Adam Green

Christian // Software Engineer // AI Enthusiast

[Adam Green @admgrn01]

It’s a little interesting that what was considered top of the line / “frontier” 6 months ago is not the go to for a lot of people’s tasks now.

Opus 4.8 is still one of my favorites and it was considered exceptional when it was the top of the charts. Since then opus 5 and fable have been released and people default to using these.

[Adam Green @admgrn01]

@omkarships Yup, prioritizing near term profits via minimizing junior headcount and lessening training needed to push off the issue for a future date or altogether assuming/hoping AI is good enough to replace those senior devs when the time comes for them to retire/be let go.

[Adam Green @admgrn01]

@Ionkosm This is why I’m more convinced than ever that maximum adaptability in the era of ai products should be focused on organizing and constructing your files and preferences around concepts and generalized ideas and not vendors and products.

[Adam Green @admgrn01]

I know it’s easy to get excited with the possibilities of AI and its easy to get distracted by the good idea fairy, but two recommendations:

1. Always start simple and iterate incrementally.

2. Stay disciplined with your original goal / requirement(s).

[Adam Green @admgrn01]

@dpascoaa For sure…

Many people were never learned how to properly plan and do things like project management before ai, and now the results will compound much faster and easier, whether they are good or bad.

[Adam Green @admgrn01]

Just curious, why don’t more people talk about the value of upfront planning when building with AI?

We hear a lot about the value of input quality (context).

We hear a lot about the projects people build.

We hear about token usage (tokenmaxxing).

But…I don’t see or hear much about knowing the process to determining what you want, why, and what the best way to achieve that is.

[Adam Green @admgrn01]

@tunguz This gives me 10 things I hate about you vibe - “I know you can be underwhelmed and you can be over whelemd, but can you ever just be whelmed?”

[Adam Green @admgrn01]

Yeah, it’s pretty crazy and I’m surprised Apple is ok with it. The part that is most frustrating to me is that Klarna told me directly that they cabt share the approval limit with me, it’s on a individual purchase basis, and there is no way to know if you’ll get approved ahead id time. Seems like a huge waste of time for everyone, including klarna to process apps.

[Adam Green @admgrn01]

@omarsar0 you frequently share a lot of solid papers. I got to wondering, do you go through and read all of these from start to finish or do you have an AI flow with them? I am trying to be more disciplined about staying on top of some of the papers coming out but find myself quickly distracted with life.

[Adam Green @admgrn01]

True intelligence is the ability to share knowledge in a manner that can be understood by others.

[Adam Green @admgrn01]

Folks that are using Hermes - thoughts on running via @nvidia nemoclaw/openshell vs running @NousResearch docker setup?

[Adam Green @admgrn01]

Just grabbed a DGX Spark and setting up for the first time. Previously, I've been doing all local AI exploration via M4 Max Macbook (which has been great).

After doing all the initial config updates, I have been prioritizing setting up containerized projects.

So far:

- vLLM container to serve my local models (have two docker compose profiles currently: Qwen3.8-27b & Ornith-1.5-35b MoE.

Next up - setting up coding harness container with the following profiles:

- Codex
- Pi
- Grok

MTF

[Adam Green @admgrn01]

Really been exploring the LLM-WIKI concept that @karpathy wrote about 5 months ago (https://t.co/2jp7M5kOnE).

I feel like this is probably highly underutilized by most people. I will be sharing some ways that I’ve been using this in the coming days.

[Adam Green @admgrn01]

@olegakbarov For sure, we are all building sweet dashboards for our data that we admire once and never talk about it again.

[Adam Green @admgrn01]

Beginning to question if main stream chatbot preference RL is prioritizing value and truth or mainly just reinforcement ego learning.

[Adam Green @admgrn01]

@merybenavente “Clear and concise” is my go to instruction annnnnnnd 50/50 I’ll either get a nice 1-3 liner or a I’ll get 5 paragraphs acknowledging it’s going to give me a clear and concise response, “no fluff”, “no filler”, “straight to the point”, “right to it” response.

[Adam Green @admgrn01]

@merybenavente Ooof 😂

[Adam Green @admgrn01]

Things I don’t see enough of when people debate the cost v reward of investing in hardware for local AI compared to cloud provider costs:

- the peace of mind not having to worry about accidentally including sensitive info in your prompts. Not to mention the more relevant context (which can include sensitive info), the better the results.

- the learning journey about workflow design, introspection and critically thinking about how and why you do things (with ai and without)