@GaryMarcus Your statement would be true if OpenAI or some other company has trained a ~10 trillion parameter model without gaining any performance over GPT-4. Do you happen to know if this is the case?
If you don’t, you have zero evidence to back your claim.
@ESYudkowsky Has it ever crossed your mind @ESYudkowsky that we will build a system that is superhuman in every way (like GPT-4, AlphaGo etc), and nothing bad happens?
Nothing is”alive”, nothing has its ”own motivations”, the whole system is as dangerous as AlphaGo is individually.
@GaryMarcus@ericschmidt You don’t know what you are talking about Gary.
No one says these are laws of physics. Scaling laws predict very well the performance of the model with more data + compute – these are the real empirical facts.
You not having access to the new models is not “empirical data”
@RomeoLupascu @pratikmehta94 @GaryMarcus@generic_void@RosenzweigJane It is market – some players supply capital and some consume capital.
Those who cannot get capital because their ideas are useless should not blame those who can
@GaryMarcus@sama Gary my man, are you aware that we can plot the increase in performance of LLM’s when we increase compute + data?
Should we stop LLM development even though we can see precisely how we could make more intelligence by scaling the models?
@GaryMarcus Gary is it true that you are currentlu hard at work coding a neurosymbolic system and developing a completely new type of AI model along the lines of your foresight?
@GaryMarcus Gary, there’s a rumour you are developing deep and rigorous theory for a novel, hybrid neurosymbolic system, which is going to beat GPT-4.
People are saying that on top of theory building and experimental setup, you are actually coding it yourself.
Is this true?
@GaryMarcus@beffjezos Looking forward to see you build this alternative future Gary, and by doing so proving to everyone that LLM’s are not the right way forward to progress beyond the current capabilities.
you can finally see "AI safety" for what it is
blatantly authoritarian total surveillance to limit the use of large language models to organizations approved by an unelected, un-policed, cabal of "ethicists"
we have to stop them at all costs
@GaryMarcus@geoffreyhinton@elonmusk Do you think you are able to provide actual risk scenarios as well as practical approaches on how to remedy those risks, or are you just jerking yourself off in public?