Adventures with ChatGPT and generative LLMs
i hope the following may be instructive. it moves my knowledge of how to use ChatGPT forwards, so i thought i would share it.
i wanted to get a thumbnail sketch of the US gas research institute, but it was shut down in 1997, which is before the massive use of the web. there were no usable hits from the top ten results from google (though GRI was mentioned and would have been useful if one were doing in-depth research on it). as it is rarely worth searching past the top ten, i switched to ChatGPT:
The institute’s research covered a wide range of topics, including:
-
- Exploration and production
- Transmission and distribution
- Storage
- End-use technologies
- Environmental impacts
- Safety
- Economics and markets
GRI was also involved in various collaborations and partnerships with government agencies, universities, and private companies to support research and development efforts.
GRI was dissolved in 1997 by the Natural Gas Wellhead Decontrol Act of 1989, which eliminated the price controls on natural gas and removed the funding for GRI.
this appears to be incorrect. GRI merged with IGT and formed the Gas Technology Institute (https://www.electricnet.com/doc/gri-merger-with-igt-creates-gas-technology-in-0001 – undated, unfortunately), which i believe still exists. this came from my web search, not the AI chat bot. to me this reinforces the necessity of checking with other sources, probably the web as first port of call. i pushed the AI bot further:
However, the Institute of Gas Technology (IGT) is a research organization that was established in 1940, and it is still active today. It is a private, non-profit organization that conducts research and development on technologies and techniques related to the production, transportation, and use of natural gas and other hydrocarbons. IGT conducts research on a wide range of topics, including:
-
- Exploration and production
- Transmission and distribution
- Storage
- End-use technologies
- Environmental impacts
- Safety
- Economics and markets
IGT works with industry, government, and academia to develop and commercialize new technologies and to provide technical support and training. It should not be confused with GRI, as they are different entities, even though they share a similar focus in natural gas research.
and now, it gives me a much more accurate answer. i had already heard that ChatGPT can get stuck in a rut and it definitely has a tendency to confirm, sometimes uncritically, the input prompt. i realise this getting stuck in an epistemological rut has happened to me a few times now. i am pretty sure that OpenAI will fix this soon.
all told, these new AI chatbots (in this case ChatGPT and Chatsonic) are doing a remarkable job. human vigilance is still very much required and i wouldn’t let them loose unaided on any decision that had real-world consequences. for doing research though, when everything must be checked and cited anyway, it speeds things along and often offers new ideas and certainly helps with clarifications and unearthing what can seem like otherwise byzantine explanations.
most of the above would be very difficult or even impossible via a web search unless you got really lucky and someone had asked and answered the exact questions, say in quora or reddit. quora answers tend to be a bit unreliable.
what is so profound about generative LLMs like ChatGPT and Chatsonic is the way they can explain things which are not clear or are superficially contradictory. i cannot be the only one that gets immense satisfaction and pleasure from understanding the world better. if i am right, this will be the single most amazing pedagogical and research tool ever invented. a small claim, i know.