3 ms·

I agree with your general advice about shipping stuff, starting from stuff you'd want an LLM or agents to do for you. Journaling is good too. I'm not sure if I

I agree with your general advice about shipping stuff, starting from stuff you'd want an LLM or agents to do for you. Journaling is good too. I'm not sure if I agree with you on the tooling end. If you're talking about tooling as in calling OpenAI APIs vs local llamaing/fine tuning your own stuff, I definitely agree with you. If you're talking about going with abstractions like LlamaIndex and Langchain, I strongly disagree. Honestly, I'm almost entirely against all abstractions outside of something like LiteLLM. Langchain, for instance, abstracts the entire learning process away from developers, which honestly translates into little communicable skills after building apps with it. Pretty much all of my learning in this space was from building a product head first and figuring out how to give myself the tools to get me there. You pick up a ton along the way. For instance, you learn a ton from trying to figure out how to build your own text splitter, chunker, and document extractors. Building from scratch is what makes you useful to companies. Understanding the why and how behind the tools is the difference between developing just another chatbot versus developing hugely value-driving solutions for businesses. There's diminishing demand for the former, tremendous demand for the latter.

j45 2y ago
Appreciate your thoughtful reply - the best input that can be given is general anyways to find a way to apply back to our own world :) That's fair about your concern about the tooling. In case your'e interested, happy to share a bit more since I come from a web background too a long time ago, but quietly kept my skills at least passable for hobby learning. Knowing how the tools work is relative to how fast they're evolving, the impossibility of keeping up, let alone finding an application for it (shipping). I spent pretty much 8-10 hours a day last year keeping up, and it felt like 1 month worth of normal tech progress was happening in 1 week, and stepping away quickly caused a gap. One viewpoint I opened up to was by solving a problem, with web app, and increasingly adding more and more AI tooling, first staring with api.. and then going beyond is critical because there's a new programming language. Prompting. Chain of thought, all that stuff. I remember playing around with GPT and having conversations and getting much better results quietly for a very long time until a CoT paper came out and I scratched my head. All to say, trust your abilty to learn and get to it. The lake of all this stuff is becoming a sea and much bigger very quikly, but the tooling is improving. If I tried to learn everything from scratch, it would be harder 2 years ago than now. Now, I agree that tying ones self too much to one framework is risky. Framework in AI are early, and don't have a shelf life of years... maybe 6-18 months. Langchain is even starting to fall out of favour for a few other that are simpler from the results of just building and shipping. You do pick up a ton a long the way. I have known python for a long time and glad I kept it around instead of taking best career practice and floating into management and leadership and getting off the code. The combination of business / problem solving experience in the real world helps a lot to balance out what to do. I'm not sure I really agree with folks looking down on chatgpt "wrappers"... web apps in a lot of ways are database wrappers then. The value is in putting computers to work for people to make their life better. So I guess part of it is deciding what this all means for you, and base your participation on it. Lots of places and ways to grow intrinsically, extrinsically and also make a difference in a way that means something to you.
- laborcontract 2y ago
  Oh yeah, i did not mean to summarily dismiss “wrappers”, a word I have a strong distaste for. The thought there was that I have seen lopsided interest in AI from companies looking to deploy it versus normies looking to use it for any of their daily tasks. > The combination of business / problem solving experience in the real world helps a lot to balance out what to do. The pace at which this space is exhausting and exciting, but the thing that makes it exhilarating is that the newness makes it possible for anyone to have an original thought on how to generate better solutions. The problem solving experience stuff is definitely a big contributor to that. It’s fun figuring out something cool yourself, kind of a secret, and then seeing an inevitable paper or langchain tool 6 months later reflecting similar thinking. I think we’ve progressed along this space in a similar way. The learning path probably is different now from than from two years ago on account of how much catch up there is to do. On top of that, thing can get stressful when you benchmark yourself against well-funded teams. Your last several sentences resonate with me because that meaning keeps the stress at bay. I’ve always loved automation as a means of empowerment in everything I’ve done, but in the prior two decades, it all felt like a toy hobby. LLMs seem close to be able to unlock that potential for everybody, and i find that incredibly energizing.

j45 2y ago

Appreciate your thoughtful reply - the best input that can be given is general anyways to find a way to apply back to our own world :) That's fair about your concern about the tooling. In case your'e interested, happy to share a bit more since I come from a web background too a long time ago, but quietly kept my skills at least passable for hobby learning. Knowing how the tools work is relative to how fast they're evolving, the impossibility of keeping up, let alone finding an application for it (shipping). I spent pretty much 8-10 hours a day last year keeping up, and it felt like 1 month worth of normal tech progress was happening in 1 week, and stepping away quickly caused a gap. One viewpoint I opened up to was by solving a problem, with web app, and increasingly adding more and more AI tooling, first staring with api.. and then going beyond is critical because there's a new programming language. Prompting. Chain of thought, all that stuff. I remember playing around with GPT and having conversations and getting much better results quietly for a very long time until a CoT paper came out and I scratched my head. All to say, trust your abilty to learn and get to it. The lake of all this stuff is becoming a sea and much bigger very quikly, but the tooling is improving. If I tried to learn everything from scratch, it would be harder 2 years ago than now. Now, I agree that tying ones self too much to one framework is risky. Framework in AI are early, and don't have a shelf life of years... maybe 6-18 months. Langchain is even starting to fall out of favour for a few other that are simpler from the results of just building and shipping. You do pick up a ton a long the way. I have known python for a long time and glad I kept it around instead of taking best career practice and floating into management and leadership and getting off the code. The combination of business / problem solving experience in the real world helps a lot to balance out what to do. I'm not sure I really agree with folks looking down on chatgpt "wrappers"... web apps in a lot of ways are database wrappers then. The value is in putting computers to work for people to make their life better. So I guess part of it is deciding what this all means for you, and base your participation on it. Lots of places and ways to grow intrinsically, extrinsically and also make a difference in a way that means something to you.

laborcontract 2y ago
Oh yeah, i did not mean to summarily dismiss “wrappers”, a word I have a strong distaste for. The thought there was that I have seen lopsided interest in AI from companies looking to deploy it versus normies looking to use it for any of their daily tasks. > The combination of business / problem solving experience in the real world helps a lot to balance out what to do. The pace at which this space is exhausting and exciting, but the thing that makes it exhilarating is that the newness makes it possible for anyone to have an original thought on how to generate better solutions. The problem solving experience stuff is definitely a big contributor to that. It’s fun figuring out something cool yourself, kind of a secret, and then seeing an inevitable paper or langchain tool 6 months later reflecting similar thinking. I think we’ve progressed along this space in a similar way. The learning path probably is different now from than from two years ago on account of how much catch up there is to do. On top of that, thing can get stressful when you benchmark yourself against well-funded teams. Your last several sentences resonate with me because that meaning keeps the stress at bay. I’ve always loved automation as a means of empowerment in everything I’ve done, but in the prior two decades, it all felt like a toy hobby. LLMs seem close to be able to unlock that potential for everybody, and i find that incredibly energizing.