Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Probabilistic Artificial Intelligence

arxiv.org

Post reply on HN
Print view View original

91–100 of 100 posts

Re: Probabilistic Artificial Intelligence

#91

Post by hohloma »

energy123 wrote: post #80
vlovich123 wrote: post #65

Earlier quoted context omitted.

Not out of the box I think; I wouldn’t trust any self-assesment like that. With enough compute, you could probably come up with a metric by doing a beam search and using an LLM to evaluate how many of the resultant answers were effectively the same as a proxy for “confidence”.

Similar to bootstrapping a random variable in statistics. Your N estimates (each estimate is derived from a subset of the sample data) give you an estimate of the distribution of the random variable. If the variance of that distribution is small (relative to the magnitude of the point estimate) then you have high confidence that your point estimate is close to the true value. Likewise in your metric, if all answers a…

Confidence != true

Re: Probabilistic Artificial Intelligence

#92

Post by energy123 »

hohloma wrote: post #91
energy123 wrote: post #80

Earlier quoted context omitted.

Similar to bootstrapping a random variable in statistics. Your N estimates (each estimate is derived from a subset of the sample data) give you an estimate of the distribution of the random variable. If the variance of that distribution is small (relative to the magnitude of the point estimate) then you have high confidence that your point estimate is close to the true value. Likewise in your metric, if all answers a…

Confidence != true

That's correct but P(true) might empirically turn out to be some f(confidence)

Re: Probabilistic Artificial Intelligence

#93

Post by _jonas »

svilen_dobrev wrote: post #64

stupid question: can a LLM (i.e neural network) tell me the probability of the answer it just spew? i.e. turn into fuzzy logic? Aaand, can it tell me how much it does believe itself? i.e. what's the probability that above probability is correct? i.e. confidence i.e. intuitionisticaly fuzzy logic? Long time ago at uni we studied these things for a while.. and even made a Prolog interpreter having both F+IF (probabilit…

[dead]

Re: Probabilistic Artificial Intelligence

#94

Post by mnky9800n »

ivansavz wrote: post #50
woolion wrote: post #14

Earlier quoted context omitted.

Aren't you describing Lulu but for the very niche case of arxiv publications that are small books but not published as books? I think you could do it in a weekend with their API.

If anyone is interested in trying this, here is some Python starter code you might find useful: https://github.com/minireference/lulu-api-client?tab=readme-... This worked four years ago when the API was still launched, but there might have been changes since, so no guarantees. Most ArXiv PDFs are probably lulu-printable out of the box, but to make a general solution, one would probably need to do some pre-processing…

also, they require that you have both the interior contents as a pdf and a pdf of the cover, so it will need to either auto-generate or offer the option to have a custom cover art.

Re: Probabilistic Artificial Intelligence

#95

Post by wat10000 »

soulofmischief wrote: post #90
wat10000 wrote: post #87

Earlier quoted context omitted.

The argument was: lots of people do X, so X has obviously inspired discoveries.

Carrots inspire science, too. https://phys.org/news/2015-12-tablet-screen-brought-aid-carr... > It was the Austrian botanist Friedrich Reinitzer and the German physicist Otto Lehmann who discovered liquid crystals in 1888 when they were experimenting with the natural substances found in carrots Carrots have also inspired art, I don't need to find an example of that. That said, if you view that premise as it being a g…

> a critical mass of people have done mind-altering substances and thus it's almost a certainty that discoveries have been made under... their influence

This much is obvious. But that sneaky “and due to” which I cut out does not belong. You’re putting a completely unfounded assumption next to a simple statistical argument and presenting them as equivalent.

Re: Probabilistic Artificial Intelligence

#96

Post by soulofmischief »

wat10000 wrote: post #95
soulofmischief wrote: post #90

Earlier quoted context omitted.

Carrots inspire science, too. https://phys.org/news/2015-12-tablet-screen-brought-aid-carr... > It was the Austrian botanist Friedrich Reinitzer and the German physicist Otto Lehmann who discovered liquid crystals in 1888 when they were experimenting with the natural substances found in carrots Carrots have also inspired art, I don't need to find an example of that. That said, if you view that premise as it being a g…

> a critical mass of people have done mind-altering substances and thus it's almost a certainty that discoveries have been made under... their influence This much is obvious. But that sneaky “and due to” which I cut out does not belong. You’re putting a completely unfounded assumption next to a simple statistical argument and presenting them as equivalent.

It's not sneaky at all, I'd appreciate a more charitable interpretation. It's not unfounded. If a substance alters someone's thoughts, it is then responsible in some part for the thoughts they have while under its influence.

Re: Probabilistic Artificial Intelligence

#97

Post by wat10000 »

soulofmischief wrote: post #96
wat10000 wrote: post #95

Earlier quoted context omitted.

> a critical mass of people have done mind-altering substances and thus it's almost a certainty that discoveries have been made under... their influence This much is obvious. But that sneaky “and due to” which I cut out does not belong. You’re putting a completely unfounded assumption next to a simple statistical argument and presenting them as equivalent.

It's not sneaky at all, I'd appreciate a more charitable interpretation. It's not unfounded. If a substance alters someone's thoughts, it is then responsible in some part for the thoughts they have while under its influence.

It's either unfounded as written (it might be sensible with more support), or it needs a notion of responsibility that's so broad as to be useless.

Everything in your past light cone alters your thoughts to some degree. To be considered responsible for discoveries, the alteration needs to be significant and be in a way that the discovery wouldn't have happened without it. Significance is pretty much a given, there are plenty of drugs that significantly alter thoughts. But "wouldn't have happened without it" is a big assumption that, at the very least, is not supported by the argument I was replying to.

Re: Probabilistic Artificial Intelligence

#98

Post by soulofmischief »

wat10000 wrote: post #97
soulofmischief wrote: post #96

Earlier quoted context omitted.

It's not sneaky at all, I'd appreciate a more charitable interpretation. It's not unfounded. If a substance alters someone's thoughts, it is then responsible in some part for the thoughts they have while under its influence.

It's either unfounded as written (it might be sensible with more support), or it needs a notion of responsibility that's so broad as to be useless. Everything in your past light cone alters your thoughts to some degree. To be considered responsible for discoveries, the alteration needs to be significant and be in a way that the discovery wouldn't have happened without it. Significance is pretty much a given, there ar…

I agree, but it's difficult to make the criteria objective. "I was drunk last night and I realized X..." vs "Last night, a molecule of alcohol hit just the right neuron and fired off a chain of thoughts that led to a discovery"... Aren't both valid?

Re: Probabilistic Artificial Intelligence

#99

Post by devmor »

soulofmischief wrote: post #98
wat10000 wrote: post #97

Earlier quoted context omitted.

It's either unfounded as written (it might be sensible with more support), or it needs a notion of responsibility that's so broad as to be useless. Everything in your past light cone alters your thoughts to some degree. To be considered responsible for discoveries, the alteration needs to be significant and be in a way that the discovery wouldn't have happened without it. Significance is pretty much a given, there ar…

I agree, but it's difficult to make the criteria objective. "I was drunk last night and I realized X..." vs "Last night, a molecule of alcohol hit just the right neuron and fired off a chain of thoughts that led to a discovery"... Aren't both valid?

If there's no evidentiary reason to assume causal relationship, I would say that's a useless distinction, personally.

There's just as much reason to assume that if you didn't drink at all, you'd have thought of it sooner.

Re: Probabilistic Artificial Intelligence

#100

Post by mitthrowaway2 »

thisisauserid wrote: post #8

Gemini 2.0 Experimental 02-05 sees this as "only" 107K tokens. Handy if you want help breaking this down. https://aistudio.google.com 'Laplace Approximation is a "quick and dirty" way to turn a complex probability distribution into a simple Gaussian (bell curve). It works by finding the highest point (mode) and matching the curvature at that point. It's fast and easy, but it can be very inaccurate and overconfident i…

You can also think of this as using the first two terms of a Taylor series approximation in log domain and throwing away the rest!
Post reply on HN