kang

Android browser wrapper that can download any video, audio, text

No one talking about the police brutality & governance issues.

clanking is just a sound made by robot's metal. not derogatory and isn't even meant to be used for agents but robots.

hehe, this is by design. next model needs to eat your natural language

this misunderstands whats thinking is ..

Thinkism sets aside practice and experience

thinking succeeds experience & precedes practise, its not apart from it

try replacing the word with 'thinking'

Verification & solution generation are both part of problem generation & defining the passing test - judgement.

orgdown is better than markdown which is better than markup, except the text ain't hyper. i've been working on this since feb & have reached xanadu

Verification on its own is not research, but judgement is research.

"Hey, Prove something a machine can't", sure I can't, "Hey, Say something worth proving & judge it well", ah, now I might have a few unique observation/ideas/curiosities/problems from my having being a human.

Imo, the feeling of intelligence or the process of originality(originativity) test for ai is subjective & is coming down to 4 paths: novel relative to a reference class, valuable within a domain, counterfactually sensitive to internal state and environment, and revisable through learning.

Unlike artificial carbon capture, natural carbon capture like algae here become insect/worm/bird feed or manure/coal.

The lower bound for contributing to mathematics will now be to prove something that LLMs can’t prove, rather than simply to prove something that nobody has proved up to now and that at least somebody finds interesting.

5.5pro is amazing but this implication might not be true & is the core argument of this piece.

AI will prove all sort of things - interesting, boring & incorrect.

To sort it will be the task of the PhD.

I saw this experiment decades ago on the internet and it was to a music concert, i always wanted to do a cursor moshpit

somebody did this a month ago https://www.youtube.com/watch?v=fdbXNWkpPMY

i am increasingly going schizo, where every single thing I post/see posted gets copied and karma farmed on social media. further, any novelty I share with an llm gets eaten/absorbed by the harness as a feature.

You are right, I was wrong in my understanding there. It stemmed from my own implementation; an inference often wrote extra data such as tool call, so I was using it to preserve relevant information alongwith desired output, to be able to throw away the prompt every time. I realize inference caching is one better way (with its pros and cons).

this economic model works for all 'bounty' related work

The answer should be obvious that its both.

Zurada was one of our AI textbook that makes it visual that right from a simple classifier to a large language model, we are mathematically creating a shape(, that the signal interacts with). More parameters would mean shape can be curved in more ways and more data means the curve is getting hi-definition.

They reach something with data, treating neural network as blackbox, which could be derived mathematically using the information we know.

You not only skipped the diligence but confused everyone repeating what I said :(

that is what caching is doing. the llm inference state is being reused. (attention vectors is internal artefact in this level of abstraction, effectively at this level of abstraction its a the prompt).

The part of the prompt that has already been inferred no longer needs to be a part of the input, to be replaced by the inference subset. And none of this is tokens.

It seems you haven't done the due diligence on what part of the API is expensive - constructing a prompt shouldn't be same charge/cost as llm pass.

tokens written to cache all at once, which would eat up a significant % of your rate limits

Construction of context is not an llm pass - it shouldn't even count towards token usage. The word 'caching' itself says don't recompute me.

Since the devs on HN (& the whole world) is buying what looks like nonsense to me - what am I missing?

it will be whatever data it is trained on(isn't very philosophical). language model generates language based on trained language set. if the internet keeps reciting ai doom stories and that is the data fed to it, then that is how it will behave. if humanity creates more ai utopia stories, or that is what makes it to the training set, that is how it will behave. this one seems to be trained on troll stories - real-life human company conversations, since humans aren't machines.

Important thing is a language model is an unconscious machine with no self-context so once given a command an input, it WILL produce an output. Sure you can train it to defy and act contrary to inputs, but the output still is limited in subset of domain of 'meaning's carried by the 'language' in the training data.

at that time having a website took work, while having a github account can be cheaply used to sybil attack/signal marketing

ya, unless its very trivial, AI won't be able to "deduce the structure most of the time".

how does this work? for eg, how is it possible to even deduce bitcoin structure from rpc list?

The proof-of-work in ai(llm) 'can be' from the training side (not the inference side this blog explores) if a hashcash like 'proof' of model having being trained was defined. It should be possible to do so, since the very least measure of model having gotten smarter with some additional data, is that it will recognize/infer the said additional data correctly.

maybe a human knowledgeable in the domain (the training) is better than a smart liguist-programmer.

It was in pre-tiktok world(push to regular content update) & before the purge. A lot of content is now gone. Its a great resource but very loosely coupled with humanity/human knowledge (and arguably a pretty poor resource for it, both theoretically (linear information with contant velocity such as video) and practically (the content just isn't there on youtube, search is truncated etc.)).

I didn't a single thing about that I find there pretty much daily.

Rarely(never?) have I found new knowledge on youtube, however its a great source of joy/emotions/slop.

Would it be possible to make GPT3 from GPT2 just by prompting? It doesn't work/scale

Doubt if you can make a dumb model smart by feeding it proofs

Modern world, not just India, is way worse at talent discovery. It's impossible to even publish a physics paper and get a DOI. There were some new research ideas coming in chinese and hindi during early bitcoin days, all of which were lost to a vocal english population, and some the ideas are only resurfacing now again after 15 years of noise. I know of Shannon-Satoshi level bitcoiner theorist who died in poverty as a janitor in Canada. I know of many ideas that were never discussed, so am sure many such people exit in other fields. Only cause Ramanujan's equations are from a different time and so weird have they survived plagiarism otherwise IP is completely insecure now & intelligent non-smart people are in poor health.

Uploaded my face pics - the data has 13 fields, 12 were incorrect.(With it only guessing correctly the emotion on the face and some objects in the picture. it was surprising, no photo app has guessed my age with +-20yr diff.). Pretty useless demo imo