In recent years, Artificial Intelligence (AI) has swiftly moved from a speculative technology to a core driver of innovation across various industries. This shift is fueled by an unprecedented explosion of data and significant advancements in computing power along with excellent scientific research. Its disruptive capabilities and massive potential for unlocking value are undeniable. If we trust research supported by McKinsey, then we are looking at an annual global GDP increase ranging from $2.6T to $4.4T. We have also seen some amazing productivity gain reports when Harvard Business School studied 758 BCG consultants to whom 18 realistic consulting tasks were given. As Harvard Business School observed, consultants who used AI (in this case, GPT-4) completed, on average, 12.2 percent more tasks, 25.1 percent quicker.

Training:

Compare it to teaching a new employee. You provide them with extensive historical data (training data), and they learn patterns, knowledge, and skills from this information

training is a one-time, resource-intensive process that requires significant computational power and time

Inference:

the new employee (now the LLM) doing their job day-to-day, applying what they learned during training to new, unseen work (user prompts)

inference is much faster and less resource-intensive than training. It's the practical use of the model to generate predictions, answers, or content based on new inputs

Anthropic recently launched their Claude 2.1 Large Language Model which has industry highest 200,000 token length. In the video I summarize of what they have launched and what does it mean for mankind.