This year there are quite a few joke papers cross-listed in the popular physics category. My favourite: "Is Winter Coming?"
Particularly memorable entries from previous years include "Novel approach to Room Temperature Superconductivity problem" and "A solvable string theory in four dimensions."
This paper with an introduction brazenly written by ChatGPT attracted a lot of attention last week. How is it that the first line of the introduction could remain in the final version without anyone (authors, editors, referees, proofing staff) noticing?
Some said this was no big deal - aren't paper introductions boilerplate junk that nobody reads anyway? Yes and no. While an expert in the field might not expect to learn anything new from reading a paper introduction, it is nevertheless important as a means for the authors to convince the reader that they sufficiently understand the context of the research and are in a position to make a novel and significant contribution.
Others argued this was an example of the failure of peer review
and the current scientific publishing system - junk papers that no one
(not even the authors!) read.
Who exactly is at fault here (apart from the authors, obviously) - the journal editors or the referees?
Actually, it is not the referees' job to proofread manuscripts! Many referees will not bother to laboriously point out all the obvious typos in a manuscript and will purely focus on the scientific content in their reports. Sloppiness that the authors fail to notice themselves will detract from the credibility of the science reported and may be more damning than scathing technical criticism by the referees that might not be adequately addressed in the final paper!
The editors should have caught this in their initial screening. One of the roles of an editor is to curate content and ensure that the valuable time of the volunteer referees is not wasted on obviously incorrect, unconvincing, or not even wrong manuscripts. At the same time, we don't want to waste the authors' time by agreeing to send the manuscript out for review and then being unable to secure willing referees!
At Physical Review A we desk reject about half of the manuscripts we receive without sending out for peer review. While this might sound like a lot, these manuscripts tend to be of much lower quality than those that are eventually published. There are several red flags that make us lean towards desk rejection:
Out of journal scope. Does the manuscript report results that are of interest to the readers of the journal? One simple way to gauge this is to check the reference list of the finished manuscript - if you are only referring to works from other disciplines, this is not by itself grounds for rejection, but it is a hint that you need to be particularly careful with explaining the relevance of your work to the journal's specific audience.
Poor presentation. Obvious typos. Ugly figures. No figures (passable in rare cases). Too many figures. Illegible axis markers. Incorrectly formatted equations and symbols. Basic stuff, but many authors sadly cannot be bothered.
Transfer after rejection from a sister journal. This one is surprisingly common, particularly for research topics which fall in the scope of multiple APS journals. Most often we see transfers from PR Applied and PRB, which have higher impact factors, so the authors decide to try their luck with PRA. But the standards of all these journals are the same, regardless of their impact factors that fluctuate from year to year. This means that rejection from PR Applied or PRB generally precludes publication in PRA, except in special cases.
No significant new physics. This is the most controversial. Who is the editor to decide what is significant - isn't that the job of the referees? We do lean towards giving the benefit of the doubt and sending out to referees for this one. The manuscripts that fail this test generally lack the "so, what?" factor - assuming all the claims are correct, have we learned anything new? It is always possible to tweak models, change terms, make them a bit more complicated, and then apply analysis tools that are standard for the field to get something that is technically correct. But the impact of such technically correct works will be limited unless they open up something new - a novel experimental platform, a way to push the limits of existing theory, and so on.
It is never pleasant for one of your articles to be rejected without
review, but it is actually the second best response you can receive! The likely
alternative would be for you to wait months before receiving a similar rejection on the basis of anonymous referee reports!
It seems every day there's another news article about how large language models (LLMs) are going to replace hundreds of thousands of jobs.
At least in science, we are safe (for now) because the model outputs cannot be trusted and must be carefully vetted for factual accuracy. Thus, we should be thinking about applying LLMs to tasks in research where accuracy is not essential.
Next week I will be giving a talk on reviewing the scientific literature to the graduate students at CQT. Whenever I have to present a somewhat dry-sounding topic it is nice to begin the talk with a semi-humorous ice-breaker to grab the audience's attention. For example, one might try to relate the topic to a popular movie or book.
This is an area where LLMs can really shine! I quickly obtained some good ideas using the 7B and 13B parameter versions of LLaMA (running locally using llama.cpp). The prompt used is highlighted in red:
It was only a matter of time - the resource-intensive, proprietary, pay-to-use ChatGPT now has a free-to-run and use competitor in the form of a tweaked version of Facebook's LLaMa model!
llama.cpp is an super-fast implementation of LLaMa in C++ obtained by reducing the numerical precision of the trained model weights.
I installed it without any trouble, in contrast to other generative models where dealing with the python dependencies and incompatible library versions can be a massive pain.
Similar to ChatGPT, the raw generated output text might look convincing to a non-specialist even though it is spouting nonsense. Thus, careful editing of its output is essential if you want to use it for anything other than your own amusement.
For example:
Prompt: Quantum neural networks and quantum kernel machines are two promising approaches for machine learning using emerging noisy intermediate-scale quantum computers [1].
Output: Quantum neural networks and quantum kernel machines are two promising approaches for machine learning using emerging noisy intermediate-scale quantum computers [1]. They exploit the inherent parallel nature of existing classical deep network architectures through encoding them into a series of unitary gates, which can then be applied to qubits. However in most cases their expressive power is severely limited by noise that quickly accumulates during gate execution and corrupts information encoded on quantum states [2]. Although methods have been developed for dealing with this issue (either by using error correcting codes or applying a particular type of operations, namely, magic state techniques), they are not fully successful in practice. In other words, current approaches to mitigate errors suffer from an excessive overhead that makes these methods impractical and therefore useless [3]. In this paper we describe novel quantum circuits for encoding NNs architectures onto qubits with a significantly reduced number of gates and without using magic states or error correcting codes. These new constructions are not only more efficient, but also less prone to errors since they employ fewer physical operations (and therefore have lower noise), which could lead them into practical use in future NISQ devices [4].
The pace of machine learning research is incredible.
Another entertaining and insightful perspective by Jacob Khurgin was uploaded to arXiv this week: Nonlinear optics: a look from the interaction time viewpoint and what it portends. It is similar in tone to his earlier article on high refractive index materials.
This is real old-school physics - using simple intuitive models to understand fundamental limits to nonlinear optical response and "the universal principle of unavailability of free lunch."
On the one hand, the take-home message is somber: no huge magic enhancement of nonlinearity is possible due to fundamental physical constraints; hype regarding various wondrous materials does not stand up to scrutiny. On the other hand, finding new "boring" materials exhibiting modest enhancements to properties such as optical damage threshold or propagation loss is still a worthy goal as a means of improving the performance of existing nonlinear optical devices including light sources based on harmonic generation, optical frequency combs, and parametric amplifiers.
Giving your model or result a catchy name greatly increases the impact of your research.
Compare the citations of Two-dimensional massless electrons in an inverted contact and Quantum spin Hall effect.
The title you give your paper is important. Don't rush it.
On a related note, those working on soliton theory will be familiar with the nonlinear wave equation
$$\partial_t^2 \phi - \partial_x^2 \phi + m^2 \sin \phi = 0,$$
which is called the Sine-Gordon equation "for obvious reasons" in Rubinstein's original analysis of its soliton solutions. A footnote in this paper gives credit for this brilliant name to Professor Martin Kruskal, who has an impressive list of scientific achievements spanning nonlinear waves, surreal numbers, and wormholes.
I was reading about sorting algorithms the other day and stumbled upon the amusing topic of pessimal algorithms, which are horribly slow algorithms that "[do] indeed progress steadily towards [their] stated goal even though [they] may have very little enthusiasm for (or even a manifest aversion to) actually getting there."
Two examples of pessimal sorting algorithms are slowsort and bogosort, which have non-polynomial scaling (factorial scaling in the case of bogosort). Slowsort uses a multiply-and-surrender approach, while bogosort chooses random permutations until it obtains the correctly-sorted list.
A simple quantum version of bogosort might be a circuit composed of a single Hadamard gate applied to each qubit, followed by measurement; this similarly samples uniformly over the solution space, repeating until the correct answer is measured.
More generally, pessimal quantum algorithms might exploit maliciously-tuned quantum interference to obtain scaling worse than the most pessimal classical algorithm, for example by engineering destructive interference to suppress the probability of measuring the correct solution.
Are there any known examples of pessimal quantum algorithms exhibiting a rigorously-proven quantum slow-down? Pessimists might point to certain variational quantum algorithms and approaches based on mapping efficiently-solvable problems to more difficult problems such as finding ground states of quantum spin networks as potential candidates.
Pessimal quantum algorithms could be one of the first killer applications of quantum computers, satisfying clients who want to find spooky quantum solutions to their problems (efficiency be damned!) and hardware developers seeking to maximise the utilisation of their shiny new quantum processors.
Will future fault-tolerant quantum computers be scaled-up versions of existing quantum processors, or will they be based on completely different materials with intrinsic fault tolerance? Dave Bacon (formerly of IonQ and Google Quantum) envisions a middle ground in which current noisy qubits may be cleverly arranged to create intrinsically-tolerant topological qubits without requiring the massive overheads of quantum error-correcting codes: Quantum Computing's Middle Way
On a lighter note: The Quantum Hype Scorecard
