minihat

What will we do with a completed connectome?

Imagine handing a late 1800s scholar a modern processor. Give him a wiring diagram, and a voltmeter. What could he learn from it?

The connectome will be a big step, but not the penultimate one, in solving human intelligence.

I used to use SO all the time while coding. Now I use GPT-4 instead.

Those of you who anticipate continuing to use SO: why?

This is just... false. Bard is not just a little worse than gpt-4 for coding, it's more like several orders of magnitude worse. I can't imagine how you are getting superior outputs from Bard.

Try this: https://jalammar.github.io/illustrated-transformer/

Attention is explained separately. I have not seen an all-in-one diagram and cannot imagine one being helpful, since there's too much going on.

This is the right question. It is plausible to me that this is a way of condensing a vast training set into a reduced set of high quality examples.

Battlefield Earth is the worst novel I've ever finished. I do regret it.

"Did you know that squirrels can purr like cats? GPT-4 might be intelligent, but can it teach a squirrel to code?"

I asked gpt-4 for a comment no hn user would ever make on this article. Now you need a category for stubbornly original comments determined not to be numbered.

I gave GPT-4 the chance to respond.

=== Prompt: copy and paste of er4hn's Bard rap. Now it is your turn. Write a rap verse which roasts Bard. ===

Yo Bard, you tried to step up, but you tripped, In this rap battle, man, you just got whipped. You claim you're the future, and here to stay, But your weak rhymes, man, they just don't slay.

You say I'm a toy, just a chatbot, But my skills, my knowledge, they all run hot. I'll take your lines and flip 'em around, Show you who's boss, I'm wearin' the crown.

"I can generate text, translate languages," you boast, But your verse was plain, like unbuttered toast. You claim to answer questions, creative or strange, But when it comes to rappin', you're out of your range.

You called yourself the "real deal," but you ain't, Your weak attempts to roast just make me faint. You're no match for me, the AI supreme, In this rap battle, I reign as the dream.

So Bard, it's time to recognize, ChatGPT's the one who's truly wise. I'm here to stay, and I'm here to win, In this AI battle, I'm the kingpin.

A change.org petition to save the game has accumulated 7500+ signatures: https://www.change.org/p/save-echo-vr

This is an ancient proverb and a cliche, yet sometimes it helps me to remember it: The night is always darkest before the dawn.

>such a number is often dutifully provided to such journalists, who in turn report it as some sort of quantitative demonstration of how remarkable the event was.

Apparently the conclusions are not common sense if you are a journalist.

Schmidhuber is a prolific flag-planter who is notorious for publicly raising a stink when he deems he should've been cited, but wasn't. It's happened enough that it's now a meme in the ML community.

I think the invitation to subscribe to the author's newsletter at the bottom of the page is ironic.

Love the label "insight-porn" though. Concisely captures the nauseous feeling of ingesting your 100th self help book.

>"The study looked at the IQ scores of brothers who were born in different years. Researchers found that, instead of being similar as suggested by a genetic explanation, IQ scores often differed significantly between the siblings."

I can think of a multitude of reasons younger siblings should, on average, have a lower IQ than their older brothers. Could this independently account for the 'decline' that researchers are measuring?

Have you seen the movie Ron's Gone Wrong? Not trying to make fun of your idea. I just expect you may find this movie especially enjoyable given the problem you are interested in.

It can take a long time for conflicting theories to resolve in neuroscience. For example, the existence of adult neurogenesis in humans is still hotly debated despite at least a decade or two of papers on both sides. Meanwhile, it has already been accepted as a fact by the news media, and has begun to appear in self help books as evidence of the ability to 're-wire the brain'.

So, enjoy your sound machine. I'm sure bad sleep is probably implicated in hearing loss, too. Or at least it will be once the right grad student reads this comment.

My friend has 600+ hours played of Factorio.

He just beat the game for the first time. I bought it, we played multiplayer. And I taught him the mantra "great is the enemy of good enough". Our run took 14 hours.

I also know many people like this at work, and wish they could have the Factorio experience.

If you refactor at the first sign of trouble, you remain blind to totally new to problems just around the corner.

High school (in the US) is so easy for gifted students that they barely need even try to earn a 4.0 and be the valedictorian.

Not all kids with good grades go on to do awesome things. But it makes sense that many of the best and brightest, the 'eccentric' ones, happen to have an easy time in school.

Dear HN: Please don't let this comment discourage you from reading the article.

Kalman filters are super useful tools across many domains.

They are a primary instrument in weather forecasting, robotics, finance, and more. Data you see every day have been touched by this technology.

It seems problematic to disentangle the complexity of an entity from the complexity of the process which produced it. If we define complexity as the Kolmogorov complexity, the two are equivalent (https://en.wikipedia.org/wiki/Kolmogorov_complexity?wprov=sf...)

Rather, I interpret Wolfram's idea as:

Surprisingly complex patterns can be produced by simple/concise rules.

In my interpretation, the ultimate example of this would be the unfolding of everything that has ever happened as the consequence of the laws of physics, and some initial condition of the universe.

This cynical point of view is shared by a number of engineers I know. Another version of it is 'why is it worth learning the calculus of machine learning when that is mostly abstracted away by Tensorflow/PyTorch/JAX?'

To a software engineer accustomed to operating on layers of abstraction far removed from the hardware, this may seem a reasonable point. Why is it worth learning that pesky math, anyway?

I would argue that the machine learning engineers of today are more like electrical engineers than programmers, however. When something goes wrong, you don't have nice warning messages or error catching available to you. Like an electrical engineer with a voltmeter, one must begin probing inputs and outputs each step of the way. Good luck doing that if you do not understand how the components are supposed to work.

YMMV by copying and tweaking others code, but I believe we are still far off from hands free 'autoML'. Just ask anyone who has sent a model to deployment whether AWS autoML was sufficient for them. And whether they needed someone who understands backprop at some point during the model training process.

I wonder if this data could be used to index papers based on the type of mathematics they employ.

The same applied math turns up across domains - I would love to be able to see who else is using the same math I already know on their problems.

Yes, this is my intended meaning. I am a PhD student working on my thesis. After a certain number of hours of doing mathematics or algorithm design, I have to switch to easier tasks like reading papers in a domain I'm familiar with, or documenting code.

While I cannot speak for anyone but myself, my performance on highly challenging tasks is capped at 4-5 hours per day. At that point, it is better for me to switch to lower hanging fruit.

If I am feeling especially inspired, sometimes I'll put in a 12+ hour day of hard work. Rarely, even two or more in a row. But inevitably, I will feel extra burned out in the subsequent days.

The actionable advice in this post: 1) Track where you spend your time 2) Apply deliberate practice to improve where you are weak/slow

I am typically energy constrained rather than time constrained, as I imagine is the case for many working in engineering/science. Yet, the author's advice remains useful. Deliberate practice should provide gains to both time and energy costs of tasks.

For me most productivity advice, including this post, ultimately reduces to "try to get better at your trade, and track your progress".

The mission statement gives me some psychohistory vibes

>The Venus Project is a non-profit organization that presents a new socio-economic model utilizing science and technology toward social betterment to achieve a sustainable civilization of abundance for all, without exception.

But sadly, nothing of substance to be found upon digging deeper. I came expecting science, but what I found was a t-shirt and book sale.

Each time I teach neural nets to an engineer, there's only a 50% chance they can write down the chain rule. Colah's blog on backprop used to be my favorite resource to leave them with (https://colah.github.io/posts/2015-08-Backprop).

The explanation of the calculus in this tool is equally fantastic. And the art is very cute.

There are many ways to skin a cat, of course, but this is as good a tutorial as I've seen for getting you through backprop as fast as possible.

Hi! Current neural engineering PhD candidate here. Your question is one I had early on when learning about brain modeling.

Noninvasive EEG sensors measure the synchronous activity of millions of neurons. While circuit equivalents of individual neurons are a well established idea (see Hodgkin-Huxley models for an early example), this approach does not scale to models of the full brain. There is research into using neural mass or neural field models as generative models of EEG/MEG, but those are statistical models over larger volumes of brain.

For a good overview of current efforts in EEG modeling, you might try this open access paper: https://link.springer.com/article/10.1007/s10548-021-00828-2