<aside> <img src="/icons/help-alternate_gray.svg" alt="/icons/help-alternate_gray.svg" width="40px" /> Here are slides to support a session on Semantic Machines run by Dominik Lukeš.

</aside>

Slides

https://docs.google.com/presentation/d/1XbS9J0JTuaWJt-iqIcFZALCWn8ydFZ2y/edit?usp=sharing&ouid=110777908713167103295&rtpof=true&sd=true

Notes

Notes on the types of semantics

• Logical Semantics: This deals with the truth or falsehood of statements. It focuses on concepts like TRUE, FALSE, AND, OR, which are fundamental to logic and reasoning. AI models that employ logical semantics can learn and operate based on predefined rules.
• Representational Semantics: This type of semantics concerns what words or statements refer to. It involves associating words with their meanings, like recognizing that "cat" refers to a feline animal, "run" describes a type of movement, and "red" signifies a color. AI models using representational semantics can learn from dictionaries or visual aids to understand the meaning of individual words.
• Relational Semantics: This focuses on how words relate to each other within a sentence or context. It involves understanding the function of words like "get," "the," "how," "whom," and "respect," which often indicate relationships between different entities or concepts. AI models utilizing relational semantics can only learn these complex relationships through exposure to extensive contextual information.

The presentation also suggests that these three types of semantics correspond to different fields of study:

• Logical semantics aligns with philosophy, likely due to its focus on truth and reasoning.
• Representational semantics relates to science, possibly because of its emphasis on defining and categorizing concepts.
• Relational semantics connects with linguistics, as it delves into the intricacies of language structure and meaning.

Notes at capabilities and limitations

• AI's Capabilities and Limitations
  • AI can perform various tasks like translation, summarization, and content generation. However, its abilities aren't consistent across all areas, creating a "jagged technological frontier". This means AI excels at some tasks while struggling with others that may seem similar in complexity.
  • AI models can be logical, representational, and relational. They can process language, translate, answer questions, and even generate creative text formats, like poems.
  • Large Language Models (LLMs), a type of AI, have limitations such as hallucination (producing false information), unpredictability, and a lack of planning and self-awareness.
  • It's crucial to be aware of these limitations and fact-check AI-generated content, especially in academic settings where accuracy is paramount.
• Using AI Effectively
  • Experimentation and learning from others are key to using AI tools effectively.
  • AI can be a useful assistant, but it's important to provide clear instructions, check its work, and remember that it can't learn or act independently. It needs human guidance.

Conclusion

• The presentation concludes by highlighting the importance of experimentation, learning from others, and understanding AI's limitations to effectively navigate the "jagged technological frontier" of AI capabilities.