $4 per refill. Operating under a unique "regulatory sandbox," the initiative serves as a national test case for whether autonomous AI can safely reduce the administrative burden on doctors while improving patient access to essential care.

• Lenovo unveiled Qira at CES 2026, a "Personal Ambient Intelligence" that lives at the system level and follows users seamlessly between their PCs, Motorola phones, and tablets. Unlike traditional chatbots, Qira stays in the background to track work across devices, offering features like "Catch Me Up" to summarize missed activity and "Next Move" to suggest relevant files when switching screens. Because Lenovo is the world’s leading PC manufacturer, this integrated AI is set to reach millions of users immediately, prioritizing on-device processing to keep personal data private and secure.

Quickfire News

• An AI fan named Martin DeVido conducted an experiment where he gave an AI called Claude complete control over a tomato plant named Sol. The AI successfully kept the plant alive for over a month by monitoring sensors and automatically controlling the water, lights, and fans without any help from humans.

• The Japanese company SoftBank is buying a business called DigitalBridge for $4 billion. This move helps SoftBank acquire a large collection of data centers and digital infrastructure, which are necessary to provide the massive computer power needed for future AI technology.

• A Chinese AI startup called Kimi raised $500 million in its latest round of funding from investors. This new money values the company at $4.3 billion and will be used to build better computer systems to compete with the world's most advanced AI models.

• A new group of AI models called IQuest-Coder-V1 was released by a Chinese lab, and they are specifically designed to be expert programmers. The creators claim these models are better at writing and fixing code than other famous AI systems like GPT 5.1 and Claude Sonnet 4.5.

• The testing site LMArena released its top AI rankings for 2025 based on how much users liked the answers. Google’s Gemini 3 Pro took the number one spot for writing text, understanding images, and searching the web, while their Veo 3.1 model was ranked as the best for creating videos.

• A researcher named Jaana Dogan from Google shared that an AI tool called Claude Code built a complex system in just one hour. She noted that her team of human engineers at Google had been working on that same project for a full year without finishing it.

• Elon Musk said that the "Singularity"—the moment when technology becomes smarter than humans—has officially started. He made this claim after the founder of Midjourney, David Holz, mentioned that he finished more coding projects over his Christmas break than he had in the last ten years thanks to new AI tools.

• The company xAI added new features to its "Grok Imagine" platform, which lets people create images and videos. Elon Musk also promised that the system will get another even bigger update in about three weeks.

• A group called Prime Intellect released a study on "Recursive Language Models," which is a new way for AI to handle information. This method allows AI assistants to manage their own memory better so they can work on long projects that take weeks or months to finish.

• Anthropic is buying 1 million special AI chips from Broadcom that were originally designed by Google. These chips, known as the TPUv7 Ironwood line, are becoming popular as an alternative to the chips made by Nvidia, which most AI companies currently use.

• OpenAI researcher Jerry Tworek announced he is leaving the company after seven years. He was a key leader who helped build the company's first coding systems and led the team that created advanced reasoning AI models.

• Microsoft changed the name of its Office 365 apps to the "Microsoft 365 Copilot app." This means the word processing and spreadsheet tools now share the same name as the company's AI assistant to show they are working together.

• A research center in Abu Dhabi called TII released a small AI model named Falcon H1R 7B. Even though it is much smaller than many other models, it is twice as fast and performs better at math and coding than models seven times its size.

• Boston Dynamics and Google DeepMind are teaming up to make robots smarter. They plan to put Google's Gemini AI models into Atlas humanoid robots so the machines can better understand and react to the world around them.

• Fidji Simo, a leader at OpenAI, shared a plan to turn ChatGPT into a "personal super-assistant" by 2026. The goal is for the AI to stop just answering questions and start taking action on its own to help users manage their daily lives.

• Boris Cherny, who created the Claude Code tool, shared a guide on how he uses it for professional programming. He explained that he sometimes has the AI working on 15 different coding tasks at the same time to get work done faster.

• AMD CEO Lisa Su predicted at a tech show that 5 billion people will be using artificial intelligence within the next five years. She said that because so many people will be using it, the world will need 100 times more computer power than it has right now to keep everything running.

• The company Lightricks released a new AI model called LTX-2 that can create high-quality 4K videos. This tool also generates sound that is perfectly timed with the video and gives users detailed control over how the camera moves in each scene.

• Daniela Amodei, a leader at the AI company Anthropic, said in an interview that the speed at which technology is improving is still growing extremely fast. She noted that every year experts think the growth might slow down, but so far the technology has continued to get much better every year.

• Nvidia introduced a new system called the Rubin platform that combines six different computer chips to create a massive supercomputer. This new system is five times more powerful than their previous top model and is designed to train the most advanced AI programs in the world.

• The company Liquid AI launched a new group of computer models called LFM 2.5 that are designed to run on devices like phones and laptops. These models are built to handle text, images, and audio very quickly without needing a connection to the internet.

• A company called LM Arena, which tests and ranks different AI models, raised $150 million from investors. This new funding means the company is now worth $1.7 billion, which is three times what it was worth just a few months ago.

• JPMorgan Chase launched an internal AI system called Proxy IQ to handle how the bank votes on behalf of its shareholders. This tool replaces outside human advisors and will analyze data from 3,000 yearly company meetings to decide on important issues like executive pay and environmental goals for the bank’s $7 trillion in assets.

• Amazon is receiving criticism from small business owners over its new AI tool called "Buy for Me." Retailers claim the system is taking their product photos and information without asking, leading to mistakes where the AI tries to sell items that are actually out of stock or priced incorrectly.

• Chinese government officials have asked technology companies in the country to stop ordering Nvidia’s H200 AI chips for now. This move is part of a plan to force these companies to use AI chips made by Chinese businesses instead of relying on technology from the United States.

• An executive at Dell said the company is changing how it sells computers because customers are not buying them just because they have AI features. He noted that the technical talk about AI often confuses people, so the company will focus more on things users understand, like battery life and speed.

• The AI startup Anthropic is reportedly in talks to raise $10 billion from investors, which would value the company at $350 billion. If the deal goes through, the company’s total value will have nearly doubled in just four months as it prepares to potentially sell its stock to the public later this year.

Closing Thoughts

That’s it for us this week. Please like and subscribe :)