silveraxe93

Every time someone disagrees with laws passed by a democracy, this argument comes back.

Did you ever speak with someone out of tech about internet spying? I did, no normie gives a shit. This is 100% the will of the people. Just take the L for what it is and accept that this is democracy working as intended.

> Kagi isn't good just because it is good, it is good because it's small enough that nobody really optimizes for ranking high on Kagi.

While I think that's like 50% of the story, it's not the only factor. For me what's amazing is being able to up/downrank results.

By shifting responsibility down into users to curate their own algorithm Kagi is more robust to bad actors generating SEO spam.

This works out of the box in VSCode?

Just open a .py file, then select the snippet of code you want to run and cmd+enter

It will open a new REPL for you (using your selected interpreter) the first time, and after that all commands are run in that same one.

Real life with worse graphics +

- Ability to reload when you fail

- You can choose your gender

- Actually, be anything. A dwarf, elf, dog, tall, short, muscled, etc...

- Novel physics (magic)

- Sooooo much more

WWI was also arguably when soldiers became helpless. Being deployed meant huddling in a trench waiting for an artillery shell to strike you out of the blue.

Nope. In order to minimise total loss it needs to be able to generate the whole spectrum of IQ.

All ~models~ measures are wrong, some are useful.

I think this result is really cool, and is another way to measure progress in AI capabilities. I don't think it says much about the absolute position of how "smart" AIs are, but it definitely has value in showing how far it's progressing.

I don't get if this is sarcasm or not... Just in case it isn't:

This is _exactly_ the argument for Musk's case. The lawyer was hired (and managed) to convince the court this argument is wrong.

Call it Franz?

That way the reference is less on the nose and also has the advantage of being a name for the chatbot.

The usual argument is:

We test developer skill by giving them leetcode problems, but leetcode while requiring programming skill is nothing like a real programmer's job.

Eh, I wouldn't call it an issue with Bayesian math itself. Just... it's an issue with reality.

At least with bayesian math you can make your priors explicit, so we the reader at least have the possibility of disagreeing with it.

If you really had no idea you can say it's 50/50 chance and plug the numbers.

That's a terrible plot. I have no idea what the x-axis or the circle areas mean.

They also compare to RLHFed GPT-4, which reduces capabilities, while their model seems to be pre-RLHF. So I'd expect those numbers to be a bit inflated compared to public release.

That got solved using Deep-Q learning. I think David Silver did loads of work on it?

Basically, instead of computing every state value like in normal Q learning. You use a Neural Network to estimate the best value.

Yeah. GPT-3.5 is useless, 4 is just on the margin of being better than using my old workflow.

Definitely worth the money. (Never run into the limit)

I installed this extension, it made me hate Twitter less by not having to use it.

https://addons.mozilla.org/en-US/firefox/addon/nitter-redire...

Ah nice. Thanks!

Could you point me to the part where it says it depends on supercomputer output?

I didn't read the paper but the linked post seems to say otherwise? It mentions it used the supercomputer output to impute data during training. But for prediction it just needs:

> For inputs, GraphCast requires just two sets of data: the state of the weather 6 hours ago, and the current state of the weather. The model then predicts the weather 6 hours in the future. This process can then be rolled forward in 6-hour increments to provide state-of-the-art forecasts up to 10 days in advance.

Right, so as a solution to them having: too much power over our lives, being unaccountable and incompetent. Is:

Giving the backing of the state over their actions. Move from being accountable to government to _being_ the government. And the competency of giant public bureaucracies!

Is it really ironic if every time he touches AI it ends up causing the opposite of what he tried to do?

I didn't see anything in the study (that I also skimmed tbh) that is clearly wrong, which makes me dismiss the results.

It's just because I don't think this is evidence enough to change my priors. i.e. My gut tells me this is wrong and I don't buy it.

The sample is 267 people, tiny enough that I'd expect an analysis to be done with a linear model with few features. They used 14 features, but had a pipeline to select "model (out of 3), feature set, and threshold for prediction".

There's _so_ many degrees of freedom there that _by default_ I assume there's leakage.

I'd love to see this paper replicated! It would be amazing if this were true. But if I had to bet on this, I wouldn't give more than 20% chance of being true.

I _highly_ doubt this would replicate. I bet they just leaked test data somehow and won't generalise.

Completely agree with you. I had amazing friends when I went to the office. I 100% prefer remote working and am not coming back, but I have to recognise that it's not better at absolutely everything.

I'm seeing a absurd amount of Arguments as Soldiers[1] in this space. No one wants to have a conversation. It's always straight into conflict, underlying intentions, etc. No regards for truth seeking.

[1] - https://www.lesswrong.com/tag/arguments-as-soldiers

He covers that in the post. If you donate your kidney, you can nominate up to 5 people to be first in line if they (or you) ever need a transplant.

Given compatibility concerns, it's probably safer for your family member for you to donate than not.

I actually read that one before! Kahneman is overconfident, as we all were before the replication crisis.

While most of the psychology field is crumbling and filled with bullshit (even though most people didn't catch on yet). There's still _some_ truth lying in there.

It might look bad in isolation, but Kahneman's work is one of the few that actually holds up to scrutiny.

This post by Scott Alexander is pretty good in summarising the few good bits left. https://www.astralcodexten.com/p/heres-why-automaticity-is-r...

---

But regarding the initial point. The goal is not to fully emulate human thinking. It might sound wishy-washy, but there's _obviously_ _some_ truth to the system 1/2 model. It's not perfect! But I think it's useful.

We as humans do most decisions without thinking (hard), but we have a way to _switch_ into a more reliable but expensive mode. We see some evidence that artificially inducing LLMs to 'think' more improves their output.

So it stands to reason that adding this capability to a model would make it better. And what better way to do it than swallowing the bitter pill and have that decision be made by the model itself, on a case-by-case basis by learning it from data.

Damn that's cool. Can we implement a version of System 1 / System 2 [1] thinking with this?

From the future work section:

> better determining the number of <pause> tokens (perhaps using model confidence)

If the number of pause are learned, using model confidence + regularisation (so that the model doesn't always use the maximum number of pauses), then we effectively have the System 1/2 switch. If it's a task the model has seen tons of times before, it just goes with the first inference pass. If it's low confidence, then it keeps expending more inference passes until it reaches an acceptable confidence threshold.

- [1] https://en.wikipedia.org/wiki/Thinking,_Fast_and_Slow

I've had many conversations exactly like yours, literally. To reasonable people being convinced that's normal, and me trying to explain how it's probably a coincidence or ad targeting.

That's what _you_ and _I_ think.

Most conversations I have with non-techy people, they end up saying "Yes" to both.

If by lowest hanging fruit, you mean in the original sense of: Equally as tasty as higher hanging fruit, but easy to pick because it hasn't been relentlessly exploited yet.

Then yeah, I agree.