The Age Of AI ‘Dumping’

🌈 Abstract

The article discusses the author's disappointment with Apple's new AI product, "Apple Intelligence," and their concerns about the broader trend of "AI dumping" by tech companies. The author argues that the high memory requirements of large language models (LLMs) make on-device AI a questionable choice, and that the free or heavily subsidized distribution of AI models by companies like Apple, Microsoft, OpenAI, and Google is distorting the market and could lead to an "AI winter" in the future.

🙋 Q&A

[01] Memory Requirements and On-Device AI

1. What are the author's concerns about the memory requirements of LLMs?

• The author is concerned that the memory requirements of LLMs, such as OpenELM with 1.1 billion parameters, are too high to be permanently reserved for an on-device AI assistant like Siri.
• Even with 4-bit quantization, the author believes that having only 16 possible states per artificial neuron is too restrictive.
• The author suggests that having an on-device AI is a bad idea and that it would be better to have the AI processing done in the cloud.

2. How does the author view Apple's approach to on-device AI?

• The author is disappointed that Apple is pursuing on-device AI with "Apple Intelligence" and believes that everything shown could be done with other AI models, including integrations with ChatGPT.
• The author questions why Apple is developing its own AI models instead of just outsourcing the AI to ChatGPT or other existing solutions.

[02] AI Dumping and Market Distortion

1. What is the author's concern about "AI dumping"?

• The author argues that tech companies like Apple, Microsoft, OpenAI, and Google are engaging in "AI dumping" by giving away access to their AI models for free or far below market cost.
• The author believes this distorts the market and reduces the incentive for companies to improve AI technology and services, similar to how a free Netflix service would reduce the incentive to improve streaming technology.
• The author is concerned that this could lead to an "AI winter" if companies suddenly pull out of the AI market or start charging market rates for their AI services, leaving developers and customers without suitable alternatives.

2. How does the author view the potential consequences of AI dumping?

• The author suggests that the worst-case scenario is that companies could suddenly pull out of the AI market or start charging market rates, causing a shock to the economy and forcing developers to completely retool their workflows.
• The author cites a report from Business Insider that tech companies plan to spend over $1 trillion on AI, but the return on investment may be disappointing, making the current AI spending unsustainable.
• The author believes that this could lead to a situation where the only innovation in the AI space comes from the giant tech companies, which are not great at innovating.