I have enjoyed this discussion and honestly am amazed I never discovered the depth of thinking represented here – a forum for writing software. (Never judge a book by its cover, right?)
I need to get back to my novel writing for now, but if I were to boil down my perspective, it would be stated thus: combining quantum coherence with neuromorphic architecture will yield a form of intelligence so vast we might not be able to imagine the implications for humanity. And it’s happening much faster than many of us suspect–unless you’re OpenAI, Microsoft, and NVIDIA. Just follow the billions being poured into it.
Back to my eleven-year-old boy who’s lost his memories…
First and most important, they haven’t yet established a quantum entanglement. They have 6,000 individual atoms, in a stable configuration – which is indeed a huge advance! – but they haven’t established the ability to compute with them. (Briefly, preparing the array with a specific waveform and allowing it to evolve in a controlled way, without breaking the quantum superposition.)
Second, individually controlling 6,000 atoms with laser tweezers in a purpose-built vacuum chamber simply is not scalable to practical computational systems with hundreds of thousands or millions of qubits. Look at the diagram of the “tweezers” that they provide, and multiply by 600,000.
And finally, they handwave their way around the issue of noise, which is the single greatest obstacle to scalable quantum computing. At the quantum level, the world is simply a very noisy place.
Well, I guess I’ll pick up my novel writing in the morning, haha.
You make good points.
Entanglement is indeed necessary for quantum computation to utilize error correction and multi-quantum-bit gates. The paper indicates the achievement of transport and control without losing coherence for long periods. A big next step is entanglement, as you’ve said. My observation is that we are good at entanglement, having accomplished it experimentally in 1972. Let’s see how good we are.
Scalability is also a huge hurdle, no doubt about it. Caltech proposes a zone-based architecture. It will be interesting to see how far this takes us. We need to keep in mind that scalability is not a binary concept, but rather an engineering continuum. The point is that the paper demonstrates that scaling is possible, and that’s huge.
Sure, noise is the big challenge with quantum computing. In noisy systems, coherence lasts just milliseconds, if that, making it very difficult to use in a computing system. In the paper, a coherence time of about 12.6 seconds with an imaging fidelity of over 99.99% was achieved. Thus, the system was able to resist noise long enough to achieve stable computing, in principle.
So, scalability and noise resistance are within our grasp.
I have no idea how soon this will be combined with a neuromorphic intelligence architecture to produce the kind of intelligence that the natural world (on Earth, anyway) has not produced and likely never will. I am, though, impressed (and yes, terrified) with the galloping progress being made.
AI will improve in discrete areas, for sure. But LLMs are limited, and there is no guarantee that increasing the computational effort will improve the oracle-like results that the marketing droids are pushing. The maths simply does not support this.
While I’m bullish on the broad scope of AI in constrained domains – which are to some extent already proven, but failing magnificently to drive a car safely – I am sceptical of the marketing claims of quantum computing. But this is a futile argument, because most folk can’t do the sums, which is where the truth lies.
Anyone who writes fiction knows that the ideas don’t come from their brain’s processing power but from some weird other space. Art is not rational.
I don’t think human-generated art is going away, I just think the number of humans who will be able to support themselves financially solely through creating art will dwindle dramatically. Art will be a hobby for most people.
My problem is that my views on what I think will happen are heavily skewed by what I hope will happen rather than a deep understanding of {the things I’d need a deep understanding of to opine intelligently}.
If I’m honest, I think you’re right.
{grumbles}
I would absolutely love you to be right. It is heavily subsidised at the moment and hasn’t had to “carry its own water,” but that’s not a reason to think it never will.
So, it is possible, but only if some people who have bet very big are very wrong.
That, unfortunately, is the argument for changing legislation to ensure a commercial market, not for why the market will fizzle out.
There quite simply is too much invested by the people who run the county and by the people that own the country and by the people that control the people that pay for the country to let this be foiled by failing to do something as simple as a minor tweak to the law at the expense of humanity’s soul.
It doesn’t need to. It just needs to be good enough to fill the “reels” section of Instagram, the “shorts” section of YouTube and whatever is on TikTok. If it entertains the masses (who have shown themselves endlessly amused by people reposting and remaking the same content over and over) then it will take over.
The most misunderstood thing about bubbles is that people think that they take everything down with them when they burst. They don’t. We still have tulips; there is still housing in Japan, and in the US for that matter; countries still trade with South America; the internet is, I believe, still alive and well (you are reading this on a “dotcom” website right now).
The bubble bursting takes out the businesses that had no solid basis and the fools that invested in them. It doesn’t destroy the underlying technology. What’s left after the burst is the viable stuff, which suddenly finds itself with little-to-no competition and a really cheap supply of resources from the ashes of the insolvent.
Now we naturally have way no of knowing for sure which bits will die and which will remain, although it’s usually obvious with hindsight so we can hazard a pretty good guess if you look without emotion or greed. You’re looking for the amazon.coms whose underlying businesses make sense, not the countless startups whose business cases were little more than a statement of a route to market without any ability to describe the problem they were there to solve.
Sure. There are lots of very interesting applications of machine learning in medicine, the sciences, and so on. But LLMs specifically have simply not demonstrated that they can do the things that their creators are claiming.
I’m not so sure. While there are certainly overly optimistic claims, I have seen some pretty remarkable results in the wild:
one friend of mine was having car trouble (they are not a “car person”) and they used ChatGPT voice activated as a “car coach” and it walked them through the process of diagnosing the problem and fixing it themselves. Pretty cool.
another friend had been going to therapy and creative coaching for years trying to get over some psychological resistance that was interfering with their productivity. An artist colleague recommended using ChatGPT as a psycological coach. My friend tried it, and because the LLM had such a comprehensive selection of therapeutic texts in its memory, it was able to zero in on the cause of the problem. My friend is now doing daily exercises from well-researched therapeutic approaches, recommended by the GPT, and reports real progress for the first time in years.
a third friend is uncluttering their garage and uses Grok to research comic book values, and pick out special issues that are especially rare and valuable. Grok even writes the ebay ads for them.
LLMs are here to stay, even if it is a bubble and it bursts.
As this thread goes even further and further from the “adding AI to Scrivener” promise in the title…
I had an interesting meeting at work today. We have an “away day” next week (it’s actually taking place in the office basement so it’s more just a “day in the office”) and there will be team awards.
Someone in the team made the excellent suggestion that it would be great if the big boss would say a few words about each award category as they’re given out to really personalise the wins and show how much everyone’s efforts and hard work means to the leadership.
Of course, they agreed and immediately instructed someone in the meeting to get copilot to write their “personal” remarks. Which now I think about it, did at least accurately demonstrate to everyone on the call how much everyone’s hard work means to them.
As someone who has benefited from a lifetime of therapy from actual humans with not only a wealth of medical and psychological knowledge but also empathy, compassion, and a desire to help, this is both terrifying and depressing. Having access to all the text means almost nothing when the LLM can’t do the human part. How does it know when to disagree as oppose to affirm? How does it know when it could be causing long term harm for short term benefits? How does it know anything other than what the text says?
Seconding this. It has already proven to be incredibly harmful to desperate and vulnerable people.
