Buy me a hot chocolate! | About Me
TL;DR - I was once a skeptic of LLMs (Large Language Models) like ChatGPT but now I use them for what they are and nothing more and as a result I am far more productive than I was before.
Consider using them, playing around with them to see how they fit in your workflow, pay for them if needed or not. They’re not paying me to tell you this so if you do or do not I won’t be sad. :-)
There’s not much to say here, that’s pretty much it (Feel free to read on for more context though). Search is Google’s to lose. Just today I was reading Snowflake’s documentation on their access control setup and how it works.
And maybe it’s because it can synthesize the right string of words to “answer” a query I find it better at answering questions than finding the same answers from Google.
Probably. But I also don’t have time to waste scouring links, reading blogs and articles, combing through documentation. What I do have is – after more than a decade in tech – an idea of how something should work, what to look for, what to verify, what direction to go, how this backend service should be constructed, what “good” looks like and what LLMs help me do is take what I envision and get a skeleton up and running far faster and quicker.
Are things that are generated by an LLM perfect? Are developers going to be replaced by farms of bots churning out code?
No.
Are they tools that will help you be more productive?
Very likely yes.
On a handful of occasions now I’ve become so flustered with wasting time wading through all the old StackOverflow answers, blog posts for a version of a package far too old, or one filled with so many annoying ads that getting to the knowledge – the thing that Google was supposed to help organize and make available to us … – that I thought: “Why don’t I just ask ChatGPT?” and the majority of the time it resulted in:
The caveat here is LLMs will make competent, capable programmers who have a grasp over their domain enough to know what “good” and correct should be – LLMs will make these developers far more productive.
Where there be dragons is for the new engineers that are just learning or still copy-pasta’ing code from StackOverflow but don’t understand what it is they copied … these are they that will suffer.
What LLMs do is reduce the tedium. It’s really kinda great.
I’ve a buddy of mine. He’s a very capable and strong Java developer at Amazon. If you’ve used Amazon.com you’ve benefited from his intelligence. When we were younger we were talking. He was learning Java at the time at university and had a question. (I knew nothing of Java having switched away from Computer Science as a major after getting a C in C++. That’s a story for another day) I said: “I know! Post your question to StackOverflow!” He responded, with a dower look on his face: “No, I’ll figure it out.” After more prodding I understood why he was so reluctant: He didn’t want to get nerd-shamed that his question was too dumb, or obvious, or that it was a dupe and to have it closed in anger. What ChatGPT is, StackOverflow could have been. But – ironically enough – it was the people that made StackOverflow toxic sometimes and why it’s not used as much now.
In sum, I find that I, the end user am the beneficiary of 10s of billions of dollars in investment causing 100s of billions in valuations chasing ~10 billion in total revenue for the whole of the LLM space … so I don’t know when or if the AI/ML/LLM hype will end. I don’t know if the – what I think is very generous – free tiers of these LLMs and their seemingly endless context windows will end but for the time being I am now a paying subscriber and have embraced it and it’s only helped me and my career.
Copyright Alex Narayan - 2024