mvcalder

There's a story sharing the front page right now about scaling laws, so it's not an unreasonable assumption.

https://news.ycombinator.com/item?id=48689744

Wait, my job is not cushy. I think hard all day long, I endure levels of frustration that would cripple most, and I do it because I have no choice, I must build the thing I see or be tormented by its possibility. Cushy? Right.

I really enjoyed the article and want to both praise and encourage the author.

You get it. It’s about value. Keep you eye on that north star and you won’t go wrong.

Whose value? How do I value? Can I reconcile disparite value? Yep, those are the right questions.

For me, I read this and want to give a shout out to Zen and the Art of Motorcycle Maintenance, but that’s just me.

I enjoyed the read, thank you.

It’s not “freedom of speech I agree with”, no one needs protection to speak agreeably.

My Massachusetts home has both radiant floor heating (water pipes in floor) and baseboard water pipe heating (separate part of the house). My son’s New York home has radiators driven by hot water. I can’t recall a home where heating was vented air and not circulating hot water in North East USA.

Nope, all human all the time.

“They say sixty-five percent of all statistics Are made up right there on the spot”

https://www.youtube.com/watch?v=IUK6zjtUj00

And there’s only one photon, vibrating like mad, singularly unable to contemplate its own magnificence.

It was an odd sight at that hour of the day. I wanted pictures as something to talk about with friends. I really wish I had been astute enough to take video.

I read several comments referring to: idiots, morons, delusional, and hysteria.

At the risk of being labeled a kook or an idiot, I photographed drones flying over my suburb of Boston neighborhood a few weeks ago. This was about 6am, definitely drones not regular aircraft. I assumed it was something flying out of Hanscom or the city mapping streets. And yes I took photos not video, sorry.

https://photos.app.goo.gl/Lwfn134LqdEp6xbG9

For the experts in the house: would it be possible for this to wrap around, so the highest and lowest level join? A torus of GoLs? Can such a thing, assuming it exists, have a finite number of layers? Just curious, this is amazing.

Does anyone know what the copyright status of LLM generated content is? That is, if I feed a NYT article into GPT4 and say, summarize this article, and then publish that summary, is there argument or precedent that says that is or is not copyright infringement? Asking for a friend.

I wrote up this analysis of cache one-hits after reading the FIFO is all you need paper.

https://www.polyscale.ai/blog/one-hit-expectation

Disclosure, I work at PolyScale.

Prepared statements work fine. Views can be tricky.

The statistical invalidation will in general disable caching on views since their results will change without any corresponding writes (because we only associate writes with the backing tables of the view). Currently our best answer to views is the new stale-while-revalidate flow which revalidates against the db while still serving a hit. With this feature enabled, one can turn off invalidation (if the resulting eventual consistency is acceptable).

More details about how invalidation works here:

https://www.polyscale.ai/blog/approaching-cache-invalidation...

Whenever this topic comes up I like to provide a citation to some work I've done:

https://towardsdatascience.com/gradient-kernel-regression-e4...

Not out of vanity (ok, a little) but because I think the idea has importance that has not been fully explored. The article's Bayesian perspective may be the whole story but somehow I don't think so. Unlike the article's author, my work left me feeling model architecture was the most important thing (behind training data) whereas they seem to feel it is ancillary.

If you, like me, often find yourself covered in pine sap. The product: Shout, for removing laundry stains will make quick work of removing sap from skin. Maybe the answer to your chemistry question is in there.

One of my favorite things to do in museums is to use Google Translate on artifacts. I find it so exciting to point my phone at a 500 year old vase and read what it says. Here are a couple examples from a trip I tool to the British Museum, it handled the Arabic and Chinese pretty well but the Rosetta Stone's Greek stumped it.

[1] Arabic: https://photos.app.goo.gl/jQw3QCdC5PrWQGkr9 [2] Chinese: https://photos.app.goo.gl/dXhsH7Xq6zpFS6Eq8 [3] Greek: https://photos.app.goo.gl/DBGWCDd8EhzuB2y67

I can't speak to the others, but pgvector indices can "break" hybrid queries. For example, if you select using a where clause specifying metadata (where genre = jazz) and order by distance from a vector (embedding of sound clip); if the index doesn't have a lot (or any) vectors in the sphere of the query vector that also match the metadata it can return no results. I discuss this in a blog post here [1].

[1] https://www.polyscale.ai/blog/pgvector-bigger-boat/

Blog post about using PolyScale with pgvector.

Bruh

Professors Boyd and Vandenberghe really broke ground with this text. Prior to this, optimization algorithms and methods were very much locked up behind a metaphorical paywall: difficult to access literature with very high barriers to entry, and strictly commercial software offerings. They brought optimization to the masses and should be celebrated for it.

If you like this sort of thing there’s a great book: Methods and Programs for Mathematical Functions by Stephen Moshier. It covers the computation of all sorts of special functions. The code is available in the cephes library but the book may be out of print.

No punch cards, but I did write the Windows version of ITSM (software accompanying the Brockwell and Davis Time Series book). The while thing was written in Fortran, the event loop and calls to the Windows API for UI and graphics. It's still the best damn visualization of a series + periodogram I've ever come across (if I do say so myself).

I trained the raccoons that visit my house at night. I started them out getting a peanuts from a water bottle. Then I tied the bottle to a rope. Then kept raising the bottle higher. At that point, I built an automated feeder system using a linear actuator activated by pulling the rope with the bottle attached. It had LEDs that were green / red to show when the feeder would / wouldn't dispense peanuts. It was all driven by an ESP32, it even had a web page on our LAN reporting how many correct / incorrect pulls were done. Over the coarse of a few nights they figured it out. Raccoons are so cool.

It will be interesting to see what insights such efforts spawn. For the most part LLMs specifically, and deep networks more generally, are still black boxes. If we don't understand (at a deep level) how they work, getting them to "conform to some semblance of reality" feels like a hard problem. Maybe just as hard as language understanding generally.

You might like tslab. It allows you to have the full notebook experience with either JavaScript or Typescript. My day to day is data analysis. JS/TS runs circles around pandas and you aren’t constrained to vectorized operations. If there were a suitable replacement for matplotlib I would leave python behind altogether.

https://github.com/yunabe/tslab

One might argue one approach is superior over the other, I'd argue they are more like duals of one another. The PolyScale approach analyzes the queries and identifies the semantic and the statistical relationships between reads and writes. The Noria approach forgoes analyzing the queries and instead maintains a materialized view like representation of where the data should-be-at-now.

The PolyScale approach does not maintain / require a separate data representation and so saves space, but on the other hand, precisely identifying the relationship between reads and writes is not possible and so the PolyScale approach must sometimes over-invalidate in the interest of accuracy.

There are scenarios in which show-me-the-data (Noria) beats show-me-the-math (PolyScale), for example, running complex queries against a relatively simple schema. There are also scenarios in which the statistical (PolyScale) approach wins, for example if the queries are relatively simple or if not all writes to the underlying data are visible.

There are additional unique features of PolyScale that set it apart. Full disclosure, I work at PolyScale.

That's an unscientific conclusion. They say the effect (of high variation) is unattributable, so it's unfair to limit it to soft sciences.

This could infect literally any human endeavor that generates measurements, hard sciences included.

Or, as another poster observes, this result itself is affected by high variation and we end up in a Godel-ian loop of "this study is false" results.

H''''': The lottery is run by corrupt officials, who want to cover their tracks. In order to do so they want to have many associates win so no single winner is subject to scrutiny. They need to communicate to this diverse group of conspirators the correct numbers to choose in a simple way, say: multiples of 9, or anti-diagonal of ticket.

Remark 4: The human mind is an amazing hypothesis generating machine. If it also knows about Bayesian statistics, it is capable of accounting for it. Paranoia is a bitch.