intelligence
Executive Summary The fundamental problem of the market for artificial intelligence is that the fundamentals are being driven by ideologically and emotionally invested actors, instead of skeptically evaluating the advancement of the current paradigm of large language models. Hence there is not even a question as to whether the merits of merely the short term profits of the AI industry overall can justify the investment as it currently stands. Nor can there be an honest self-evaluation of the advantages and disadvantages of LLMs. LLMs stand on the far side on the cognitive balance between subsumption and symbolic and thus have a spikey intelligence. Hence because of the inherently spikey nature of LLM, it is not able to be commoditized efficiently because it is not of equal use in every context. Transformer architectures excel at a superhuman level at smoothly transitioning from different isomorphic forms. Hence it is most proficient at translation, coding, and mathematics, skills that derive from a form of literacy but not efficiently a robustly general intelligence. Because it is always easier to persuade than necessarily demonstrate the truth of a proposition, reward hacking means any vaguely defined goal will be short-circuited. Finally AI is part a broader trend in western economies of the systematic underinvestment in labour. After hollowing out the bottom through underfunded education, immigration as a labour market stop gap, and generally the decline of a diversified economy, AI is a final attempt to right everything. Dedicated AI supporters, often the ones leading frontier labs and the Silicon valley hedge funds that invest in them, believe in a near-theological commitment to the singularity and the post-scarcity economy. There is nothing that be done to convince them to deleverage properly and hedge their risks. Chipmakers like NVIDIA are then betting on this religious devotion, and the fact they are manufacturers of strategic commodities. NVIDIA believes because they control the pipeline and the market share that the US government will be the backstop and lender of last resort to any collateralized spending they encourage through vendor-financing, no matter the risk.
https://www.youtube.com/watch?v=2ZU3j4GQ4K8 Human motivation doesn't work well on mostly unsupervised. Uncanny valley of task replacement
-> task supplementation is non-linear (don't get typical economies of scales) Code Writing (well founded) versus Code Review Speed + Feature Bloat vs Optimization vs Maintainability
A Short Stay in Hell Can you write a paragraph with misspellings
Sure! Here's a paragraph with intentional misspellings:
I recieve alot of emails everyday, and its hard too seperate the important ones from the junk. My freind told me I should definately use a filter, but I havent had time to set it up. Untill I do, I'll just contiue too scroll thru my inbox, hopeing I dont miss anything importent. Its truely a hassle, but I supose thats just how it is somtimes.Can you write a paragraph with misspellings
Here's a paragraph full of malapropisms:
My uncle is a real pineapple of the community — always ready to offer his input, even if nobody asks for it. Last week, he tried to explain the whole situation, but honestly, it was all Greek to me, like some kind of undiscovered continent. He said we needed to nip it in the butt before things got out of hand, and that we should take everything he said with a grain of assault. I told him he was having a nervous breakdown, but he insisted he was just being a devil's avocado. Still, you can't help but love the guy — he's a wolf in cheap clothing, but his heart's in the right place.https://en.wikipedia.org/wiki/Eggcorn https://en.wikipedia.org/wiki/Malapropism
AI + Humans have different failure modes -> Mistakes that they make
https://www.youtube.com/watch?v=xmoIrwTsZbg
Reward hacking is the exploitation of a vaguely defined question Is that objective subjectivity?
A standard answer is correct, and lowest energy answer
There are many other answers that are equivalent but might be better at guessing at the higher order associations
The more incomplete answer
The more incomplete answers are better after a certain bound (where the problem is solved)
Minimum viable answers: a terse but correct answer is often better than a fuller one because it respects the reader's time and forces clarity — the incompleteness is a feature, not a bug, once correctness is established.There is nothing outside the (con-text) window. Il n'y a pas de hors-verrine.
Preference function over all theories that varies the length of the minimal universal theory Is that objective subjectivity?
A sufficiently complete answer contains every simplier question
How useful is generality, is subjective
https://plato.stanford.edu/entries/xenophanes/
Conditions for rationality form the conditions for civility
If irrationality is decided from the heuristic of the inability to convince someone of fact that is the difference between them
Either a superrational or an irrational subject is given to the same result
Therefore a rational subject is irrational to a superrational subject
And a irrational subject superrational to the rational subject? Is that objective subjectivity?