Davos Memo: 10 things you need to know about AI

The following is a guest post by John deVadoss.

Davos in January 2024 was all about one topic: AI.

The vendor was selling AI. Sovereign nations were promoting AI infrastructure. intergovernmental The organization was deliberating about the regulatory implications of AI. Business leaders were exaggerating the promise of AI. Political heavyweights were debating the impact of AI on national security. Almost everyone I met on the Main Promenade was speaking fluently about AI.

Nevertheless, there was an air of hesitation. Was this the real deal? Here are 10 things you need to know about AI: the good, the bad, and the ugly, based on several of my presentations at Davos last month.

1. The correct term is “generative” AI. Why “create”? While previous waves of AI innovation were all based on learning patterns in data sets and the ability to recognize those patterns when classifying new input data, this wave of innovation is based on learning large models (aka ‘collections of patterns’). do. , you can use these models to creatively generate text, video, audio, and other content.
2. No, generative AI does not hallucinate. When a large, previously trained model is asked to generate content, it does not always include a completely complete pattern to direct its generation. If the learned pattern is only partially formed, the model has no choice but to ‘fill in the blanks’, and what are observed as so-called hallucinations occur.
3. As some of you have observed, the output produced is not necessarily repeatable. why? This is partly because generating new content from learned patterns involves randomness and is an inherently probabilistic activity. This is just a fancy way of saying that generative AI output is not deterministic.
4. In fact, non-deterministic content creation lays the foundation for a core value proposition when applying generative AI. The best place to use it is in use cases that involve creativity. If creativity is not required or required, the scenario is likely not suitable for generative AI. Use this as a litmus test.
5. Creativity in the little things provides a very high level of precision. A good example is the use of generative AI in software development to generate code that developers use. All in all, creativity allows generative AI models to fill very large gaps. This is why, for example, we tend to see miscitations when asked to write a research paper.
6. A common analogy for generative AI is the Oracle of Delphi. The trust’s statement was vague. Likewise, generative AI output is not necessarily verifiable. Ask questions about generative AI. Don’t delegate transactional tasks to generative AI. In fact, this analogy extends beyond generative AI to all AI.
7. Paradoxically, generative AI models can play a very important role in the realm of science and engineering, even though they are not typically associated with artistic creativity. The key here is to combine a generative AI model with one or more external validators that are responsible for filtering the model’s output, until the combined system generates these validated outputs as new prompt inputs for subsequent creativity cycles. It is used as. desired result.
8. The widespread use of generative AI in the workplace will lead to a modern-day disruption. Between those who use generative AI to exponentially improve their creativity and performance, and those who give up their thought processes to generative AI and become increasingly marginalized and inevitably dismissed.
9. The so-called public model is largely tainted. All models trained on the public internet are further trained on content at the far end of the web, including the dark web and more. This has serious implications. One is that the model may have been trained on illegal content, and the second is that the model may have been infiltrated with Trojan horse content.
10. The concept of guardrails in generative AI is fatally flawed. As mentioned in the previous point, once a model is contaminated, there is almost always a way to creatively induce the model to bypass the so-called guardrails. A better approach is needed. A safer approach; It’s about driving public trust in generative AI.

As we witness the use and misuse of generative AI, we must look inward and remind ourselves that AI is a tool, nothing more, nothing less, and we must look to the future and shape the tool appropriately. Tools shape us.

The post Notes from Davos: 10 Things You Need to Know About AI appeared first on CryptoSlate.