According to Dario Amodei, an AI researcher and entrepreneur, AI coding may match the best human coders by late 2026. He believes that AI systems will reach a very serious level of coding capabilities by the end of 2025.

Amodei acknowledges that the prospect of AI surpassing human coders is both exciting and threatening. On one hand, he recognizes the intellectual achievement of human coding, which is a source of pride and identity. However, he also acknowledges that it is "wrong to tell people that [AI surpassing human coders] is coming or to try to sugarcoat it."

Amodei further suggests that 2026 or 2027 could be the threshold moment for the development of Artificial General Intelligence (AGI), which would have profound economic, societal, and security implications. He cautions that this timeline is not set in stone, but the risk of such rapid progress in AI during a geopolitically uncertain time is a cause for concern.

Overall, Amodei's predictions highlight the rapid advancements in AI capabilities, particularly in the realm of coding, and the need for careful consideration of the implications of these developments.

According to leading AI researchers and executives, 2026-2027 could be a pivotal period for the development of Artificial General Intelligence (AGI):

• Dario Amodei, co-founder of Anthropic, believes AI coding may match the best human coders by late 2026, and AGI could be achieved by 2026-2027.
• Elon Musk predicts AGI will likely be achieved by 2026 at the latest, enabling capabilities like self-driving cars and virtual assistants that can perform almost any task.
• Demis Hassabis, co-founder of DeepMind, estimates AGI is 3-5 years away, with key capabilities like reasoning, long-term memory, and creative invention still needing development.
• Mark Zuckerberg aims for AI to handle half of Meta's coding by 2026.
• Eric Schmidt believes the "San Francisco consensus" is that AGI will be achieved within 3-5 years, with major breakthroughs in programming, math, and reasoning in the next 1-2 years.
• Nvidia CEO Jensen Huang outlined roadmaps for their next-generation "Hopper" and "Ampere" AI hardware platforms launching in 2026-2027, promising massive performance and efficiency gains.

However, not all experts agree on the timeline. Yan LeCun, a prominent AI scientist, believes AGI is at least 5-6 years away, cautioning that the field has historically underestimated the difficulty of achieving human-level AI.

Overall, the consensus among industry leaders points to rapid AI progress in the next 3-5 years, potentially culminating in the emergence of AGI-level capabilities by 2026-2027, with significant implications across industries and society.

DeepMind has stated that in 2026, they are aiming to combine their Gemini AI and ViLO (Vision-Language-Object) models. The goal is to create multimodal models that can truly understand the physical world.

Gemini was designed from the start to be a multimodal system, supporting Google's vision of a universal digital assistant that can help in real-world contexts. This aligns with the broader industry trend towards "omni-models" that can handle a variety of media types like text, images, audio, and video.

Meanwhile, ViLO learns about the physical world by analyzing large amounts of YouTube videos and other online content. By integrating Gemini and ViLO, DeepMind hopes to develop models that have a deep, holistic understanding of the real-world environment.

This convergence of multimodal and physical world understanding capabilities is seen as a key milestone towards more advanced, general-purpose AI systems in the coming years.

Mark Zuckerberg, the co-founder and chairman of Meta (formerly Facebook), has spoken recently about his expectations for AI in 2026. He wants AI to do half of Meta's coding by 2026.

At Llamicon, Zuckerberg had a discussion with Satya Nadella, the CEO of Microsoft, about the future of AI and coding. Zuckerberg stated:

"The big one that we're focused on is building an AI and a machine learning engineer to advance the llama development itself, because I mean our bet is sort of that in the next year probably, you know, I don't know, maybe half the development is going to be done by AI as opposed to people, and then that will just kind of increase from there."

This indicates that Zuckerberg believes AI will play a significant role in Meta's coding and software development processes, with the potential for AI to handle up to 50% of the company's coding tasks by 2026.

According to the predictions made by various AI experts and industry leaders, we can expect to see significant advancements in AI capabilities within the next 3-5 years:

• Programming: Eric Schmidt, the former CEO of Google, believes that in the next year, the vast majority of programmers will be replaced by AI programmers. He also predicts that within 2 years, we'll have AI systems that can perform at the level of top graduate-level mathematicians.

• Mathematics: Eric Schmidt suggests that within 3-5 years, we'll have AI systems that can perform at the level of the smartest human mathematicians, physicists, artists, writers, and thinkers.

• Reasoning: Demis Hassabis, the co-founder and CEO of DeepMind, believes that while current AI systems are quite capable, they still lack certain attributes like robust reasoning, hierarchical planning, and long-term memory. He expects that we'll achieve AGI (Artificial General Intelligence) capable of exhibiting all the cognitive capabilities of humans within the next 3-5 years.

• Creativity: Demis Hassabis also notes that a key benchmark for AGI is the ability to invent new hypotheses and conjectures, like Einstein did with relativity, rather than just proving existing ones. He believes we're still a few years away from achieving this level of creative and inventive capability in AI.

Overall, the consensus among these AI experts is that we're on the cusp of a major breakthrough in AI capabilities, with systems surpassing human-level performance in a wide range of cognitive tasks within the next 3-5 years. However, some, like Yan LeCun, caution that the timeline may be longer and that we shouldn't underestimate the difficulty of achieving true AGI.

Nvidia CEO Jensen Huang provided details on Nvidia's roadmap for their next-generation AI platforms, Ruben and Vera Ruben, slated for release in the second half of 2026 and 2027 respectively.

Ruben is Nvidia's new GPU architecture, featuring a custom CPU called Vera designed for advanced AI applications. Ruben is expected to deliver significant performance improvements over Nvidia's previous Blackwell chips, particularly for AI training and inference.

Some key highlights:

• Vera Ruben CPU: Twice the performance of Grace, with more memory and bandwidth, yet only 50W power consumption.
• Ruben GPU: Brand new CX9 GPU, along with new networking (NVLink 6), and memory (HBM4).
• Vera Ruben NVLink 144: Connects to 144 GPU dies in the second half of 2026.
• Ruben Ultra (2027): Scales up to 15 exaflops of performance, 4.6 PB/s of bandwidth, and 300 racks with 2.5 million parts.

Nvidia is focused on scaling up performance and efficiency before scaling out, with Ruben delivering a 900x increase in flops over the current Blackwell platform. This roadmap demonstrates Nvidia's commitment to advancing AI hardware capabilities in the coming years.

One of the key predictions for 2026 is the emergence of AI models that are no longer stateless, but actively learning and improving through experience. Aidan Gomez, the co-author of the influential 2017 research paper "Attention Is All You Need," believes that we will see breakthroughs in continual learning, where AI models can learn from their interactions and mistakes, rather than being frozen in time.

Gomez explains that the current status quo is that AI models are built with significant investment, but then remain static, unable to learn and grow. However, he anticipates that this will change in the near future, where models will be able to learn from their experiences, much like humans do. When a model fails at a task, it will be able to learn from that failure and remember the correct way to perform the task, continuously improving over time.

This shift towards continual learning is expected to unlock dramatic improvements in the value and capabilities of AI systems. Rather than being limited to a fixed knowledge base, these models will be able to adapt and grow, providing more personalized and effective assistance to users. This could lead to significant advancements in areas such as enterprise applications, where AI-powered assistants can learn and improve alongside the users they serve.

Overall, the transition towards continually learning AI models is seen as a crucial step in the evolution of artificial intelligence, allowing these systems to become more dynamic, adaptable, and valuable in real-world applications.

According to Imad Mustaq, the founder and former CEO of Stability AI, by next year (2026) we should be able to get an "01 level model" on a smartphone that only consumes 20 watts of electricity. This is comparable to the power consumption of the human brain.

Mustaq explains that the cost of these powerful AI models will be just a few pennies per unit of intelligence, compared to the massive energy requirements and infrastructure costs that would have been needed in the past. He notes that the efficiency curve for AI is rapidly improving, making it feasible to run advanced AI systems on low-power devices like smartphones.

This prediction challenges the notion that AI development will require massive data centers and energy-hungry hardware. Instead, Mustaq believes we will see a dramatic shift towards highly efficient, low-power AI models that can be deployed widely, even on consumer devices. This could unlock new applications and use cases for AI technology in the coming years.

While many prominent AI leaders have predicted significant advancements in AI capabilities by 2026, Yan Lecun, a renowned AI scientist, offers a more cautious perspective. Lecun believes that achieving true artificial general intelligence (AGI) - a system that matches human-level intelligence across a wide range of tasks - is likely still 3 to 5 years away, even in the best-case scenario.

Lecun acknowledges the rapid progress in AI, including advancements in language models and other AI architectures. However, he cautions against the overly optimistic timelines proposed by some industry figures. Lecun notes that the history of AI is marked by repeated underestimations of the challenges involved in developing truly general intelligence.

He emphasizes that for AGI to be achieved within the next 5-6 years, a number of critical factors must align, including the successful development and scaling of the latest AI techniques, continued hardware advancements, and the absence of any major unforeseen obstacles. Lecun suggests that a more realistic timeline for AGI may be in the range of 5-6 years, rather than the 1-2 year predictions made by some AI leaders.

This cautionary perspective serves as a counterbalance to the more optimistic forecasts, highlighting the inherent uncertainties and challenges in the pursuit of artificial general intelligence. While the AI field continues to make rapid progress, Lecun's view underscores the need for a measured and realistic assessment of the timelines involved in achieving this ambitious goal.