I'll make sure to be there 👍

Provided they can get hold of some reasonably big tin... They can host some pretty powerful stuff now locally, and even fine tune it to your needs.

I guess it depends on the use case.

You can configure local LLMs and tools such as Cursor to use those local models for inference.

Using AI tools doesn't need to be a "I'm terrified of a data breach" conversation provided you build the right infra. 👍

Just host LLMs locally? 🤔

Appoint someone to operate in your stead with the same aplomb that you have over the years...

"The Guru is dead... Long live The Guru!"

It's dependent on the mission-criticality of the result.

You'd not want an "ish" answer for nuclear reactor safety, or spacecraft re-entry for example.

I get that.

But you'd be surprised at how many people want to see the end-product... First.

The way we consume information has changed; Example being short-form media, but it applies to everything.

Show someone being able to achieve their end-goal quickly - Go through the detail later if they're still interested.

We've become an instant-gratification society, that wants the short-form answer first. So give people what they want early, and then let them stay for the detail later.

Better still...

...It's good as a potential hire to check the companys GitHub repos to see:

- Commits made in the last 6 months
- How many were made by the interviewers

This put me off joining a company who offered when I realised they'd not done _anything_ since getting VC money for 6 months...

... And if you don't believe this and you think "i'm not using #AI" - Just remember every time you use a captcha that is "identify the traffic light/bicycle/motorcycle" - You're providing training data to train driverless cars on image recognition by labelling it for them.

We are in the "training/correcting" phase of #AI adaptation right now.

For those of you using AI tools, you know this to be true; If something is free, you are the product. Your prompts will be being used to train models. 1/2

Autonomous agents. You don't need a 14-day-week if you've got an agent alongside you, doubling your throughput/quality.

That's the theory. Remove onerous work, to amplify existing workflows.

In practice, it'll be a LOT of correction/training initially which will regress throughput/quality.

Welcome! If you're not already on, get yourselves onto @walkingrandomly.bsky.social's HPC Starter pack :)

Point 22 flies directly in the face of the Turing Institute in discussions around redundancies.

Already mentioned it - One of the biggest risks when it comes to AI is the risk of it creating a two-tier environment.

In the #HPC / #AI arms race... Who loses? It's the businesses who can't afford to generate business benefit from it because they're not large enough to.

Think #SMEs.

So... It's an action plan that says "Let's build infrastructure capacity to make ourselves attractive to tech companies"

"Let's do a renewal initiative to re-industrialise de-industrialised areas, just re-zoning them for a different industry."

It's a path-of-least-resistance approach.

Thanks, I was trying to ascertain some of the more weird and wonderfuls, things like the multitude of ARM CPUs out there but I think i've got it now and I can strip it out of /sys :)

If you want to connect with plenty of people, getting onto @walkingrandomly.bsky.social's HPC Starter Pack would be a good shout