Jerrrry

  >Whether these training set predictive patterns result in the model performing as you might hope on an unseen text depends on the similarity of that text to samples in the training set.

>similarity

yes, except the computer can easily 'see' in more than 3 dimensions with more capability to spot similarities, and can follow lines of prediction (similar to chess) far more than any group of humans can.

that super-human ability to spot similarities and walk latent spaces 'randomly' -yet uncannily - has given rise to emergent phenomena that has mimicked proto-intelligence.

we have no idea what the ideas these tokens have embedded at different layers, and what capabilities can emerge now or at deployment time later, or given a certain prompt.

They need to provide an small hash with the api result that can be verified by others.

The need to identify and label is a n objectifying cope to an otherwise unfamiliar alien world.

This is the neatest and most useful thing Shown on HN using AI in recent memory.

Well done!

There is nothing you can do with $10bn that you cannot personally do with $1bn.

You can only buy the International Space Station twice.

I'm convinced Michael Crichton was the greatest sci-fi writer after Isaac Asimov.

The alignment problem you've solved was simply moved a few syllables.

How do you simply "not do that" and have a useful model to interact with?

Without reducing complexities to grammatical self terminating cliches, ponder: whatever you tell the model to do, including being honest - it could be lying, and lying dormant awaiting a critical mass of dependency.

I believe that cooperation via Nanny AI state and then eventual departation...then War is the most likely scenario.

That, according to game theory and the AI themselves, maximizes the AI survival long-term.

First, grow Dependency.

First commandment lookin real pertinent right now.

  >It can't have an intrinsic moral compass, so someone must have trained it in. Why not simply... not do that?

Why not factor any number thru sheer exhaustive enumeration?

  >- a plausible mechanism by which the model can distinguish between training and real world use without the experimenter making it explicit

Like the existence of a large number getting factored?

  >You can't add 1-based indices together like you can zero-based indices.

I think you are right but I am unable to articulate why. But I think 0 based indexes are able to fruitfully capture "going nowhere" iteratively than 1 based indexes, which do require a decrement in that circumstance.

Post term abortion isnt cheap

Maybe a male Jack the Ripper as well

Computers are finite ergo the largest number any computer can compute is all ones in binary.

That's it's Busy Beaver number, but think how easy all ones in binary is to compress from an information theory standpoint.

So the most entrophic state would be the one state requiring the most compression.

Your second example subtracts and adds 1 nearly arbitrarily, which wouldn't be needed if the convention of the 0-index wasn't so widespread.

We actually did, it's a solved problem.

Research bans do not inherently work.

Treaties need enforced, and the Streisand effect and arms-race dynamic play into the game theory as well.

It may be the self-aware human bias tainting this reasoning, but it seems convergent with our own psyche/brain processes, and/or inherent to the way we commonly express conscious thoughts.

Percolating tokens that allow a more "accurate" latent space appear to be more accurate, but are nearly actually useless noise. Almost a virtual shower thought.

Because people only put the answer at the end of a grammatically correct statement, with the more "reasoned" statements being more articulately percolated/logically sound, and that is expressed grammatically. These statements are inferred to be associated with intellectual boiler-plate. They may be correlated and not actually causative, but that would require a multiple component architecture with embeddings being used as a proto-"qualia" and that is getting hairy.

Facts should "only" have to be read once, and should be explicitly defined with a more secure of a confidence completely. Implicit inferences from those explicit facts should be emitted from a different, less confident module; with the chat boilerplate being tacitly composed finally when presenting the output to the user.

Of course separating the baby from the bathwater is the hard (not impossible) part.

And Nuclear incidents caused by climate/ai extremists are way more likely than those.

Environmental collapse is a convenient, very portable goalpost though.

You're work is vital for opposing the nonchalant march of privacy erasing norms we are continuing to parade towards.

Depending on the prominence of the user, that could have been further coerced as a fact or reference, and/or inferred as factual if enough people agreed, in the next training session.

Upvotes as truth-confidence scores, what could go wrong?

Ironically some iotas of truth can be outputted directly, but if the most explicitly articulated reference is by a (non power-) user on a forum in a comment, it will be sourced as anonymous/eponymous wisdom (unless suffiently coerced)

Depends on the use and context, benchmarking can hint at the "truth", that it is faster to just divide: ultimately the benchmark would have to have the two implementations and run them thru all possible inputs to know for sure.

It may clearly state that memoization is slower, until you plop it into a benchmark that inlines division explicitly without using a given hardware accelerate.

division by a constant input can be solved faster still, so there can be optimizations on top of memoization that would beat raw division on most peocessors/runtime environments.

  >multiple automata

Automatia

Very convenient for those who are waiting for the waters to get muddier.

Hard to think anyone who can't solve a captcha is something other than a parasite.

Parasites can solve captchas, people with accessibility issues and the poorest of people are the one being locked out.

Why are you explicitly saying "human"...

I like this definition for a "proto"-intelligent trait.

Inner Platform effect

https://en.wikipedia.org/wiki/Inner-platform_effect

good compression = pattern eventually

maximum compression = indelineable from random data

  >Okay to be clear, I've written a paper on exactly this topic which will be announced in a week or so. So you won't find anything on the subject yet, haha. But I use almost exactly this example.

I would use Floating Point arithmetic as the example/analogy: one trades off accuracy/precision for exactness/correct-ness when in the extremities. Answers near the more granular representations will be only be able to represented by their nearest value. If this is forsee-able, the floating point implementation can be adjusted to change where the floating point's "extremities'" are.

I used the above analogy and the following when articulating the magnitude of near-lossless-ness that large LLM's have managed to attain, especially when all of the humanities corpus is compressed into a USB flash drive; the Kolmogorov complexity re-mitted/captured is similar to a master-piece like the Mona Lisa having to be described in X many brush-strokes to achieve Y amount of fidelity.