Jonathan Rudderham

Go local, or go loco.

[Jonathan Rudderham @codeRunnerUK]

@ivanfioravanti It sends me to something called “Locally AI” yet the post claims it’s “LM Studio’s mobile app”. That’s not going to confuse people at all.

[Jonathan Rudderham @codeRunnerUK]

I asked Claude that question - given what you know about me, will AI replace my job in 12 months. It gave me a long detailed answer saying that it wouldn’t. To which I replied that it just goes to show that in all the months we’ve been working together, it knows nothing about me. What it thinks of as my job (coding, AI, etc) is my hobby.

[Jonathan Rudderham @codeRunnerUK]

@ivanfioravanti What’s the problem with the 64K+ context? I’m planning on adding this to my long context coherence test (up to 173K words).

[Jonathan Rudderham @codeRunnerUK]

@ivanfioravanti @filipstrand Just something I vibe coded so I can select from models, image gen, image edit, outpaint, upscale, remove backgrounds, just the usual stuff. It’s a bit of a flaky Streamlit-Python app which needs redoing to be honest. I have a similar app for old SD models, and for LTX2 video.

[Jonathan Rudderham @codeRunnerUK]

@loktar00 @Alibaba_Qwen Did Qwen3.5-9B have a crown? In my long context coherence tests, 9B is definitely “on the bubble”. It can be good but also terrible. I wouldn’t want to rely on it.

[Jonathan Rudderham @codeRunnerUK]

@ivanfioravanti I’m still just using Claude Code but, even if I wasn’t, there’s no force on Earth that’ll get me using Grok again.

Lol. I guess I’m lucky that, right now, the Sonnet model is working okay for my purposes.

[Jonathan Rudderham @codeRunnerUK]

@PussyBlaster10k @seangeng You’re right, of course. Running a laptop on a battery - whoever heard of such a thing. I apologise for suggesting such a crazy idea.

[Jonathan Rudderham @codeRunnerUK]

@venkatarangan 128GB in mid-2026, really? What is this - the Dark Ages? 😂

[Jonathan Rudderham @codeRunnerUK]

I asked Claude if it thought my job would be replaced by AI within 12 months. At first it said I’m safe, and then explained its reasoning based on all the AI and LLM work we’ve done.

That made me realise that for everything it knows about me, it knows nothing about me. Because that’s my hobby, not my job.

I’m a warehouse worker. So then Claude says I could be replaced by AI.

I then explained what “warehouse worker” actually means - rather than the “monotonous drone” belief that many people (and, AI it would seem) have.

Then it noted that it was highly unlikely AI or automation could replace me.

The primary issue is that humans adapt and react in real time to real world issues and problems. An AI needs to be trained on all the fail scenarios. Yet AI is typically trained in best-case scenarios, not how to handle worse-case scenarios.

People need to stop thinking that jobs they don’t understand are just “mindless drones doing repetitive tasks”. Productive manual workers are the backbone of the country. We were in the past, we are now, and we will be in the future.

[Jonathan Rudderham @codeRunnerUK]

@Daniel_Rubino If it was made for “AI devs”, it wouldn’t be limited to 128GB. This is 2026. 128GB is so last-year.

[Jonathan Rudderham @codeRunnerUK]

@yiningkarlli $2K with 16GB RAM won’t be a rival to anything. $4K for 128GB RAM won’t find a market because people prepared to pay that will already have bought a DGX Spark or clone, or Mac Studio.

It’s mid-2026. This thing should have 192GB at least, and be under $3K, but it won’t be.

[Jonathan Rudderham @codeRunnerUK]

Initially, I didn’t understand it and even now I only partly understand it. What baffles me is someone’s attachment to an AI controlled by a central money-obsessed company that’s always going to bow-down to government overreach.

Using a local model that you, at least, create your own system prompt, backed by RAG, all the way up to training your own - I can understand that. You’ll never lose that model.

But using frontier models is leaving yourself open to the whims of bean counters and government overreach. That’s not going to end well no matter which way you look at it.

[Jonathan Rudderham @codeRunnerUK]

Are there really that many people with their own datasets clamouring to train LLM models on their desk? And, if so, why aren’t they using 256GB or 512GB machines? I can’t image they were all sitting around twiddling their thumbs waiting for yet another limited-to-128GB release.

[Jonathan Rudderham @codeRunnerUK]

I can’t believe we’re in mid-2026 and they bring out a new machine that’s still limited to 128GB RAM.

As much as I’d love to explore the world of CUDA, the 128GB is the deal-breaker.

[NVIDIA RTX Spark @NVIDIARTXSpark]

This is the NVIDIA RTX Spark Superchip. A new beginning for personal computers.

Designed for creators, AI developers, and gamers, RTX Spark brings over 30 years of NVIDIA innovation to slim Windows laptops and small, ultra-efficient desktop PCs.

[Jonathan Rudderham @codeRunnerUK]

I can’t believe we’re in mid-2026 and they bring out a new machine that’s still limited to 128GB RAM.

As much as I’d love to explore the world of CUDA, the 128GB is the deal-breaker.

https://t.co/r9O6dG6EfX

[Jonathan Rudderham @codeRunnerUK]

@LeakerApple I can’t believe we’re in mid-2026 and they bring out a new machine that’s still limited to 128GB RAM.

As much as I’d love to explore the world of CUDA, the 128GB is the deal-breaker.

[Jonathan Rudderham @codeRunnerUK]

@gailcweiner The first time I saw “video moderated due to UK laws” from Grok, I realised that frontier AI was slave to government overreach. Six months later, people in other countries are finally realising the same.

The only solution is to go local, or not use AI at all.

[Jonathan Rudderham @codeRunnerUK]

I wasn’t going to use any of this on my regular machines. When I decided I wanted to try Claude Code and similar, I bought a used M1 MBA. I keep it off my private network, it only has access the MBA so, if it strays outside its folder, it can’t find anything else. One day, if I get too deep into these things and it causes trouble, I’ll just nuke the thing and start afresh.