The rapidly decreasing costs of AI and the acquisition of Mosaic by Databricks are driving significant changes in the industry. Mosaic's vision to make it cost-effective for companies to train and fine-tune their own models aligns perfectly with Databricks' goal of helping companies rapidly adopt machine learning to outpace the competition. This trend is fueled by two factors: algorithmic improvements from companies like MosaicML and the decreasing costs of GPUs.

The rapidly decreasing costs of AI and the acquisition of Mosaic by Databricks are driving significant changes in the industry. Mosaic's vision to make it cost-effective for companies to train and fine-tune their own models aligns perfectly with Databricks' goal of helping companies rapidly adopt machine learning to outpace the competition. This trend is fueled by two factors: algorithmic improvements from companies like MosaicML and the decreasing costs of GPUs.

Training costs have decreased by 10x in less than a year. For example, it now costs $50k to train stable diffusion and $200k to train a high-quality LLM. MosaicML has published several papers demonstrating their ability to train stable diffusion for only $50k. This significant reduction in training costs is driven by the efficiency of algorithms and the decreasing prices of GPUs.

In August 2019, the cost of a Nvidia T4 GPU for one hour was $0.95, but today it is only $0.35 per hour. This 3x decrease in GPU costs over the span of three years has contributed to the affordability of training models. As a result, inference costs have also dropped by 10x in 16 months, from $0.006 per 1k tokens for Curie generations to $0.0005 per 1k tokens for Curie quality generations today.

These decreasing costs have significant implications for SaaS companies. Most SaaS companies target 80% gross margins, with COGS accounting for 20% of revenue. Cloud costs typically make up 50% of COGS for software companies, meaning that cloud costs should only account for around 10% of revenue. However, some companies are spending as much on cloud costs as they are on generative AI features, which are additional expenses. The decreasing costs of AI training and inference models will help alleviate this financial burden for companies.

Furthermore, the decrease in training costs will lead to more model providers and increased competition at the model layer. This pricing pressure may also prompt more companies to opt for open-source model providers, as they become more cost-effective and accessible.

In addition to the decreasing costs of AI, there is another challenge that individuals face when it comes to utilizing their highlights effectively: Highlight Dementia. Highlight Dementia refers to the frustration of going back through highlights and realizing they no longer hold meaning or value. This issue arises when people lack a reading strategy beyond simply highlighting without context.

To prevent Highlight Dementia, it is crucial to add context to your highlights. When making a highlight, try to note down a couple of words explaining why you are highlighting a particular sentence. This context can be in the form of what the sentence makes you think of, any confusion it may cause, or if someone else would find it interesting. Creating context around your highlights goes beyond mere markup; it enriches the meaning and relevance of the highlighted text.

There are two approaches to reading that can help prevent Highlight Dementia: Synchronous Active Reading and Asynchronous Active Reading. Synchronous Active Reading involves having index cards, a notebook, or a database open alongside the text, allowing you to immediately write down quotes and thoughts. On the other hand, Asynchronous Active Reading involves reading with only a book or paper and a pen or highlighter. With this approach, you need to revisit your highlights later and transfer them to a better storage medium, such as a permanent note storage.

It is crucial to actively read and create context around your highlights while reading. Additionally, ensure that you transfer your notes and highlights to a permanent storage medium, either immediately or by setting a reminder for later. Liberating your highlights from their original location and enriching them with personal context is essential for effectively utilizing them.

In conclusion, the decreasing costs of AI and the acquisition of Mosaic by Databricks are driving significant changes in the industry. The combination of algorithmic improvements and decreasing GPU costs has led to a 10x reduction in training costs. This trend will continue to place pricing pressure on closed-source model providers, leading to increased competition and the potential for more companies to adopt open-source models. Additionally, individuals must be conscious of preventing Highlight Dementia by adding context to their highlights and utilizing active reading strategies. By actively reading, creating context, and transferring highlights to a permanent storage medium, individuals can maximize the value of their reading and prevent the frustration of meaningless highlights.