Artificial intelligence (AI) has become a buzzword in recent years, with more and more companies investing in AI technology to improve their businesses. However, for those who are not familiar with the field, the jargon used by AI insiders can be overwhelming. In this article, we will break down some of the most commonly used terms in AI and explain their meanings.
AGI
AGI stands for “artificial general intelligence.” It is a concept used to describe a significantly more advanced AI than what is currently possible. AGI would be able to do most things as well as or better than most humans, including improving itself. For some, AGI is equivalent to a median human that you could hire as a coworker, and they could do anything you would be happy with a remote coworker doing behind a computer.
AI Ethics and Safety
AI ethics refers to the desire to prevent AI from causing immediate harm. It often focuses on questions like how AI systems collect and process data and the possibility of bias in areas like housing or employment. AI safety describes the longer-term fear that AI will progress so suddenly that a super-intelligent AI might harm or even eliminate humanity.
Alignment
Alignment is the practice of tweaking an AI model so that it produces the outputs its creators desired. In the short term, alignment refers to the practice of building software and content moderation. But it can also refer to the much larger and still theoretical task of ensuring that any AGI would be friendly towards humanity. The dataset used to align these systems and determine whose values and bounds are set by society as a whole, by governments.
Emergent Behavior
Emergent behavior is the technical term used to describe when some AI models show abilities that were not initially intended. It can also describe surprising results from AI tools being deployed widely to the public.
Fast Takeoff or Hard Takeoff
Fast takeoff or hard takeoff is a phrase that suggests that if someone succeeds in building an AGI, it will already be too late to save humanity. AGI could happen soon or far in the future, and the takeoff speed from the initial AGI to more powerful successor systems could be slow or fast.
GPU
The chips used to train models and run inference, which are descendants of chips used to play advanced computer games, are called GPUs. The most commonly used model at the moment is Nvidia’s A100.
Guardrails
Guardrails are software and policies that big tech companies are currently building around AI models to ensure that they do not leak data or produce disturbing content, often called “going off the rails.” It can also refer to specific applications that protect the AI from going off-topic, like Nvidia’s “NeMo Guardrails” product.
Inference
Inference is the act of using an AI model to make predictions or generate text, images, or other content. Inference can require a lot of computing power.
Large Language Model
A large language model is a kind of AI model that underpins ChatGPT and Google’s new generative AI features. It uses terabytes of data to find the statistical relationships between words, which is how it produces text that seems like a human wrote it.
Paperclips
Paperclips are an important symbol for AI safety proponents because they symbolize the chance an AGI could destroy humanity. It refers to a thought experiment published by philosopher Nick Bostrom about a “superintelligence” given the mission to make as many paperclips as possible. It decides to turn all humans, Earth, and increasing parts of the cosmos into paperclips. OpenAI’s logo is a reference to this tale.
Singularity
Singularity is an older term that refers to the moment that technological change becomes self-reinforcing, or the moment of creation of an AGI. It is a metaphor, and literally, singularity refers to the point of a black hole with infinite density.
Stochastic Parrot
Stochastic Parrot is an important analogy for large language models that emphasizes that while sophisticated AI models can produce realistic seeming text, that the software does not have an understanding of the concepts behind the language, like a parrot.
Training
Training is the act of analyzing enormous amounts of data to create or improve an AI model.
Understanding the jargon used by AI insiders is essential for anyone who wants to engage with the AI community. The terms we have discussed in this article are just the tip of the iceberg, and there is much more to learn. However, by learning these terms, you will be better equipped to understand the latest developments in AI and contribute meaningfully to the conversation.
Leave a Reply