It's not about who wins the battle of models, rather if the models employed would theoretically (symbolically) be a true representation of an AGI with potential for ASI.
I think that LLMs on their own simply won't hack it. You may be satisfied with the tradeoffs in commercialized value, but there are researchers who are capable of predicting the quantum limits of current "solutions". The profiteers profit while old-school scientists slog away, ever waiting to pounce on their insights. Human behaviour, as self interested as it is, suffers the Icarus complex. In your argument, most likely you"ll have to aggregate all those costs, placing most-correct AGI beyond the reach of the lifetimes of all the players of the day. Where does it leave this symbol of human ambition? In a negative value, in an infinite loop. Simply, because we have proven anf persist in a model which proves that we have no respect - or perhaps insufficient understanding - of the cosmic perfection in the conservation of energy. No one model will do, not unless in its holism it would generate a net, positive value. The number of that value would always be equivalent to 1. Scientists are beginning to understand the simplicity of this thought. It's about pattern languages, not brute force. Is an LLM a pattern language? If so, is it sufficient to express all aspects of a "language" for describing and specifying and managing AGI evolution in, holistically? If not, what is lacking and how can it be realized? I think nature is pragmatic. It adds and it subtracts. If AGI is a symbol of a natural system, then do the sum. On Sun, May 19, 2024, 02:54 Matt Mahoney <mattmahone...@gmail.com> wrote: > On Thu, May 16, 2024, 11:27 AM <ivan.mo...@gmail.com> wrote: > >> What should symbolic approach include to entirely replace neural networks >> approach in creating true AI? Is that task even possible? What benefits and >> drawbacks we could expect or hope for if it is possible? If it is not >> possible, what would be the reasons? >> > > Surely you are aware of the 100% failure rate of symbolic AI over the last > 70 years? It should work in theory, but we have a long history of > underestimating the cost, lured by the early false success of covering half > of the cases with just a few hundred rules. > > A human level language model is 10^9 bits, equivalent to 60M lines of code > according to my compression tests, which yield 16 bits per line. A line of > code costs $100, so your development cost is $6 billion, far beyond the > budgets of the most ambitious attemps like Cyc or OpenCog. > > Or you can train a LLM with 100 to 1000 times as much knowledge for a few > million at $2 per GPU hour. > > > *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + > delivery options <https://agi.topicbox.com/groups/agi/subscription> > Permalink > <https://agi.topicbox.com/groups/agi/T682a307a763c1ced-M5d7336a46b79663a410d119c> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T682a307a763c1ced-M892519d3918783ea7007180d Delivery options: https://agi.topicbox.com/groups/agi/subscription
