When that happens typically the company starts optimizing for sucking money from the government. From the point of view of the consumer Intel would be finished.
Has anyone tried to see what the output looks like if the model is never allowed to be uncertain?
For example, whenever certainty drops below a threshold the sampler backtracks and chooses different tokens. Such that at the end every single token had an above threshold certainty.
I doubt it would entirely eliminate undesirable outputs, but it would be interesting.
Tragedy of the commons is why short term is all that matters and will ever matter to non-ideological investors.
If an action that hurts the stock short-term but will help int he long-term needs to be performed why would you as an investor enact it or even stay for the ride?
You are better off either opposing it or selling your stock and then waiting to see if someone will enact the changes, then you have the "insider" information to know that the short-term stock drop was a good thing for the long-term and rebuy the shares cheaper.
The race for true AI is on and the fruits are the economic marginalization of humanity. No game theoretic actor in the running will shy away from the race. Anyone who claims they will for the 'good of humanity' is lying or was never a contender.
This is information technology we are talking about, it's virtually the exact opposite of nuclear weapons. Refining uranium vs. manufacturing multi purpose silicon and guzzling electricity. Achieving deterrence vs. gaining immeasurable wealth, power and freedom from labor.
This race may even be the answer to the Fermi paradox - that there are few individual winners and that they pull up the ladders behind them.
This is not the kind of race any legislation will have meaningful effect on.
The race is on and you better commit to a faction that may deliver.
Due to the limitations of gradient descent and training data we are limited in the architectures that are viable. All the top LLM's are decoder-only for efficiency reasons and all models train on the production of text because we are not able to train on the thoughts behind the text.
Something that often gives me pause is the consideration that it is actually possible to come up with an architecture which has a good chance of being capable of being an AGI (RNNs, transformers etc as dynamical systems) but the model weights that would allow it to happen cannot be found because gradient descent will fail or not even be viable.
Tornado Cash does this. And you can find articles on how it functions online. You can even read the smart contract that directly implements it.
Roughly: you have 2 secrets that you hash together and the central authority adds the result you disclosed to a list (either to print money or as part of a transaction to transfer money). To spend a note you reveal the hash of one of the secrets (to be added to a list of nullifiers to prevent double spend) and you do ZKP to demonstrate that you possess both of the secrets to *some* note from the public hash list and that the nullifier for that note is what you claim it is. Central authority rejects if nullifier is present in the list.
There are some other approaches to such a system, I believe the Tornado Cash one is the most elegant though it limits you to a discrete number of note denominations.
Note that the proof system Tornado Cash uses is not secure to a quantum computer and such a device will allow to "print money" - in reality, drain the smart contract.
It is currently possible to use ZKP's to set up via a central authority a digital cash system where the bank notes are all anonymous and all transfers are anonymous.
The central authority in this scenario cannot discriminate between transactions - any function that would compare two or more transactions cannot glean any useful information that would allow to discriminate. And and security of the anonymity of past transactions will be reducible to the security of the cryptographic hash function used (the next best thing to Information-theoretic security). As for forging money, depending on what ZKP approach is used even a quantum computer will be insufficient.
The central authority can still print money and can obviously shut the entire system down.
It is interesting to ponder whether or not some government will decide to take such a step and surrender all control (except for the nuclear option) over how their currency is used. It will certainly boost demand for the currency.
Psychopaths are incapable of feeling empathy. I merely restrict empathy to family, friends and people who are in a position to reciprocate. I do feel pity for people who succumb to the mental compulsion to empathize with everyone.
mRNA can be used as indirect vaccine, by manipulating cellular factories to produce a foreign object that draws the attention of the immune system.
There is no known pathway of persistence in this mechanism, and there is good reason to believe such a mechanism cannot exist as it would be a highly versatile attack vector and viruses would have evolved to be much more virulent by using it. Retroviruses do complex work to infiltrate the DNA repository of a cell, if all it took is to drop some mRNA it would have been exploited and 'patched' a billion years ago.