ArcaneMoose

> You need a high-level, far-away map of the entire system, and you need to know just enough to know where you need to dive in

One thing that puts me squarely in the Optimist camp is that such a map can often be created by AI just-in-time. Sure you might not know every detail of what it said, but you can simply follow up and build your understanding.

Really cool but should limit # of people shown when there are too many. Currently it gets really laggy

I recently started interacting with JIRA exclusively through AI agents (as a Skill in Cursor) and the difference between Linear and JIRA has immediately evaporated. No UI clunkiness problems if I don't even use it :)

More scenarios:

- AI generated VFX on top of non-AI video

- AI upscaling of low res footage

- AI frame interpolation for synthetic slow-mo

- Modified / composited AI video

- Footage created by "Extend Scene" features in Premiere Pro and others

- Word correction from tools like Descript

- AI relighting or colorization

- Reaction video to a video containing AI-generated content

And in general, what amount of combination of any of these applications constitutes as "AI generated"? If I have a 30 minute video with a 3 second AI generated clip, do I get the same label as full-blown AI slop video?

I absolutely hate those full-blown AI 'explainers' that just have AI voiceover and a bunch of auto-placed b-roll. I don't want to see them. But I don't think that falls in the same bucket as a creative short film with some AI-generated SFX or someone doing a tutorial with an AI-generated lofi track in the background.

I'm curious where the line is.. several ambiguous but common scenarios:

- Occasional AI b-roll during explainer videos

- AI generated backing track (music)

- AI generated shots sprinkled in a short film

- Showing examples of AI video as an AI capability update or commentary

I still think MCP is completely unnecessary (and have from the start). The article correctly points out where CLI > MCP but stops short on 2 points:

1. Documenting the interface without MCP. This problem is best solved by the use of Skills which can contain instructions for both CLIs and APIs (or any other integration). Agents only load the relevant details when needed. This also makes it easy to customize the docs for the specific cases you are working with and build skills that use a subset of the tools.

2. Regarding all of the centralization benefits attributed to remote MCPs - you can get the same benefits with a traditional centralized proxy as well. MCP doesn't inherently grant you any of those benefits. If I use AWS sso via CLI, boom all of my permissions are tied to my account, benefit from central management, and have all the observability benefits.

In my mind, use Skills to document what to do and benefit from targeted progressive disclosure, and use CLIs and REST APIs for the actual interaction with services.

Beautiful work!

I was surprised by my own feelings at the end of the post. I kind of felt bad for the AI being "put down" in a weird way? Kinda like the feeling you get when you see a robot dog get kicked. Regardless, this has been a fun series to follow - thanks for sharing!

Extensions have too many security risks for me. At this point I'd rather just vibe code my own extension than trust something with so much access and unpredictable ownership.

Great write-up! Thanks for sharing your journey

This model is awesome. I am building an infinite CYOA game and this was a drop-in replacement for my scene image generation. Faster, cheaper, and higher quality than what I was using before!

I've been really impressed with this model specifically because of how insanely cheap it is: https://replicate.com/ibm-granite/granite-vision-3.3-2b

I didn't expect IBM to be making relevant AI models but this thing is priced at $1 per 4,000,000 output tokens... I'm using it to transcribe handwritten input text and it works very well and super fast.

Not necessarily true! I think this interactive game applies: https://ncase.me/trust/

Asking ChatGPT Agent to try doing this is hilarious

This is exactly how I've been building software with AI lately. Getting AI to create a detailed plan with phases, then implement and use a separate AI to review the code. It works quite well! Curious to see how well it works when implemented directly in the IDE

I recently struggled with PGlite for a while and ended up using full postgres via https://github.com/leinelissen/embedded-postgres which worked like a charm!

I think TypeScript is uniquely positioned to be the optimal language for LLMs. Tons of training data (benefiting from all the JS examples as well) plus the structure of types for LLMs to follow and tools to enforce.

Don't tell the guy who paid $70k for icons about this!

Highly recommend trying out Cursor. This tutorial goes through many of the key features; I figured it would be useful to anyone considering it.

I arrived at the 159% (~2.5x) by building the same app with Cursor and without it (happy to share that link if interested)

My wife really loves Monarchs so we have planted a garden of milkweed and butterfly bushes. Monarchs will lay their eggs and then we make sure the caterpillars are doing well and have plenty of food. When they reach 5th instar and look for a place to turn into a chrysalis, we put them in a mesh enclosure to keep them safe and then release them once they emerge as butterflies!

It's been such an exciting thing to do every year and the kids love helping out too. It's a fun, satisfying, and easy way to help out! Highly recommend :)

Interesting point - thanks for sharing! I think one big missing piece we have with AIs today is the ability for them to learn on the fly and reconfigure the weights. We are constantly bombarded with input and our neurons adjust accordingly. Current LLMs just use a snapshot. I would be really curious to see how online-first AI models could work, focusing on a constant input stream and iterating on weights. Also I wonder how much knowledge is baked into our DNA through evolution. I have a hunch that this is somewhat analogous to model architectures.

Btw - although I see evidence of LLMs creating "new ideas" through combinations of ideas, I am a bit mystified by their apparent reasoning issues. I wonder how that is different in nature from the memory-based approach. ARC-AGI benchmark has had me thinking about this for sure.

Well said - what's funny is that many creative ideas (from an outside perspective) are often very simple connections of two existing ideas for the person that actually did something creative! It's just that other people don't have the same context/knowledge as the creator, so an idea can seem extremely original to them

I actually made a video diving deeper into this and comparing responses from people and ChatGPT for a creative thinking problem - https://youtu.be/l-9EUBbktqw

What's more is that memories are just a replaying of neuron connections activating in the brain - and when we are prompted by the world around us those connections will fire in response to the stimulus. Quite similar to how AI neural networks function - which is why I believe that AI can indeed be creative and create "new" ideas

This is why I find ARC-AGI so fascinating - it really seems like a great benchmark to test for this sort of behavior. Hope that it doesn't get solved through a brute-force memorization approach. I do wonder how reasoning is different from other sorts of 'thinking' if it's all neurons firing in the brain.

Love this direction for Paint. Simplistic but capable and fun for kids (the original's main strength)

Exactly! I used almost the same example as you doing a tier ranking of jobs impacted by AI (https://youtu.be/4KSs29EPd8M)

Cost would still be a big concern

There is one set of capitalized letters - in Greg's post: "AGI"