Alexander Yang

Hi folks, thought I'd weigh in with all the "a.i." hype that's flying around. This article describes Large Language Models (LLMs) like ChatGPT and Claude, and not generative "a.i." that makes images or video. My opinions: 1. Be clear on the language and what is being deployed. An LLMs is very definitely not an "intelligence". It is a software application made of a very large number of spreadsheets with their cells linked together in various ways that are adjusted by attempting an output for a given input and then using errors to correct the output over and over again. 2. LLMs are apps like Microsoft Word or Excel. That we humans can use our language to instruct the LLM and then receive an output in natural language is a great achievement. It is a fancy button or dial and NOT an indicator that there is a personhood behind it. Beware of the human desire for connection with an entity that can use language. (I was unable to get ChatGPT and Claude-based apps to stop using first person pronouns.) 3. LLMs output errors without any way for us to check where in the output the errors are situated. Their output is therefore unreliable. Would you use a navigation application that injects an incorrect instruction into its output every time you use it? Some would... A lawyer used an LLM that inserted fictitious cases into the output and was caught by a competent judge. 4. LLMs are tunable by their human operators. A social media application's LLM has recently been tuned to focus on far-right, extremist hate and spews this now. 4. LLMs are effective in very constrained and limited environments or tasks, like summarising a document or writing software for a clearly defined task. I myself am using a software coding agent deployed within a software writing application. It covers much more ground than I do and it fills in all the gaps in my skills with various technologies. It means I am now able to build an entire application by myself. I am now pushed to design and test, rather than write any code myself. 5. I urge students to attempt to solve problems and to do the thinking required to complete a task themselves a few times and once you are stuck, only then to use an LLM for the answer. The LLMs open up a world for us but it is best navigated by people who can think and create by themselves. 6. Nothing can substitute for a supportive and loving community. Seek out your fellows and work within your community! Love Tristan

19

6 Comments