AI was everywhere in 2024. Here are the biggest stories of the year.

OpenAI debuted AI models in 2024 called o1-preview and o1-mini that can tackle harder problems by working through solutions. The company offered o1-preview for general users, while o1-mini provided a faster, cheaper option for writing code, according to the company's official system card.

These tools mark an important shift in AI's capacity for reasoning, Northwestern University researchers explain. Instead of just giving quick answers like earlier AI, these new models work through problems step-by-step, more like how humans solve complex tasks.

OpenAI reported that tests showed significant improvements in the system's abilities. The model could solve 83% of complex math competition problems, while older versions only solved 13%, illustrating how this slower, more careful approach dramatically improved results. In analyzing chemistry, physics, and biology problems, the new model outperformed PhD-level scientists in problem-solving proficiency.   

However, these advances worry experts in the field. In an interview with Newsweek, computer science professor and AI pioneer Yoshua Bengio called the improvement in AI's reasoning and potential for deception "particularly dangerous" and called for better regulation.

Two Nobel Prizes recognized AI's growing impact on science in 2024, marking the first time artificial intelligence received such prestigious recognition. John Hopfield and Geoffrey Hinton received the Nobel Prize in physics for laying the groundwork for modern machine learning, while Google DeepMind CEO Demis Hassabis and researcher John Jumper shared half the chemistry prize for using AI to solve a 50-year-old protein structure problem, according to the Nobel Committee.

The awards highlighted AI's potential and risks should the technology fall into the wrong hands. Hinton, dubbed the "godfather of AI," uses his Nobel platform to warn about the technology's potential for "getting out of control," The New York Times reported. He had previously told MIT Technology Review that he was deeply concerned that AI will surpass a human's ability to learn, creating a superintelligence that could too easily cause widespread harm, manipulation, or warfare in the hands of bad actors. 

Hassabis compared AI's risks to climate change, telling The Guardian that "we can't afford the same delay with AI."

"This is more a Nobel moment for AI risk, rather than for AI itself," Bhaskar Chakravorti, dean of global business at Tufts University, wrote in Foreign Policy. He noted the prizes served as recognition of AI's transformative growth and a warning about its unchecked development.

Apple, Samsung, Microsoft, and Google began building AI directly into everyday devices, bringing it to broader audiences in 2024. Features like enhanced photos, text-to-speech tools, personalized recommendations, and myriad other AI optimizations helped integrate AI into everyday tasks.

Adoption varied widely, however. In CNBC's bi-annual survey of executives on its Technology Executive Council in October, 79% said their company was using Microsoft Copilot AI—though many questioned its $30 monthly per-user cost.

Google's Gemini chatbot attracted 42 million active users and 1.5 million developers since its May launch, according to the Business of Apps.

Apple followed with its own AI Intelligence that works across devices, though it's initially limited to its newest phones and computers with specific chips, the company announced in October. Samsung introduced Galaxy AI, which lets users translate conversations in real time and edit photos with simple taps, according to the company.

The focus shifted from standalone AI tools to integrated features, but questions about data security were raised. While Apple emphasized the importance of privacy through on-device processing, sending personal data to company servers could expose it to government agencies, employees, or bad actors, according to security experts The New York Times interviewed in June.

"We should be really looking at the cost benefit in terms of what we give up and what we get in exchange," said Hoff, who shares insights on AI and digital intelligence with his 14,000 TikTok followers.

"Tools like Google's suite are free because they're tracking our information to sell to advertisers. Now, large language models are aggregating all our data at once, putting everything into an algorithmic black box that nobody really knows how works."

Despite significant concerns, businesses increasingly embraced AI tools in 2024. "Business executives were, I think, resistant," Hoff said. "And now there's definitely an opening and receptiveness and a fear of missing out that exists on the private side."

When it launched in June 2024, Claude 3.5 Sonnet changed how coders work, quickly becoming a Silicon Valley favorite. The AI model solved 64% of coding problems in internal testing, according to Anthropic, the AI safety and research company behind Claude. The system could write new code and update old programs with fewer errors than previous versions, making it one of the most proficient—and popular—models powering AI-based software development tools such as Cursor.

In October, Anthropic announced new features that let Claude use computers similarly to humans. The model could move a mouse, click buttons, and read screens to complete tasks. Tests by GitLab, a major software development platform, showed a 10% improvement in development tasks with these updates, spurring debates about AI's growing role in software development and its impact on programming jobs. Many current top-scoring AI software development agents are already based on Claude. Letting the model control computers directly could unlock even further productivity gains for businesses.

In addition to coding, the model has been praised for its uncanny analytical ability and capacity to understand what users want. Many users have reported success using Claude as a sounding board to help them think through complex problems and make decisions in their personal lives. 

Skateboarding cats, dazzlingly beautiful crystals, and politicians in compromising situations are just some of the ways Flux, a new AI image generation system from Black Forest Labs, churned up viral buzz around its hyper-realistic images.

The system launched in August with $31 million in funding from Andreessen Horowitz. The model uses 12 billion parameters to create images, making it more powerful than previous systems, according to the company. Advancements in parameters like fractal dimension and relative smoothness all contribute to Flux's ability to render extraordinary detail in human features, deft animations, and high-quality images.

The model quickly drew the attention of the tech world. X (formerly Twitter) chose it to power image generation in its Grok-2 AI system, making the technology available to millions of users. However, this wide availability raised concerns about potential misuse, particularly around creating misleading images of political figures and spreading false content.

Story editing by Alizah Salario. Additional editing by Kelly Glass. Copy editing by Kristen Wegrzyn. Photo selection by Ania Antecka.