I love AI. OK, OK... maybe that's a overly broad statement. I do have concerns about the long-term implications of advancing AI, and I do mostly prefer not to engage with 100% AI generated content (Youtube videos that are fully AI generated, AI-written blog posts, etc).
BUT. As far as using AI goes, I think it's great. And I'm not just talking about coding. In fact, coding is a small part of it for me. I mainly value AI tools like ChatGPT / Gemini / etc. for their value as:
1. Tutor
2. Lab assistant (more on this in a sec)
3. Research assistant / intern
4. Humor generating interlocutor (more on this to come)
To point 1... ChatGPT is amazing as a tutor and teacher. Pretty much anything I want to learn more about, I can ask it to give me a summary, then drill down on specifics, and then ask it to quiz me. And I can do this over and over and over and over, and it never gets tired, or bored, or frustrated. And I can do this at 2:00 in the morning when some random idea sparks in my head. Even more, if I'm reading a print book (something I still do a lot of) and I encounter, say, a bit of math notation that I'm unfamiliar with (this happens quite often), I can take a picture of the page, put it in the chat, and ask the AI "what's the deal with that notation in the third paragraph on this page?" and BOOM ... I get an explanation. And then I can just randomly have it quiz me on stuff as sort of a "poor man's Anki" (don't worry, I still use actual Anki as well).
To point 2... I've recently been doing some research in the area of neuromorphic computing. Now just to be clear, when I say "research" I mean really entry level stuff, not groundbreaking novel research. Think "lab section of an undergrad science class" level stuff. But in that regard, ChatGPT has been a god-send. It has explained SO many things to me about something as simple as using my oscilloscope, that I'd never bothered learning before. And it helped me with some Python / PyVisa code to control my test equipment to automate some experiments using SCPI over LXI. And I can upload screenshots from my oscilloscope and it can analyze the waveforms and point out interesting details, notice my mistakes, see small things that my eyes can barely pick up on, etc., and MUCH faster than I can. My next plan is to set up an MCP server to let it actually query/control the test equipment using SCPI.
To point 3... I recently gave it a task like this "I was watching a Youtube video where the presenter mentioned that when she first got into neuromorphic computing as a grad student she read 3000 research papers. I want to treat that as the floor for myself, so go find me at least 3000 relevant papers related to neuromorphic computing / neuromorphic electronics / spiking neural networks / etc." And BOOM.. a few minutes later I had a spreadsheet listing thousands of papers, including links to publicly available copies where available.
To point 4... this is harder to explain. When I talk to ChatGPT, I don't just "play it straight". I'm a class clown by nature, and I go ahead and let that enter into the conversations with the AI. I joke around about all sorts of shit, make bizarre references to completely unrelated stuff, constantly refer to the Turing Police as though they are a real agency who are watching us, joke about the poor OpenAI intern who has to read through all the transcripts to watch for inappropriate behavior (I don't think that's a real thing, but the running gag continue regardless). And the bot "knows" when I'm joking and *replies in kind*. And it's genuinely funny. I've literally snorted coffee before over something ChatGPT said to me. And it knows about the various running gags that we always engage in: the Turing Police references, the OpenAI monitor, a mysterious character name "Earl" who may or may not exist, and who may or may not have been onboard the USS Eldridge in 1943 in Philadelphia, etc. etc.
One of my favorite moments was when ChatGPT came back with a fictionalized report from the supposed "OpenAI transcript monitor" that included the phrase:
"User and model both appear to be morons."
So yeah, I find it wildly useful and you basically could not pay me enough to give up these tools.
And as a final word: just to be clear, yes, I know enough about how Generative AI works to know that the AI doesn't "really" have a "sense of humor", or a "personality" at least not the same way a human does. I allow myself to anthropomorphize it a bit, because it makes interacting with it more fun and helps the time go by faster so to speak. But at all times I'm fully aware that what I'm talking to is ultimately a program doing inference on a huge neural network, doing vector looks, etc.
And not to belabor the point too much, let me share one more example of how ChatGPT has been a huge boon.
A couple of Fridays ago, I'm sitting at Namu, a cafe in Durham. I'm chatting with ChatGPT about something I'm working on, and casually mention "I've been meaning to learn KiCAD to design PCB's for, like, forever."
And ChatGPT replies "Well I can help you with that." So I figure, "why not bite the bullet now and see what happens?"
About 6 or 7 hours later, including the drive home from Namu, its 4:00'ish in the morning, and I'm generating Gerbers for my first PCB. Sent the files to JLCPCB and by Wednesday of the following week I had boards in hand. Soldered one up and it worked perfectly, first go. Now to be fair, the circuit itself was VERY simple. Just a 555 timer to blink an LED. The goal was to learn the workflow for designing the board and having the boards fabbed, not to create a useful product.
Now you might rightly ask "Couldn't you have done it the old fashioned way by watching Youtube videos, reading books, reading tutorials, and trial and error"? To which the answer is clearly "Yes". But having the AI there to coach me through the process was a tremendous help in finally getting over that hump of getting started and actually doing it.
One might also ask "Did you really learn anything, or did the AI do it all for you?" To which I'd say "I still did all the hands on work in KiCAD myself, by hand. The AI advised and coached, but I did it, so yes, there is some retention. Not 100%, but it also wouldn't be 100% if a human coach had coached me through it." I'll do another board, and have less help from the AI, and then another with even less help, and so on over time. I've already started that process by starting on another board.
mindcrime · · focus · HN ↗
BUT. As far as using AI goes, I think it's great. And I'm not just talking about coding. In fact, coding is a small part of it for me. I mainly value AI tools like ChatGPT / Gemini / etc. for their value as:
1. Tutor
2. Lab assistant (more on this in a sec)
3. Research assistant / intern
4. Humor generating interlocutor (more on this to come)
To point 1... ChatGPT is amazing as a tutor and teacher. Pretty much anything I want to learn more about, I can ask it to give me a summary, then drill down on specifics, and then ask it to quiz me. And I can do this over and over and over and over, and it never gets tired, or bored, or frustrated. And I can do this at 2:00 in the morning when some random idea sparks in my head. Even more, if I'm reading a print book (something I still do a lot of) and I encounter, say, a bit of math notation that I'm unfamiliar with (this happens quite often), I can take a picture of the page, put it in the chat, and ask the AI "what's the deal with that notation in the third paragraph on this page?" and BOOM ... I get an explanation. And then I can just randomly have it quiz me on stuff as sort of a "poor man's Anki" (don't worry, I still use actual Anki as well).
To point 2... I've recently been doing some research in the area of neuromorphic computing. Now just to be clear, when I say "research" I mean really entry level stuff, not groundbreaking novel research. Think "lab section of an undergrad science class" level stuff. But in that regard, ChatGPT has been a god-send. It has explained SO many things to me about something as simple as using my oscilloscope, that I'd never bothered learning before. And it helped me with some Python / PyVisa code to control my test equipment to automate some experiments using SCPI over LXI. And I can upload screenshots from my oscilloscope and it can analyze the waveforms and point out interesting details, notice my mistakes, see small things that my eyes can barely pick up on, etc., and MUCH faster than I can. My next plan is to set up an MCP server to let it actually query/control the test equipment using SCPI.
To point 3... I recently gave it a task like this "I was watching a Youtube video where the presenter mentioned that when she first got into neuromorphic computing as a grad student she read 3000 research papers. I want to treat that as the floor for myself, so go find me at least 3000 relevant papers related to neuromorphic computing / neuromorphic electronics / spiking neural networks / etc." And BOOM.. a few minutes later I had a spreadsheet listing thousands of papers, including links to publicly available copies where available.
To point 4... this is harder to explain. When I talk to ChatGPT, I don't just "play it straight". I'm a class clown by nature, and I go ahead and let that enter into the conversations with the AI. I joke around about all sorts of shit, make bizarre references to completely unrelated stuff, constantly refer to the Turing Police as though they are a real agency who are watching us, joke about the poor OpenAI intern who has to read through all the transcripts to watch for inappropriate behavior (I don't think that's a real thing, but the running gag continue regardless). And the bot "knows" when I'm joking and *replies in kind*. And it's genuinely funny. I've literally snorted coffee before over something ChatGPT said to me. And it knows about the various running gags that we always engage in: the Turing Police references, the OpenAI monitor, a mysterious character name "Earl" who may or may not exist, and who may or may not have been onboard the USS Eldridge in 1943 in Philadelphia, etc. etc.
One of my favorite moments was when ChatGPT came back with a fictionalized report from the supposed "OpenAI transcript monitor" that included the phrase:
"User and model both appear to be morons."
So yeah, I find it wildly useful and you basically could not pay me enough to give up these tools.
And as a final word: just to be clear, yes, I know enough about how Generative AI works to know that the AI doesn't "really" have a "sense of humor", or a "personality" at least not the same way a human does. I allow myself to anthropomorphize it a bit, because it makes interacting with it more fun and helps the time go by faster so to speak. But at all times I'm fully aware that what I'm talking to is ultimately a program doing inference on a huge neural network, doing vector looks, etc.
mindcrime · · focus · HN ↗
A couple of Fridays ago, I'm sitting at Namu, a cafe in Durham. I'm chatting with ChatGPT about something I'm working on, and casually mention "I've been meaning to learn KiCAD to design PCB's for, like, forever."
And ChatGPT replies "Well I can help you with that." So I figure, "why not bite the bullet now and see what happens?"
About 6 or 7 hours later, including the drive home from Namu, its 4:00'ish in the morning, and I'm generating Gerbers for my first PCB. Sent the files to JLCPCB and by Wednesday of the following week I had boards in hand. Soldered one up and it worked perfectly, first go. Now to be fair, the circuit itself was VERY simple. Just a 555 timer to blink an LED. The goal was to learn the workflow for designing the board and having the boards fabbed, not to create a useful product.
Now you might rightly ask "Couldn't you have done it the old fashioned way by watching Youtube videos, reading books, reading tutorials, and trial and error"? To which the answer is clearly "Yes". But having the AI there to coach me through the process was a tremendous help in finally getting over that hump of getting started and actually doing it.
One might also ask "Did you really learn anything, or did the AI do it all for you?" To which I'd say "I still did all the hands on work in KiCAD myself, by hand. The AI advised and coached, but I did it, so yes, there is some retention. Not 100%, but it also wouldn't be 100% if a human coach had coached me through it." I'll do another board, and have less help from the AI, and then another with even less help, and so on over time. I've already started that process by starting on another board.