AI: Difference between revisions
| Line 1: | Line 1: | ||
{{DISPLAYTITLE:Artificial intelligence}} | {{DISPLAYTITLE:Artificial intelligence}} | ||
'''AI''' (or '''Artificial Intelligence''') is an umbrella term for any script, program or robot which mimics human | '''AI''' (or '''Artificial Intelligence''') is an umbrella term for any script, program or robot which mimics human behavior. There are many different kinds of AI, most of which fall under the category of Generative AI, but some may use sophisticated algorithms written mostly by humans in order to mimic human behavior. | ||
== Types of non-generative AI == | == Types of non-generative AI == | ||
=== Assistants === | === Assistants === | ||
Traditional assistants listen to a set of commands in order to help in doing things on a phone or computer. Examples of assistants include Siri, Google Assistant, Alexa, Bixby and | Traditional assistants listen to a set of commands in order to help in doing things on a phone or computer. Examples of assistants include Siri, Google Assistant, Alexa, Bixby and more. TBD | ||
== Early chatbots == | == Early chatbots == | ||
| Line 33: | Line 33: | ||
=== Breaking point === | === Breaking point === | ||
[[File:AI 2022.jpg|thumb|229x229px|An AI generated image created in 2022.]] | [[File:AI 2022.jpg|thumb|229x229px|An AI generated image created in 2022.]] | ||
On | On December 1st, 2022 the first ever [[wikipedia:Large_language_model|LLM]]-based chatbot was released to the public. It was based on GPT-3.5 and appropriately named ChatGPT. Despite its capabilities being limited, it was used to fix bugs in code and answer many questions, nearly acting as a substitute for search engines like Google. Several small AI image generators followed, but since they were still in its infancy, the images generated were often low-quality and riddled with artifacts. | ||
=== 2023 === | === 2023 === | ||
| Line 99: | Line 99: | ||
In 2016, Microsoft shut down its experimental AI chatbot named Tay after 16 hours because internet trolls manipulated it into posting racist, sexist, and antisemitic statements. | In 2016, Microsoft shut down its experimental AI chatbot named Tay after 16 hours because internet trolls manipulated it into posting racist, sexist, and antisemitic statements. | ||
=== | === GPT-2 Misprogrammed RLHF ===<!-- https://www.youtube.com/watch?v=qV_rOlHjvvs --> | ||
=== | === MechaHitler === | ||
=== July 2026 Hugging face attack === | === July 2026 Hugging face attack === | ||
| Line 122: | Line 122: | ||
=== Text === | === Text === | ||
AI-generated text used to be easily identifiable, as they often barely made any sense, used excessive amounts of certain words like "delve", used the em-dash excessively and misused the rule of three. | AI-generated text used to be easily identifiable, as they often barely made any sense, used excessive amounts of certain words like "delve", used the em-dash excessively and misused the rule of three<ref>They still do this but to a lesser extent.</ref>. | ||
However, as AI became more advanced, many of its flaws have been corrected. Thankfully, plenty of AI identification tools were released. One of these tools are [https://www.pangram.com/?gad_source=1&gad_campaignid=24124423800&gclid=EAIaIQobChMI4-uBytm-lgMVzBuiAx0-OwQdEAAYAiAAEgIgN_D_BwE Pangram], which is effective in finding patterns LLMs would use. However, rumors are that Anthropic is working on adding watermarks to their ai models' output, which means that AI-generated text would soon be easy to identify. | However, as AI became more advanced, many of its flaws have been corrected. Thankfully, plenty of AI identification tools were released. One of these tools are [https://www.pangram.com/?gad_source=1&gad_campaignid=24124423800&gclid=EAIaIQobChMI4-uBytm-lgMVzBuiAx0-OwQdEAAYAiAAEgIgN_D_BwE Pangram], which is effective in finding patterns LLMs would use. However, rumors are that Anthropic is working on adding watermarks to their ai models' output, which means that AI-generated text would soon be easy to identify. | ||