Leading AI Models

Last year, Meta's Llama 2 was seen as a simpler AI model, but the new model, Llama 3.1, released in July, has been reviewed as one of the best current models. Llama 3.1 is capable of outperforming many closed-source models in tasks like following instructions and solving math problems.

Now there are three major players in the large language AI model space. Meta’s Llama 3.1, ChatGPT-4 and Google’s Gemini are all trained on very large datasets to be able to find patterns in text that allow for tasks like answering questions, generating text, and summarizing information. Llama 3.1 uses one of the largest databases, which should make it one of the strongest. ChatGPT-4 is free and it is starting to move past just text inputs and responses by allowing users to input images, video, and voice. Google’s Gemini’s model is set up with a Mixture-of-Experts architecture. While other models use one network, Gemini has multiple smaller networks with different strengths which could lead to more efficiency.

I believe having multiple companies in the AI space will lead to more innovation. AI is evolving quickly, and with more competition for creative ideas, the technology will become stronger. Now it is important for companies as well as other players like universities and governments to think about ethical and safety concerns as AI grows and becomes more powerful.

Sources: https://www.wired.com/story/meta-ai-llama-3/ (July 23, 2024)

https://techfundingnews.com/metas-llama-31-vs-chat-gpt-4-vs-google-gemini-which-one-is-better/ (August 8.2024)

Image Source: https://techfundingnews.com/metas-llama-31-vs-chat-gpt-4-vs-google-gemini-which-one-is-better/

Neuralink:The ability to control technology with the brain

Elon Musk is back at it with exciting developments at his controversial company Neuralink. Neuralink will potentially enable humans to wirelessly link the brain to technological devices. The groundbreaking technology could revolutionize how we treat people with severe disabilities like quadriplegia. Neuralink’s product would interpret a person’s neural activity and transmit it through a phone or other device. At this point in time, Elon Musk and the others working on this product have tested it on monkeys and have recently implanted their first device in a human. Although this product could advance the medical industry, there are also some potential pitfalls. The brain is a very fragile part of the body and it is risky to tamper with it. Putting 1024 electrodes distributed across 64 threads could definitely cause some long term health issues. While many are upset about the use of monkeys in testing, I am much more concerned about hackers or other parties with bad intentions having control of someone’s brain. Despite all of the risks, Neuralink could significantly enhance the way that people with severe disabilities can function through day to day life.

GraphCast: The Future of Forecasting

The inconsistency and chaotic nature of weather makes it very difficult to predict. Similar to many other areas, Artificial Intelligence(AI) is revolutionizing the way we forecast weather. Over the past few years, Google has developed GraphCast, which has quickly become one of the most accurate forecasting systems. GraphCast uses data from the last 20 years to conceive more accurate weather predictions. First, GraphCast chooses more that a million grid points covering the surface of the earth. For each grid point, GraphCast predicts 11 different variables including temperature, humidity, wind speed, wind direction, and mean sea level on the surface. Not only does GraphCast predict day to day weather, it also predicts extreme weather events with more accuracy than other weather systems. GraphCast provided more accurate predictions on more than 90% of 1380 test variables compared to the HRES, which is a popular weather simulation system. GraphCast has started a new era of weather predictions, and will only improve as time goes forward. AI innovation continues to change the world.