Vibe Slopping at 1000 tokens per second

Yeah honestly having max cognitive capability is #1 for me. Faster tokens is a distant second. I think anyone working on creating valuable unique IP feels this way.

This us where agents actually shine. Having a smart model write code and plan is great and then having cerebra’s do ask the command line work, write documents effectively instantly and other simple tasks does sped things up quite a bit.

Here's the system Prompt Template they use : https://github.com/microsoft/vscode-copilot-chat/blob/4c72d6...

"- cursor position marked as ${CURSOR_TAG}: Indicates where the developer's cursor is currently located, which can be crucial for understanding what part of the code they are focusing on."

I was not aware that was a thing and useful to know. Thanks!

Interesting to hear how others use these tools. I often phrase things as “this line/method” which implies the tool knows where my cursor is.

I use the in-line prompt when I’m talking about a specific area. In the chat I always explained in words what part of the code I’m talking about. This tidbit of information will change how I use chat.

Isn't that needed for the tab completion?

I assumed they had this already, but I began to suspect it didn't actually exist. Disappointed to learn I was right bc half the time copilot pretends it can't read my code and at all asks me to go look for stuff in the code.

You know it's going to be a great article when the design is from 1995

XHTML 2.0,WML,SOAP, APPN,WAP...for each new technology there's thousands of failed protocol.

Not at all. GPT-4o is image output - this model (and previous Qwen release QvQ - https://simonwillison.net/2024/Dec/24/qvq/) are image input only with a "reasoning" chain of thought to help analyze the images.

Model is insane but the RPM limit is insane too.

I have ADHD and while you're right that everyone experiences some of these things occasionally, for those with ADHD our emotion regulation system is fundamentally different.

As a kid, you build your identity and coping mechanisms through emotional experiences, but when your emotion engine is 'broken' or works differently, you develop differently.

The intensity, frequency, and impact of these experiences for someone with ADHD is far beyond what neurotypical people experience.

It's not about occasional forgetfulness or distraction - it's about a brain that's structurally and functionally different, affecting every aspect of daily functioning.

Getting diagnosed isn't about finding an excuse, it's about finally understanding why basic things others find easy have always been so much harder for you.

So today is Qwen. Tomorrow a new SOTA model from Google apparently, R2 next week.

We haven't hit the wall yet.

Qwen 3 is coming imminently as well https://github.com/huggingface/transformers/pull/36878 and it feels like Llama 4 should be coming in the next month or so.

That said none of the recent string of releases has done much yet to "smash a wall", they've just met the larger proprietary models where they already were. I'm hoping R2 or the like really changes that by showing ChatGPT 3->3.5 or 3.5->4 level generational jumps are still possible beyond the current state of the art, not just beyond current models of a given size.

> met the larger proprietary models where they already were

This is smashing the wall.

Also if you just care about breaking absolute numbers, OpenAI released 4.5 a month back which is SOTA in base model, planning to release O3 full in maybe a month, and Deepseek released new V3 which is again SOTA in many aspects.

> We haven't hit the wall yet.

The models are iterative improvements, but I haven't seen night and day differences since GPT3 and 3.5

Yeah. Scaling up pretraining and huge models appears to be done. But I think we're still advancing the frontier in the other direction -- i.e., how much capability and knowledge can we cram into smaller and smaller models?

Because 3.5 has a new capability which is following instructions. Right now we are in 3.5 range in conversation AI and native image generation, both of which feels magical.

Tbh such a big jump from current capability would be ASI already

Google's announcements are mostly vaporware anyway. Btw, where is Gemini Ultra 1? how about Gemini Ultra 2?

It is already on the LLM arena right, codename nebula? But you are right they can fuck up their releases royally.

I guess they don’t do ultras anymore, but where was the announcement for it? What other announcement was vaporware?

Asymptotic improvement will never hit the wall

We've slid into the upper S curve though.

Any more info on the new Google model?

I have ADHD and I've decided that it's part of my identity.

I was diagnosed late at 28. When I got diagnosed, my psychologist told me that I had to reassess my life. Many behaviors that people had misinterpreted as laziness, carelessness, or lack of commitment were actually manifestations of my ADHD.

Friends who thought I didn't care when I forgot plans, teachers who believed I wasn't trying hard enough, and colleagues who saw me as disorganized - they were all seeing untreated ADHD symptoms, not character flaws.

Understanding this was liberating because it meant I wasn't fundamentally flawed as a person. I had to rebuild myself, my confidence - it was a new start in life.

It's a process to relearn and teach yourself that you can do it now. Labeling publicly, saying to your friends and family that you are ADHD makes it so that you OWN your change, you OWN your disability.

tldr, ADHD as an IDENTIY is for me : Reclaim control over your narrative instead of letting others define your behaviors

Create accountability for yourself and set realistic expectations with others

Remove shame from the equation by openly acknowledging your challenges

Enable yourself to access appropriate accommodations and support systems

You can build your own MCP client https://modelcontextprotocol.io/quickstart/client

You will need an ANTHROPIC_API_KEY though.