On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? https://dl.acm.org/doi/10.1145/3442188.3445922 >
I enjoyed the real world concerns they gathered but found the octopus analogy a bit weak.
Ask HN: Which recent research paper blew your mind?
141–150 of 174 posts
Re: Ask HN: Which recent research paper blew your mind?
#142More is Different by P. W. Anderson (1972)'arguing that “at each level of complexity entirely new properties appear” — that is, although, for example, chemistry is subject to the laws of physics, we cannot infer the field of chemistry from our knowledge of physics.' The paper https://cse-robotics.engr.tamu.edu/dshell/cs689/papers/ander... Also its impact https://www.nature.com/articles/s42254-022-00483-x
> that is, although, for example, chemistry is subject to the laws of physics, we cannot infer the field of chemistry from our knowledge of physics I mean we could , with infinite computing power and enough time to look into every interesting phenomenon (and to evaluate the corresponding multi-particle Schrödinger equation numerically) but there are simply too many such phenomena, Schrödinger equations are tough to s…
no, that is the point of the paper, even with infinite computing power you could not predict. it is a very strong case against reductionism in science. The phenomenological examples he considers in the paper cannot be derived from basic principles of quantum mechanics.
Re: Ask HN: Which recent research paper blew your mind?
#143https://arxiv.org/pdf/1703.04234.pdf
On why Oppenheimer didn't get the Nobel for his black hole paper decades before some else got it for a rediscovery.
Re: Ask HN: Which recent research paper blew your mind?
#144"Overview of SHARD: A System for Highly Available Replicated Data" it's the first paper to introduce the concept of database sharding. It was published in 1988 by the Computer Corporation of America. It is referenced hundreds of times in many classic papers. But, here's the thing. It doesn't exist . Everyone cites Sarin, DeWitt & Rosenb[e|u]rg's paper but none have ever seen it. I've emailed dozens of academics, libr…
Going through the bibliography of other people's papers and theses, looking for papers that you better cite "for good luck", or because "you gotta cite that one" is a classic PhD student behavior (I've done it) and it's not terribly surprising that something like this can happen. In fact I'd expect it to be much more widespread...
Sometimes it's just an attempt to do the due diligence of citing the primary reference rather than the reference that cited the reference that cited the primary reference. I experienced this recently with a very widely cited basic fact in a very hard to come by technical report from the early 80s. I'd have bet dollars to cents it did in fact contain the statement everyone claimed it did.. however, just in case I made an inter-library loan request to actually read those couple sentences.
Re: Ask HN: Which recent research paper blew your mind?
#145Earlier quoted context omitted.
Surely something about the technique would make it impractical for enriching uranium though... you're not going to be able to produce kilograms of 20% U235 with this are you?
Not with off-the-shelf biological centrifuges, no. I don't know enough about centrifuge design to judge how easy/hard it is to build a 60,000 G centrifuge that can handle liters instead of milliliters of liquid per cavity. The other even more exotic possibility is to use something like this to enrich ordinary reactor grade plutonium to weapons grade plutonium-239. The amount of mass to process with plutonium is order…
Could someone knowledgeable describe the differences between this new method and “traditional” enrichment of nuclear isotopes? If I’m not mistaken the latter also uses very high RPM rotation, and I believe I have read that the hardest part around managing it is the needed precision/balance of rotating something heavy at such high speed — a tiny imbalance can cause a wobble and make the whole thing break.
What changes?
Re: Ask HN: Which recent research paper blew your mind?
#146Earlier quoted context omitted.
Oh nice thanks! I am slooooowwly working on my own language that is planned to use work from both you and the Effekt team, so i may send you all an email in the future when i start having questions.
Sure, it would be great to chat! If you are interested in contributing to Koka at some point, there might also be opportunities there :)
Honestly probably not. I mean never say never.
But I have really limited free time and only work on a language for really specific reasons. Understand, nothing else seemed to try to solve the problem i have at hand.
That said, if the offer is a solid long term employment at good salary, i am open to negotiations. My email is on my blog and my profile iirc
Re: Ask HN: Which recent research paper blew your mind?
#147Earlier quoted context omitted.
Going through the bibliography of other people's papers and theses, looking for papers that you better cite "for good luck", or because "you gotta cite that one" is a classic PhD student behavior (I've done it) and it's not terribly surprising that something like this can happen. In fact I'd expect it to be much more widespread...
Funny, in my high school literature class I clearly remember being chastised for having sources in my works cited; but not warranting their inclusion with an actual reference in the work. It's kind of wild that LLMs and other models/sequences will be able to quickly suss out which papers have high levels of referential integrity.
LaTeX will only include the sources in the bibliography if they are used at least once in the document.
Re: Ask HN: Which recent research paper blew your mind?
#148Placebo Effect Grows in U.S., Thwarting Development of Painkillers https://www.scientificamerican.com/article/placebo-effect-gr... The most interesting thing is that "placebo responses are rising only in the United States."
I wonder if there is any correlation with either political or socio-political volatility, e.g., bread and circuses, conspiracy theories, cults, political dysfunction, and an inability to double-blind test painkillers.
Sounds plausible. Would be interesting if we found a way to measure it on a large scale and track it over time. A National Beliefe Index of sorts.
Re: Ask HN: Which recent research paper blew your mind?
#149I suppose recent is subjective, but this one totally blew my mind. It's a 260 year old academic publication, which is older than the USA:
Mr. Bayes and Mr. Price. “An Essay towards Solving a Problem in the Doctrine of Chances. By the Late Rev. Mr. Bayes, F. R. S. Communicated by Mr. Price, in a Letter to John Canton, A. M. F. R. S.” In: Philosophical Transactions of the Royal Society of London 53.0 (1763), pp. 370–418. DOI: 10.1098/rstl.1763.0053. URL: https://doi.org/10.1098/rstl.1763.0053
Re: Ask HN: Which recent research paper blew your mind?
#150Earlier quoted context omitted.
This seems like an early paper and agreed with the consternation. The paper, in a modern context and based solely on the abstract and having been in the community, is chipping at the "uninteresting" part of the problem. Around that time, program synthesis started switching to SMT (satisfiability modulo theory) methods, meaning basically a super powerful & general SAT solver for the broad search ("write a wild python…
The paper claims that there is no neural / deep learning based solver that performs well on regular expression inference. Calls to Coq and Z3 will be very slow and not competitive with GPU compute.
Ironically, one of its few comparisons was FlashFill / Excel.. which had been going the way of NNs and now LLMs for years now. The paper largely skips measured and non-superficial comparisons, so I wouldn't draw comparative conclusions based on their writeup.
RE Coq/Z3 vs GPUs, I think about them as different kinds of things, and thus wouldn't compare directly. A solver like z3 encapsulates search tricks for tasks where brute force approaches would effectively run forever. Their tricks are symptomatically necessary -- effective GPU acceleration requires finding a clean data parallel formulation, and if a more naive algorithm is picked for that GPU friendliness, but at the expense of inflating the workload asymptotically, the GPU implementation would have high utilization, and high speedups vs a CPU version... but would still be (much) worse than Z3. Instead of comparing them as inherently different approaches, the question is if the practically relevant core of Z3 can be GPU accelerated while preserving its overall optimization strategies.
Which takes us back to the broader point of this thread of GPUs for synthesis... LLMs are successfully GPU accelerating program synthesis of more than just regex. It's amazing that half of github code now uses them! Closer to this paper, folks are making inroads on GPU accelerating SMT solvers that the more powerful (classical) program synthesizers use, e.g., https://github.com/muhos/ParaFROST .