For more than a week, social media has been flooded with screenshots of a chatbot. ChatGPT, a conversational interface for OpenAI’s GPT-3.5 AI system, amazes experts with smart answers and is ridiculed by ethicists for its vulnerabilities. Even Sascha Lobo, often not the fastest of all checkers, felt compelled to proclaim the “end of irrelevant artificial intelligence” on Spiegel Online. Oh well.
In the text I piqd, Ian Bogost correctly describes the AI marvel chatGPT as a toy for the infinities of text, a synthesizer for knowledge, regardless of whether it is true or factually false knowledge.
ChatGPT isn’t a step along the path to an artificial general intelligence that understands all human knowledge and texts; it’s merely an instrument for playing with all that knowledge and all those texts. Play just involves working with raw materials in order to see what they can do. You play a game, or an instrument, to avail yourself of familiar materials in an unexpected way. LLMs are surely not going to replace college or magazines or middle managers. But they do offer those and other domains a new instrument—that’s really the right word for it—with which to play with an unfathomable quantity of textual material. (…)
we should adopt a less ambitious but more likely goal for ChatGPT and its successors: They offer an interface into the textual infinity of digitized life, an otherwise impenetrable space that few humans can use effectively in the present.
An instrument for textual infinities that can generate terms of services as a Shakespearean sonnet, but which, due to its unreliability, does not mean the end of Google or even institutional education, as is often written about in the frenzy of hype. (Side note: I have long been in favor of keeping digital devices out of educational institutions, if only because handwritten notes aid learning. The development of knowledge synthesizers that can produce in 2 seconds a conventional essay on the French Revolution that cannot be written by a human written essay supports my position. Let them write by hand. Problem solved.)
ChatGPT, for all its miracles, is still a stochastic parrot, albeit the most evolved at the moment.
Yesterday I have in my newsletter I still wrote about the text I piqd here that it is one of the smarter pieces of opinion about ChatGPT and it is already obsolete today – at least on the outer edge of AI development: In the paper Discovering Latent Knowledge in Language Models Without Supervision, scientists have developed a kind of lie detector for large language models, and can examine the truth content in a trained AI model, completely independently of its generated outputs. Here one Twitter thread by AI scientist Collin Burns about how this “lie detector” works. (Fun fact: Collin Burns is a former world record holder for solving the Rubik’s Cube. I wouldn’t be surprised if a Rubik’s Cube wizard cracked the alignment problem.)
The successor to GPT-3.5 will be released in a few weeks. GPT-4, like its predecessors, will be vulnerable to prompt hacking and nonsense outputs. But it is foreseeable that AI systems will very soon achieve all performance elements of the LLM benchmarks for human intelligence.
The clearer breaking latest news (Artificial General Intelligence) is on the horizon, the more pressing the millennia-old fundamental questions are being asked in completely new contexts: What is intelligence? What is the difference between human intelligence and its simulation? What does this difference mean? How can we deal with the blurring line between simulation and reality? And what is reality; and what is consciousness; and what are we anyway?