Breaking GPT-4 Monopoly in 2024 with DeepSeek Reducing Training Costs

Explore 2024's AI breakthroughs: GPT-4 monopoly ends, multimodal LLMs rise, training costs plummet, and AI applications expand globally.

In 2024, we've made incredible progress.

The year 2024 was a milestone for generative AI development. We witnessed the rise of OpenAI Sora, the rapid decline in large model service prices, and the surge of open-source models catching up.

This rapid, multifaceted development fills us with confidence about the widespread application of new AI technologies in the next wave.

For engineers and researchers in the AI field, would they view this year differently?

As the new year begins, some have provided detailed summaries of the advancements in large-model technologies over the past year.

In 2024, we observed:

• GPT-4’s monopoly was completely broken.

• Some GPT-4-level models can now run on laptops.

• The cost of large model services collapsed due to competition and optimization.

• Multimodal vision became commonplace, with audio and video models emerging.

• Voice and real-time camera modes are transitioning from science fiction to reality.

• Building an app with a single prompt is now widespread.

• Universal access to the best models lasted only a few months.

• AI "agents" still seem a bit distant.

• Evaluation is truly critical.

• Apple Intelligence is disappointing, but the MLX library is outstanding.

• Emerging “reasoning” models are making waves.

• The best LLMs currently come from China, with training costs under $6 million.

• The environmental impact has improved but is also becoming more significant.

• Synthetic training data works effectively.

• LLMs have somehow become harder to use.

• Knowledge distribution remains uneven.

• LLMs need better criticism.