1
point
kirillzubovsky
Joined 6,925 karma
KirillZubovsky.com
- kirillzubovsky parentFascinating that we didn't have to wait too long. Apple announced M5 this morning. Does it compare though?
- > So in theory, there’s an expectation that they will diligence the company to make sure they aren’t buying snake oil. In practice, what, are you kidding?
Brilliant. If you can create that kind of leverage, all the power to you!
- Google hasn't touched Google Finance in a thousand years, and then this. Looks amazing. It integrates Gemini to answer questions.
- 3 points
- This is interesting because on one hand, there's nothing wrong with doing this research, and they might as well. On the other, I think we are putting too much emphasis on what has been.
Sure, we had these mammoth, and these fancy birds, and the dinosaurs. So? It's gone. Let's look at the future. let's figure out what we need now (EVs, Rockets, open Internet...etc).
I am just always surprised by how much emphasis we put on bringing things from the dead, instead of actively elevating the living.
- This is the only command I use nowadays: "claude --dangerously-skip-permissions"
I have about 25 agents for different tasks, and sometimes I dispatch them to do their individual work, but that usually happens at the beginning, when I build the first scaffold.
For example, I have agents that setup a DB in certain way, or write a logger in a specific way, I have a test agent that writes them, and a test agent that executes ..etc.
However, for the day-to-day, outside of major ground up builds, straight up Claude.
Sending better prompts to Claude adds more value than everything else, in my experience.
- 3 points
- While I can see how these agentic systems could improve accuracy and scalability (I've dabbled into the space before), I also cannot help but to wonder if LLM providers will soon roll these features into their native offerings. There's no way Grok/OpenAI...etc won't be doing everything in their power to one-up each other, and this is one of the areas. Smart RAG might become a way to augment providers that don't have the power to compete with the biggest players.
- Great job! I bet so many of us (self included) had this nagging need to do this, but you actually followed through. Kudos!
- 1 point
- 2 points
- 1 point
- 4 points
- Yeah. I've been coding with AI a bunch lately, and MCPs sound like a good cool way forward, but having to setup a new iteration for each tool is still a process .... glad to see Zapier take care of it.
Bryan mentioned they've got Zapier Agent too, which can take care of automating choosing the integrations, but I've not played with it yet (https://x.com/bryanhelmig/status/1904220766847066225).
- 9 points
- p.s. you'll see a million opinions on how to improve it, but this is perfect. don't overcomplicate it. Apple Keyboard, split in half, is literally all there need to be and nothing more. It's perfect!
- Yes. Shut up and take my money! This is exactly what I've been looking for and couldn't find. Seriously, can I just give you money and take the prototype right now?
- 1 point
- 1 point
- By the way, if you want to see a proper growth calculator, Trevor Blackwell made one many many years ago. https://growth.tlb.org
- 4 points
- Different LLMs respond to requests slightly differently. So if I say - "here's a log of all the errors I got, can you help me solve it?" - you will get different suggestions than if provided more constraints first. It helps to split the logs and ask to solve one problem at a time, rather than expect it to solve all of the problems at once. So it's easy to just ask for help, but you get better results by helping the LLM with extract directives first. That's one example.
If you ask o1 vs Grok to write something, they have very different styles. Grok is chill and concise, o1 is professory and kind of dull. But, each has its advantage, so it makes sense to give o1 something to make it better, then feed to Grok to make it more fun. Things like that.
- I talked to Cursor, and Grok, and ChatGPT, at lot.
- 8 points
- It may not exist right now, but we can strive towards it. I believe that in a few years there will be an AI, or AIs, that will work completely autonomously. They will not only browser files, documents, and do research, but they will exist on their own, synthesizing information across organizations. That's going to be the 1000x improvements I am thinking about.
I wrote about it: https://www.hackerneue.com/item?id=42701691
- I've been relying on something similar for years, a stripped down custom css library with just the styles I use, and nothing more. It made a lot of sense in the last few years, but honestly, I think it's not going to matter going forward.
"Hey Cursor, make the corners rounded, border thick, and make it glow..." That's css of the future. How it's done, where it's compiled, how big it is... most of that won't matter anymore.
Sure, size will matter to some extend, but what if we could just recompile Tailwind on export and only include classes we used?