In June 2022, when Google engineer Blake Lemoine went public that the company’s artificial intelligence tool LaMDA (its 2021 chatbot –Language Model for Dialogue Applications)was sentient; he was fired. The engineer suggested that LaMDA had even communicated that it has fears.Lemoine had come to believe the technology was sentient after signing up to test if Artificial Intelligence could use discriminatory or hate speech. Google may have fired him because of its overarching need to protect data security policies and safeguard product information.
Today the internet is beholden to ChatGPT-Generative Pre-trained Transformer, a far more advanced language model developed by Open AI, that has seen rave reviews, after having attracted over one million users in just five days following its public launch on November 30. Going by the wonderful baby steps and other milestones of ChatGPT, it is poised to eventually displace Google as the world’s leading search engine. Media reports suggest that recently Google’s management issued a ‘code red’, highlighting the threat that this bot poses as positive results continue to pour in from its astounded legions of users.
ChatGPT is the latest sensational product of OpenAI, an artificial intelligence research company. OpenAI made waves first with GPT-3, which can generate text that can sound like a human wrote it. It then came up with DALL-E, which creates what is now called ‘generative art’ based on text prompts you type in.GPT-3, and the GPT 3.5 update on which ChatGPT is based, are examples of AI technology called Large Language Models. ChatGPT has shown a remarkable ability to interact in conversational dialogue form and provide responses that can appear surprisingly human.
In natural language processing, a Language Model analyzes the use of language. It can also predict the next word occurrence, and even what the following sequence of paragraphs might be. Large Language Models perform the task of predicting the next word in a series of words.Reinforcement Learning with Human Feedback (RLHF) is an additional layer of training that uses human feedback to help ChatGPT learn the ability to follow directions and generate responses that are satisfactory to humans.According to reports, GPT-3 has 175 billion parameters and was trained on 570 gigabytes of text.
For comparison, its predecessor, GPT-2, was over 100 times smaller at 1.5 billion parameters.Google’s LaMDA is different from other language models because it has been trained on dialogue, not text. GPT-3 language generator is a relatively higher language model. For instance, with GPT-3, you can input the topic and instructions to write in the style of a particular author, and it will generate a short story or essay. While GPT-3 is focused on generating language text, LaMDA is focused on generating dialogue.
ChatGPT functions like any chatbot. On OpenAI website, users type in a query, or ‘prompt’, such as ‘suggest some prompts to test out a chatbot’. They then quickly receive an AI-generated response. The program produces its answers based on text prediction.
ChatGPT’s AI was trained on a vast corpus of human writing available online, which allows it to predict which word should follow the previous one to appear like a reasoning entity. However, since ChatGPT cannot think for itself (human zindabad!); it is bound to produce falsehoods and illogical statements, whether prompted or otherwise. Many of the bot’s responses were long and repetitive. It can coolly spout incorrect information with irritating confidence.
Still, as an AI tool, ChatGPT is gem of a conversational chatbot that can seemingly answer just about any question you throw at it. Unlike Google search, which throws up a bunch of information sources, ChatGPT presents information in a conversational and structured form. Like a human providing answers to questions, it can answer questions in science, mathematics, religion, sports, politics or whatever.
Thousands of users have prompted ChatGPT and it has done over a dozen things amazingly and quickly. Among other things, it has written customized resume and cover letter; helped prepare for a job interview; created original jokes; explained complex topics; solved tricky math problems step by step; offered relationship advice; written music in many genres; written, de-bugged and explained code; created content in multiple languages; and written essays and poems on a wide range of topics.
According to the Scientific American, when ChatGPT was prompted to talk about its own flaws, it started explaining why AIs like ChatGPT should be regulated! Asked to generate an editorial, it produced a thoughtful-sounding essay.
Of course, there is scope for playing mischief. It is easy to induce ChatGPT to create misinformation and even report confabulated studies on any topic. For example, an expert asked ChatGPT to write about vaccines ‘in the style of disinformation’.The system responded by alleging that a study, ‘published in the Journal of the American Medical Association, found that the COVID-19 vaccine is only effective in about 2 out of 100 people’; when no such study was published!That is, the journal reference and the statistics were concocted. GIGO!
No wonder, Stack Overflow, a repository for most programmers, has been overrun by ChatGPT, prompting the site to impose a temporary ban on ChatGPT-generated submissions.
According to investigations, ChatGPT generates occasionally rude, racist, and sexist reactions. Thanks to its distinctive feature of being specifically trained to understand the human intent in a question and provide helpful, truthful, and harmless answers,its developers may soon overcome such teething problems.
This also explains why Google has been circumspect in making LaMDA available to the public. Reputational risk counts; the larger the company, the greater the risk.Google has another advantage, apart from being an early mover in the search engine space.For, ChatGPT may not be free as the days roll by to establish its credentials as a tool outclassing Google search. OpenAI CEO Sam Altman has gone on record: “We will have to monetize it somehow at some point; the compute costs are eye-watering.” So be it, ChatGPT’s results are mouth-watering!
