neolefty

From the article:

> Auto-brewery syndrome may carry substantial social, legal, occupational, and medical consequences for patients, and awareness of this syndrome is essential for clinical diagnosis and management.

I had assumed both — greater volume and also lower temperatures. The more qbits you are trying to get to cooperate, the less thermal noise you can tolerate, so the cooler you have to make it.

Yes, more qbits also take up more space, but I hadn't thought of that as a major factor — but it certainly could be if they are physically large! Overall I think the bigger issue is cooling.

The article mentions error correction as an alternative to increased coherence among the qbits. Perhaps what that really means is "to increase meaningful interaction among qbits, make it as cold as you can, and then error correct until you get a meaningful answer." My intuition is that they are both a battle against entropy — the need for error correction will also increase exponentially with the number of qbits, simply because the number of possible combinations increases exponentially.

And the even larger outcome of all this is that if this intuition bears out, quantum computing will have no fundamental advantage over conventional computing — which also has an exponential cost for a linear increase in bits, for computations such as factoring large numbers.

That's also plausible. My guess is the same logic would apply — you would need exponentially more error correction for a linear increase in qbits.

Well, we are always using our internal models for work like this. When our internal models and what we say don't match reality, we might call it "hallucination" or "getting it wrong" or even "lying".

When we get it right — our internal model matches reality, and we also say something that matches reality — we call it "truthfulness", and it's super valuable.

But I think there are two really different sources of hallucination / inaccuracy / lies. One is that the internal model is wrong — "Oops, I was wrong. The CVS isn't on Main St." The other is when we decide to deceive. "Haha, I sent you to Main St. CVS is really on Center Rd." Two very different internal processes with the same outcome.

We think of truthfulness as boolean — whether in humans or AI models — but it may be more useful to think of it as a skill. You can become more truthful by aligning your inner model with reality and then also by checking it with ground truth. A challenging and intrinsically imperfect process either way!

Maybe you mean "work from an internal model of the world"?

DNA could also be considered 6-bit, since codons are 3 base pairs: https://www.genome.gov/genetics-glossary/Codon

The article describes both. The 1,0,-1 networks are "1.58 bits" — that is, log2(3). But yeah it mostly focuses on 1-bit networks.

The more certain the domain, the more that is possible. If you have a document database that you trust, great. For example a support desk's knowledge base. And especially if you have an escape valve: "Did this solve your problem? If not, let's escalate this to a human."

But if you are searching the Internet, you'll find multiple answers — probably contradictory — and the next step is to ask the model to judge among them. Now you want all the intelligence you can muster. Unless you really trust the search engine, in which case yeah a small model seems great.

Not the size, but the temperature. If you have to cool to a microkelvin for a certain number of qbits to retain coherence, how low do you need to go to add one more qbit, and how much energy will that require?

My thermodynamic instinct says that the cooling effort required rises with the resolving power — which is exponential with the number of qbits. But it's just instinct, not grounded very well in science or engineering.

> The final section was of particular interest to me; Gil Kalai's work on quantum error correction is very interesting to me and I am in the camp that believes that quantum computing is not possible in any useful sense; in particular a quantum computer will not be capable of being significantly more powerful than a classical computer, in the quantum supremacy sense.

My question too. I've had a vague feeling about this for a long time, waving my hands about thermodynamics with "It must get exponentially harder per qbit to eliminate thermal noise by cooling down closer to absolute zero," and I'd really like to get past my hand-waving and see what the dynamics really are.

Probably responding to this from a parent comment:

> Shoving 100 cattle into a 1 acre feed lot for their entire lives

That would make sense if you knew how hard things were going to be and if you could predict external responsibilities.

For example if I plan to meet a friend tomorrow, but tonight my child unexpectedly gets sick, I can call the friend and say, "Sorry my child got sick. How about next week?"

That might happen multiple weeks in a row, but eventually (assuming a patient friend) we'll get together and renew our connection. On the other hand, if I played it safe, and said, "I'm too busy for friendships now that I'm a parent," we would both have less friend-time.

This market is troubling. But I have a different question:

What does the long game look like for raw training data? How will AIs maintain the quality of their diet?

To compare, web search started — in the early days of Google — as a huge win because so much valuable information that was scattered around became findable. But over time it has become whac-a-mole with spam and AI copypasta, and now it's a struggle to keep returning good results, for any search engine.

The point is you won't feel it (aside from air resistance) on the way down.

How does it scale to such a large context window — is it publicly known, or is there some high-quality speculation out there that you recommend?

Per another comment here, it's probably autoimmune related. So it's complicated!

Yes, I think the research team is saying exactly that:

> "The concept that many other viral infections that lead to post-viral syndromes might drive blood vessel leakage in the brain is potentially game changing and is under active investigation by the team.” — Matthew Campbell, Professor in Genetics and Head of Genetics at Trinity, and Principal Investigator at FutureNeuro.

The article says it's both.

> "For the first time, we have been able to show that leaky blood vessels in the human brain, in tandem with a hyperactive immune system may be the key drivers of brain fog associated with Long COVID. ..." — Prof. Matthew Campbell, Professor in Genetics and Head of Genetics at Trinity, and Principal Investigator at FutureNeuro.

As for healing, I'd think blood vessels could! The article doesn't say whether it's an autoimmune problem, so I suspect the researchers are being conservative or don't know?

> "The concept that many other viral infections that lead to post-viral syndromes might drive blood vessel leakage in the brain is potentially game changing and is under active investigation by the team.” — also Prof. Campbell

So cool! I especially like the "122-millionth cousin, 28 million times removed" description and the fact that it has many different pictures for "Human".

Here's Geoff Hinton discussing spiking neural nets: https://youtu.be/2EDP4v-9TUA?si=E4D5YNGQGdYSiTIy — from an interview with Pieter Abbeel in June 2022.

From the paper:

> Spiking Neural Networks (SNNs) have been an attractive option for deployment on devices with limited computing resources and lower power consumption because of the event-driven computing characteristic.

But training them requires new techniques compared to continuous neural networks — SNNs aren't differentiable and therefore you can't back-propagate (as I understand it; please correct me if I'm off here).

Lex Fridman too: https://www.youtube.com/watch?v=OgIo36F6Fsg — great 2-hour conversation.

Thunderbolt docks too — with a single cable they can drive two 4k displays plus provide power to the laptop.

I'm running an M1 Max with two Thunderbolt docks, and each drives 2 4k displays, runs great, although it's kinda overkill. But it does require the docks; you can't connect directly.

Presumably UBI would come from taxes. So it's wealth redistribution. It would reduce the difference between richest and poorest.

Mike?

So did they follow up?

Anecdotally, on a desktop machine with something like 64GB RAM, when I switched from non-ECC to ECC (and 128GB RAM), the number of unexplained crashes seemed to drop from one every couple of weeks to one every 6 months or so. Could have been from a software upgrade though, probably Ubuntu 18 to Ubuntu 20.

Now I'm on an M1 MBP with 32GB RAM, and I get unexplained crashes roughly every 3-4 months, but I've been chalking it up to MacOS. Behavior also seems to degrade after a month or two of uptime. Hmm, maybe it's time for Apple to go ECC ...

SEEKING WORK | Indiana, USA (Eastern timezone) | REMOTE | Full stack & ops, mentoring, team facilitator

Remote: Yes, but can be on-site 1-2 weeks per quarter.

Technologies: React or other modern front-end framework, Python, .NET, all relational databases and provisioning workflows.

Resume: https://neolefty.org/resume/resume.txt, https://github.com/neolefty

Blog: https://neolefty.org

Email: bbbaker@alum.mit.edu

Hi, I'm Beecher, an experienced engineer (22 years) who can help a team gel or navigate a difficult transition — I lead by example, mentor, listen, make friends, and communicate. Fully remote since 2018. Experience with startups, established companies, and scientific research. I love a good startup, can wrangle a crisis, and navigate legacy situations / upgrades.

Just as gaming hardware has taken over compute, gaming journalism must take over reporting!

It is so crazy how good they are. I love my Apple hardware (M1 Max, four external displays FTW) and tolerate my Apple software (Mac OS).