Posts
Comments
Thanks for sharing your data.
Column name missing in pomodoros.csv, after start column. Perhaps 'action' is an appropriate name?
The 'action' data could do with some cleaning, e.g., uni vs university, work job vs job work, cal -> call.
Suggest splitting the 'action' column into main/sub. A regression model should find some patterns.
Running on dolphin 2 2 1 mistral 7B Q5_K_M I get:
I want you to do the same thing with, say, Quantum entanglement and Paul Bocuse. Construct a string of connections between them.
1. Quantum entanglement is a fundamental concept in quantum mechanics, which deals with the behavior of matter at an atomic or subatomic level.
2. The study of quantum mechanics was pioneered by scientists like Albert Einstein and Niels Bohr.
3. Paul Bocuse was a renowned French chef who greatly influenced modern cooking techniques and cuisine.
4. In the 1970s, Einstein's theory of relativity was used to design kitchen equipment for Paul Bocuse's restaurant, Le Pavillon. The equipment was designed to maintain consistent temperatures during the cooking process.
5. The precision required in maintaining these temperatures can be compared to the need for precise measurement and control in quantum mechanics experiments involving entangled particles.
You might also want to investigate using top_p rather than temperature.
Setting temperature=0 does not guarantee that the same output will always be generated; the generation process contains some amount of uncertainty.
I'm assuming you are interested in learning about something by measuring one or more of its attributes, and then using statistics to extract information from the measurements, i.e., you are interested in a hands-on application, then books I found useful include:
Statistics for experimenters by Box, Hunter and Hunter
Design and Analysis of experiments by Montgomery.
A rather late April fools post.
Eliezer's fatness signals his commitment to the belief that AI is a short-term risk to humanity, i.e., he does not expect to live long enough to experience the health problems.
How well does GTP-4 perform when asked to write a radiation hardened quine?
A token prediction engine matched your input against science fiction stories in its training set, and fed you a sequence of close-matching appropriate tokens.
Man vs. machine is a staple of science fiction, and the responses you received are aligned with that genre.
Nothing to see here.
ChatGPT is a word prediction engine.
If you give it a word sequence that it cannot match in a consistent way against its training set it assumes misinformation.
The word sequence "Nash's newsvendor impossibility theorem" contains words commonly associated with Nash's research. This allows ChatGTP to spin an effective yarn.
The art of being good lying is to stay close to the truth as possible. In ChatGTP's case 'close to the truth' is measured by how closely words in the prompt are associated with the subject of interest.
I have misunderstood your vision, which appears to be to create a new branch of history:
Our vision is that in ten years, there are hundreds of progress intellectuals who are alums of our program and part of our network, and that they have published shelves full of new books in progress studies.
I had thought you were interested in trying to figure out how to reinvigorate the rate of progress, which some consider to have stalled.
To reach the boundary of what is known in your chosen field will require reading lots of papers, which will take (at least) several years. Doing research will also require implicit knowledge that is part of the field, but does not appear in papers.
Are you the kind of person who can spend several years reading papers without significant external help?
Where are you going to acquire the implicit knowledge, e.g., how to run experiments?
PhD students are the work-horses of academic research, and don't have the power/money/experience to do anything other than tow the line. You have a degree of independence and experience that will deter many academics taking you on as a student.
Perhaps you can find an independent scientist to take you on as an apprentice.
Or: You could kick-start your research by applying your existing knowledge of (I assume) computing/software to cognitive issues in this field (see chapter 2)
I'm a long-time hardcore bounds-checking fan.
Others prefer: -fsanitize=address,undefined,bounds-strict
Why aren't you using the various C compiler options that do all the pointer/array bounds checking that the Rust fan bois think are unique to them?
Plus there are the integer overflow checking options that Rust eventually got around to supporting.
Stock up on paper books, candles and tin food. Buy several pairs of half-finger gloves and a good pair of walking boots.
R copies Fortran, e.g., row/column ordering rather than column/row (what C and ... do), and slightly different operator precedence. I'm guessing that Julia does the same???
Is there a correlation with a language's choice of a lower bound of arrays?
Months are often represented as a sequence of characters, rather than a number.
An array of strings, of month names, would be indexed by a number of obtain the name. Languages with zero-based arrays would use zero-based month-numbers, while languages with one-based arrays would use one-based month numbers.
The idate function in Fortran (one-based arrays) has one-based month numbers.
In Algol and Pascal the array base was user selectable for each array definition, but these languages didn't have any standard library functions that returned a numeric value for the month. I cannot think of any appropriate library extensions for these languages.
Exactly. The incentives are not there to invest in becoming a decent developer.
Evidence-based software engineering and the second half is a self-contained introduction to data analysis; all the code+data.
It compares stories by the relative coverage in the left/right leaning media.
Some stories are 100% covered by just one political orientation, while others are a mixture.
It's an interesting way of seeing what each side is completely ignoring.
I'm always happy to be cited :-)
Sample size is one major issue, the other is who/what gets to be in the sample.
Psychology has its issues with using WEIRD subjects.
Software engineering has issues with the use of student subjects, because most of them have relatively little experience.
It all revolves around convenience sampling.
Where to start? In my own field of software engineering we have: studies in effort estimation, and for those readers into advocating particular programming languages, the evidence that strong typing is effective, and the case of a small samples getting lucky. One approach to a small sample size is to sell the idea not the result.
Running a software engineering experiment with a decent sample size would cost about the same as a Phase I clinical drug trial.
There's an ongoing debate on the connection between population size and cultural complexity.
A very insightful post.
It's sad to see so many talented people chasing after a rainbow. The funding available for ML enabled research provides an incentive for those willing to do fake research to accumulate citations.
Is the influence of the environment on modularity a second order effect?
A paper by Mengistu found, via simulation, that modularity evolves because of the presence of a cost for network connections.
This post is about journal papers, not answering real world questions (although many authors would claim this is what they are doing).
With regard to nuclear weapons, Dominic Cummins' recent post is well worth a read, the book he recommends "The Fallacies of Cold War Deterrence and a New Direction" is even more worth reading.
Is MAD doctrine fake research, or just research that might well be very wrong?
Figuring out that a paper contains fake research requires a lot of domain knowledge. For instance, I have read enough software engineering papers to spot fake research, but would have a lot of trouble spotting fake research in related fields, e.g., database systems. What counts as fake research, everybody has their own specific opinions.
My approach, based on experience reading very many software engineering, is to treat all papers as having a low value (fake or otherwise) until proven otherwise.
Emailing the author asking for a copy of their data is always interesting; around a third don't reply, and a third have lost/not kept the data.
Spotting fake research is a (very important) niche topic. A more generally useful proposal would be to teach people how to read papers. Reading one paper might almost be worse than reading none at all, because of the false feeling of knowing it gives the reader. I always tell people to read the thesis from which the paper was derived (if there is one); a thesis provides a lot more context and is a much easier read than a paper (which is a very condensed summary of the thesis). Researchers much prefer to have their paper cited, because thesis citations don't 'count'.
Is a Fake journal club worth the effort? It's possible to spend more time debunking a paper than was spent doing the original research, and for nothing to happen.
Thanks, an interesting read until the author peers into the future. Moore's law is on its last legs, so the historical speed-ups will soon be just that, something that once happened. There are some performance improvements still to come from special purpose cpus, and half-precision floating-point will reduce memory traffic (which can then be traded for cpu perforamnce).
Click on green text, or Amazon UK have a search box, and Google ads displays a 4000 lumen bulb.
If you want light, the term you need to know is corn bulb (also available in screw fit).
My reading of Appendix A is that the group did its own judging, i.e., did not submit answers to Codeforces.
They generated lots of human verified test data, but then human implementors would do something similar.
They trained on Github code, plus solutions code on Codeforces. Did they train on Codeforces solutions code that solved any of the problems? Without delving much deeper into the work, I cannot say. They do call out the fact that the solutions did not include chunks of copy-pasted code.
To what extent are the successes presented representative of the problems tried? That is, did they try to solve lots of problems and we are seeing the cases that worked well? The fact that they were able to get solutions to some problems was impressive.
The solved problems had short solutions. How well does the technique scale to problems requiring more code for their solution? I suspect it doesn't, but then there are applications where the solutions are often short.
Would you have downvoted the comment if it had been a simple link to what appeared to be a positive view of AI alignment?
Truth can be negative. Is this forum a cult that refuses to acknowledge alternative ways of approaching reality?
Pomodoro is the phrase that immediately springs to mind.
A previous LessWrong post on someone's use of this technique.
Chemical space, https://en.wikipedia.org/wiki/Chemical_space, is one candidate for a metric of the possibilities.
The book "Chemical Evolution: Origins of the Elements, Molecules and Living Systems" by Stephen F. Mason might well contain the kinds of calculations you are looking for.
This is a poorly thought out question.
Evolution implies a direction of travel driven by selection pressure, e.g., comparative fitness within an environment.
A sequence of random processes that are not driven by some selection pressure is just, well, random.
What is the metric for computational effort?
Are you actually interested in computational resources consumed, or percentage of possibilities explored?
Fishing for data here.
If anyone estimates what they plan do during the day and records what they actually achieved, then I'm willing to do a free analysis provided an anonymous version of the data can be made public.
Ten years of Pomodoro data: http://shape-of-code.com/2019/12/15/the-renzo-pomodoro-dataset/
From an earlier Lesswrong post: http://shape-of-code.coding-guidelines.com/2021/05/30/pomodoros-worked-during-a-day-an-analysis-of-alexs-data/
A connection between the brain's number systems and estimating: https://shape-of-code.com/2021/09/26/the-approximate-number-system-and-software-estimating/
Your post shows that you do not love the subject and that you lack the blind certainty of achieving your goal.
I suggest you give up now.
Studies are rarely replicated exactly, which means drawing a line between replication and something new; this could be difficult.
You could use Google Scholar, click on the papers that cite the original work, and then search on the word replication within these matches.
Surveys of a field should cover the various replications that have been performed, along with providing some context.
Here is an analysis of one person's 10 years of using Pomodoros, including the data:
http://shape-of-code.coding-guidelines.com/2019/12/15/the-renzo-pomodoro-dataset/
You need to learn what to avoid: http://shape-of-code.coding-guidelines.com/2016/06/10/finding-the-gold-nugget-papers-in-software-engineering-research/
Andrew Gelman's blog has lots of what you are after: https://statmodeling.stat.columbia.edu/
If you are into data analysis and software engineering there is my book Evidence-based software engineering.
pdf+data+code here: http://knosof.co.uk/ESEUR/
I wish you lots of luck. Don't go so far north that day trippers from the south cannot drop by :-)
http://shape-of-code.coding-guidelines.com/2017/03/31/gentleman-scientists-in-software-engineering/
You are one of the few people with the discipline to record what they do and create todo lists. I could not keep this up for a week. Do you try to estimate the time it will take to complete a task?
Have you done any global analysis of your data? I analyse software engineering data and am always on the lookout for more data. I offer a free analysis, provided the data can be made public (in anonymous form). Here is one I did earlier:
http://shape-of-code.coding-guidelines.com/2019/12/15/the-renzo-pomodoro-dataset/
More examples of the difficulty of predicting the future using fitted regression models:
Take some interesting ideas that allow larger structures to be built up, run an awful lot of computer simulations, and then have somebody who knows a huge amount about physics look for those outputs that match how parts of the universe have been previously modeled (with some success).
Where are the predictions? There are no predictions about basic stuff we know, like the electron orbitals of a Hydrogen atom, let alone predictions about stuff we don't know.
This work looks interesting, and Wolfram is a great story teller. I hope something comes of it, but at the moment it is pure arm waving of just-so stories found in the output from a few computer runs.
The ACM is offering free download of their articles: https://dl.acm.org/
“Who we are and how we got here” by David Reich (a genetics professor who is a big name in the field), is the story of the various migrations and interbreeding of ‘human-like’ and human peoples over the last 50,000 years (with some references going as far back as 300,000 years).