Phoenix: Open-Source NuBus FPGA Accelerator for 68040 Macs

[stepleton]

Decades ago I read a quote from Larry Ellsion. The founder of Oracle suggested to design a distributed way to interconnect computers, to prevent a monopoly of parties that might be interested in centralised control (as they exist, today).

People sure do change. "Citizens will be on their best behavior because we are constantly recording and reporting everything that’s going on." Larry Ellison, Oracle Financial Analyst Meeting, 12 September 2024.

He posited in that meeting that AI systems ceaselessly monitoring feeds from ubiquitous surveillance video cameras would make it possible.

 

[CC_333]

People sure do change. "Citizens will be on their best behavior because we are constantly recording and reporting everything that’s going on." Larry Ellison, Oracle Financial Analyst Meeting, 12 September 2024.

He posited in that meeting that AI systems ceaselessly monitoring feeds from ubiquitous surveillance video cameras would make it possible.

Ugly Thought!

That sounds WAY too much like Skynet!

c

 

[Builder68]

Ugly Thought!

That sounds WAY too much like Skynet!

c

Skynet wanted to wipe us out, this sounds way more like George Orwell’s 1984, where Big Brother just wants to watch us 24/7 to make sure we're behaving!

Instead of killer robots, it's basically the high-tech, AI-powered version of the "tele screen" in every room. Larry Ellison is basically describing a digital Thought Police!. Honestly, this is old news; it already caused a massive wave of criticism back in 2024 when he first made those comments.

 

[CC_333]

True points.

And it's excellent he got criticized for his comments. We need more of that! The tech CEOs need to know that we don't like what they're doing!

c

 

[adespoton]

A big limitation is that the freely accessible LLMs have limited knowledge of vintage hardware and retro programming so struggle to produce usable results. To get good results you have to spend time preparing the AI with rules, information, examples, documentation, then spend time, much time, walking it through.

It is an amazing tool and a force multiplier, allowing an individual to do the work of a team, but in the same way that an Airbrush doesn't make you a better artist than a crayon, you still need to put in time, effort and skill.

Long and short, ChatGPT is probably useless at HyperTalk out of the box, but if you provide documentation, and use agents to review each others code, develop tests to define success and sit interrogating and troubleshooting the problems with the AI, you'll potentially write the next Myst in a lot less time than it took Cyan.

The problem with the post at the start of this thread was that the quality of the hardware element was a bit low. The files were presented as first pass solutions requiring a little tweeking... But in fact didn't even open in the software they were supposedly for. That... Was... discouraging for a project. I'm been polite with that description to be honest, and ignoring the manner it was presented.

Interestingly, most of the current LLM models appear to have ingested the complete Inside Macintosh and the Think C reference books. The problem is that until you get to the frontier models, the context window is too small to handle analysis and disassembly on the level required here.

If you've got access to the latest security models from Anthropic or OpenAI... well, you aren't going to be using them for this purpose, but they could theoretically fit the bill (at large expense).