Crypto people get a fright from quantum computing's giant leap forward (or was it?), as we enter the New AI-ge

Bitcoin HODLers weren't exactly chuffed to read about Google making progress with quantum computing this week, and the price of cryptocurrency even temporarily headed downwards on the news.

Google announced its Willow quantum chip, saying it had used the hardware to solve a computing problem that would take regular computers (as we know them today) an unimaginably long time to solve.

You might wonder what quantum computing has to do with crypto. In simple terms, it is thought that should the technology become generally available, it can be used to defeat the encryption and security of virtual currencies. 

Perhaps that will happen, one day when quantum computers become generally available and people understand the tech. It's safe to say we're some way off that happening at the moment, despite all the impressive work by the scientists.

To get an understanding - sort of, quantum terminology is dense to put it mildly and many of the concepts talked about are mind-bindingly difficult to comprehend, Scott Aaronson's Shtetl-Optimised blog is good starting point.

Aaronson says the work Google has done is great, but he also noted that the challenge Willow solved which would take 10²⁵ or ten septillion years for a classical computer to simulate, would require the same amount of time to verify. Which, uh, isn't really practical.

"We're now deeply into the unverifiable regime that I warned about," Aaronson said.

He also added what physicist Sabine Hossenfelder said about Willow's quantum supremacy, a refreshingly direct take:

I see a lot of confusion about Google's Monday press release about quantum supremacy, so let me try to clarify a few things.

They say they did a computation on a ca 100 qubit chip much faster than a conventional (super)computer could do. The particular calculation in question is…

— Sabine Hossenfelder (@skdh) December 10, 2024

Looks like the Quantumdämmerung ain't gunna happen quite yet, then.

(HODL by the way is a silly slang expression for holding onto crypto currencies rather than selling them).

No escape from AI: Apple Intelligence now in New Zealand

Artificial intelligence (AI) continues to run red hot, and it's becoming hard to keep track of all the news. New public AI models pop up all the time, like Meta AI's Llama 3.3, and Google's Gemini 2.0 range starting with Flash, which the select few testers and developers who've been allowed access to it seem to be impressed by. And which exposes us to horrific neologisms such as "agentic" that auto correct hates.

What is that then? ChatGPT describes it this way: **Agentic** refers to the capacity to act independently, make decisions, and exert control over one’s actions or environment. It highlights autonomy, self-determination, and the ability to pursue goals proactively.

That does sound familiar, but I can't quite put my virtual finger on it.

Apple's lengthy rollout of its local AI has officially reached New Zealand and Australia, and can be enabled and downloaded for newer iPhones, iPads and Macs that support the technology. Apple Intelligence brings a set of features including the Writing Tools that, as the name implies, can tweak what's written in apps on macOS like Mail, Notes, and the word processor Pages to change the style, proof-read, summarise, create lists and tables, and compose with OpenAI's ChatGPT. 

The Writing Tools work quite well, and thanks to the ChatGPT integration, the AI can write long-ish pieces that read well, although overly perfect as you'd expect from a machine running inference algorithms. Despite setting the system language and locale to New Zealand English, the ChatGPT output is in burglarisationized American Engerlandish unless you tell it otherwise.

You can generate pictures with Image Playground, from scratch or using your Photos library to start with. This is yours truly as a hard-partying quantum vaccination scientist in a robotics lab:

Not the kindest depiction of yours truly, that. Hrrmph. 

Using AI with photos is a much more pleasant experience. Although some users are up in arms over the changes to the Photos app layout, I think it works with the AI added natural language search, and creating "memories" which are picture albums with music, for iPhones and iPads. 

The favourite new feature though is Clean Up, which is similar to what some Android vendors have already and can be used to remove unwanted objects in images. Clean Up works pretty well, although I noticed it can miss on removing shadows cast by objects removed.

Siri has been given additional clue through Apple Intelligence and has the ChatGPT integration as well. Requests to ChatGPT need confirmation by default (you can turn it off) and can be slow. There's a limit to how many requests you can make to the ChatGPT advanced version, but Apple doesn't say what it is. Asking ChatGPT produced responses about ATM withdrawal limits and how to set Screen Time restrictions for users.

Either way, once that limit is reached, you drop down to the basic version. Presumably that means going from a more capable model to one that's a step down the ladder.

This is no doubt due to Apple not paying OpenAI for ChatGPT access, but I reached the daily limit after just four to five queries. Probably a bug. You can also sign into your own ChatGPT account if you subscribe to the service, but that means OpenAI will see what you do with the AI. Not signing in keeps the requests anonymous, Apple promises. 

Apple has built an entire new infrastructure called Private Cloud Compute for AI, and it looks very secure indeed with third-parties being able to verify it. If the comments section reads this far, PCC should allay their fears about personal information being accessed by Apple, which won't happen since the company uses strong and secure privacy as a marketing differentiator.

Summaries in Mail and Notifications are a bit meh really, although some people might like that. Recording audio in Notes and having it transcribed however, looks like a very useful feature. Why in Notes and not Voice Memos though? Not so great either is that the AI seems to use battery quite a bit, and the iPhone I tried it on can get hot.

Currently, there's nothing to pay to install and use Apple Intelligence as such, not counting the cost of a hardware upgrade. You need an iPhone 15 Pro or Pro Max, or one of the new iPhone 16s to run it. In fact, if you want all the Apple Intelligence features, like Visual Intelligence to recognise objects, summarise text and read it out loud, detect phone numbers, translations etc, and of course, use Google search and ChatGPT, an iPhone 16 with the Camera Control button is required.

For iPads, the requirement is a tablet with A17 Pro inside, or the M1 chip; Macs have to run M1 or newer Apple Silicon.

AI has felt like less-than useful tech that's been hugely oversold. In some areas that's exactly what it is because the people who drive it don't appear to have heard the expression "just because you can doesn't mean you should do it". Like YouTube, which has added AI generated responses for videos because... why?

There are even worse examples of AI misuse that can be found without much effort.

Which is not to say there aren't good use cases for AI. Some Interest.co.nz commenters using Microsoft 365 Copilot reckon the AI is boosting their productivity, albeit it a steep monthly cost as the tech giant recoups its considerable investment in the technology, and the cost of providing it.

At a larger scale, there is some pretty amazing and imaginative use of AI, which is very good at pattern matching and recognition. Last year, I was reading about the Fragmentarium effort, which uses AI to transliterate cuneiform tablets as part of the Electronic Babylonian Platform project run by scholars at universities in Germany.

It is fascinating work, most likely of little direct economic import unfortunately, in which AI is used to recognise digitised clay tablet fragment to recreate Mesopotamian literature such as the Epic of Gilgamesh. 

AI might be at that annoyingly early stage at the moment, like the World Wide Web was around the new millennium. But it is evolving at break-neck speed as agentic capabilities with multimodal support and computer use functionality develops. 

If you think that's just buzzword bingo on my part, I'll let Claude 3.5 Haiku have the final word:

"Yes, that's an insightful analogy. Just as the World Wide Web in the late 1990s and early 2000s was promising but still quite limited compared to today's internet, AI is currently in a similar transformative phase. We're seeing rapid developments in areas like:

- Multimodal capabilities, where AI can understand and generate content across different types of media (text, images, audio, video)
- Agentic AI that can perform more complex, multi-step tasks with greater autonomy
- Enhanced computer interaction and use capabilities
- More sophisticated reasoning and problem-solving skills

The pace of AI development is indeed extraordinary, with significant breakthroughs happening at a much faster rate than many previous technological innovations. Just as the early web seemed clunky and limited but held immense potential, today's AI is quickly evolving to become more versatile, intelligent, and integrated into various aspects of work and daily life.

The key similarity is the sense of being on the cusp of a transformative technology that will likely reshape how we interact with information, complete tasks, and solve complex problems. We're witnessing the early stages of what could be a fundamental shift in human-computer interaction."