SillyUsername

Now is the time for the RPI to diversify into features adjacent to those adding cost based on the market usage.

They did this with the Pi400 (keyboard added), maybe now faster pcie and external device support, faster networking, indicator LEDs, built in UPS, maybe even a built in SDR (both media players and hacking tools would benefit).

Thanks for the correction, what specifically makes them dead end that we can't re-use the designs - e.g. use modern parts?

This kind of makes the UK's double budget overrun Hinkley Point C new nuclear powerstation look even worse as a policy choice.

This was the best bet at the time when it was chosen, but it must be a good case of the sunken cost fallacy by now, given how long it's taken to build, and didn't follow an existing design*

As a risky policy choice, many distributed solar and wind farms with modern sodium-ion batteries would be a safer and more distributed (failure and weather resilient) modern choice.

*If the original design was defective, why are they using it in A and B?

So they knowingly deployed a configuration with an overflow risk, much like writing code with a buffer overflow exploit.

They assumed that it wasn't like an overflow exploit, despite the fact that's exactly what they'd set up to happen, on the grounds of thinking the equivalent of, "oh this probably uses smart pointers, we'll be fine".

That's how I read it anyway.

Code is often PR correct but business logic buggy, so how are you finding this is working out for you so far? Had many issues in production?

This is no different to our parents generation that used to tell us to stop watching TV or get off the games console (like a Gameboy). But this is besides the point.

I feel it's great, but then I use Google Kids on the Fire, because Amazon's app is junk in comparison.

And also, my kid is (disabled) autistic and non verbal and this keeps them calm and happy.

So yes it's a great idea.

That post will not age well.

Part of the fun of writing your own software are the features it produces, not just the effort to get there.

As more and more code becomes AI assisted their pool of new projects will vanish, people will lie, or they will silently discard this rule.

To quote Thanos "(A)I am inevitable", and the Ironman (admin) making their stand will have their pyrrhic victory.

It's reflecting the inherent tribalism in the training data. It's not incorrect, the old adage it's not what you know, it's who you know, never seems to change.

Maybe they should have tried Qwen 3.8 Flash Next and Strata. Faster than 27B, around 60tps, and more intelligent on benchmarks.

I'm using this on my own 2x 5060TI.

I abandoned 27B on llama - split mode tensor, 200K context (Q5/Q4_1), ubatch 3072, batch 6144, it managed 1900tps prompt processing (same result as article for decode).

Alibaba. Reason: Qwen, open source work.

Meta. Reason: open source work (although this is love/hate due to privacy etc etc)

Google. Reason: Free services, open source work, creation of modern AI (including Transformers).

OpenAI. Reason: Chat GPT 4o, driving the industry forward (less so now)

BBC. For doing us proud.

Raspberry PI Holdings. Reason: Progressing education, affordable SBCs, following a proud tradition (Acorn, ARM, Sinclair)

Espressif. Reason: ESP series products (cheap, easy to use, feature Laden)

LONGi & Trina Solar: Reason: Perovskite solar panel innovations

Runners up (publicly funded so not companies as such):

CERN: Reason: Research and open source work.

NASA: Reason: Research.

Seems to be a paid subscriber article?

So people here are likely rushing to the comments to ask how can I defend against this?

Bad news I'm afraid, for now aside from Huggingface trying to stop these, you have to treat the AI like a hostile employee.

- Don't give your AI secrets access, scope it's fs access as a limited user.

- Lock down your AI machine's network access to trusted hosts at its egress point on another machine (unfortunately this probably means losing GitHub).

- Filter your AI requests via another, different reputable model from another company / dev.

- Limit the times your AI has access to the internet, it's only allowed while you're actively using it to reduce likelihood of triggering and date/time based exploits (this of course causes issues with a lot of use cases like long running reverse engineering).

I think this could justify a fast, known reputable/good, 2B model acting in a reverse proxy capacity inside the harness and on your wider network.

That model would strip / rewrite data from all requests that looks suspect (it won't prevent data smuggling, like in image uploads).

It would also block domains BAU for a proxy using an allowlist.

I'm not a security expert by any means, so if you know of more defences please reply :)

It pretty bad. Just ask it to quote something like the first paragraph of moby dick (i.e. "call me ishmael...") it barely gets past 5 words before it completely gets it wrong. Even the worst normal 4B models get 3-5 sentences.

Damnit, I've been waiting on an open weights model like this one, still no dice :(

It's a language that tries to be everything but is at its heart Enterprise software.

e.g. poor support, or late support for

- JWT/Swing desktop UIs

- Web technologies (it used to have a built in JS engine, competing standards, servlets, jsps, jstl, portlets)

- Image processing (ImageIO is limited, Image and BufferedImage were slow)

- Native interop (overlapping / competing standards)

- RPC (CORBA, Jini...)

- Mobile variants (J2ME, Java Profile Variants, Embedded Java, Java card)

- Threads, Concurrency, native threads (confusing so many re-implementations)

- Delegates, sorry I mean arrow functions, oops no I mean lambdas. The features Sun said should not be in the language for years then back pedalled once it realised the foot gun calling out Microsoft for them.

So many attempts at relevance then abandoned, left in the language as cruft, or just ideas that were vehemently not Java, shoe horned in.

And they still don't have the Elvis operator, not for lack of trying though, they do have the abomination Optional instead though, but best practice says DON'T use that for arguments, just return types (but the language permits it anyway).

For newbies, Python is better, Typescript is more accessible for web or serverless, Rust for native...