Despite marketing efforts hyping superintelligence around the corner, advanced reasoning abilities etc. I don't see much more than folks beating disingenuous benchmarks by modifying training sets, memory, and all manner of parameters, specific to domain of benchmark being tested. Yes, there was a jump in what we thought was state-of-art and yes, I see the incremental changes from there. But I don't see the hype/budgets justified yet, even though I can't tel what will happen if you throw half the planet's power into a city sized datacenter. Anecdotally, most software engineers I've spoken to echo the following:
https://www.cio.com/article/3540579/devs-gaining-little-if-anything-from-ai-coding-assistants.html But maybe we need to train these things for some years for domains that are not as narrow as advertising or chess/go, where you can have the programs play against themselves a billion times in a day with clear results/reward schemes. For those broader domains, the importance of real data is the currency. I'm not sure we should keep feeding it to silicon valley for free, to have them charge us for using the refined (by users and their real data!) tools later. On Saturday, September 28, 2024 at 11:58:15 PM UTC+2 Brent Meeker wrote: > > > > On 9/28/2024 4:52 AM, John Clark wrote: > > On Sat, Sep 28, 2024 at 12:19 AM Brent Meeker <[email protected]> wrote: > > * >> Albert Einstein went from understanding precisely nothing in 1879 to >>> being the first man to understand General Relativity in 1915, and you knew >>> that the human genome only contains 750 megs of information, and yet that >>> is enough information to construct an entire human being.* >> >> > > *> Also takes certain nourishment and environment, which is not zero >> information.* >> > > *Nourishment is not information, * > > Sure it is, not all calories are equal. And growing an embryo into a baby > isn't done your kitchen sink. > > *and energy will not be a problem for an AI, that's why God made nuclear > reactors. * > > *>> Aaronson: Come on! 256^750,000,000 is vastly greater than the number >>> of possibilities one could search through within the lifetime of the >>> universe.* >>> *Me: I agree, and yet it's a fact that random mutation and natural >>> selection managed to stumble upon it in only about 500 million years. * >> >> >> *> More like 3.5 billion years. * >> > > *If we're talking about intelligence then in the 3.5 billion year history > of life the first 3 billion years were irrelevant because during that time > there simply wasn't any. * > > First, even bacteria and archea exhibit rudimentary intelligence. So > advanced intelligence wasn't built on nothing. Rocks had the same start > 3.5 billion years ago, they didn't develop intelligence. > > > > *It was only about 500 million years ago that multicellular animals came > on the scene and anything even vaguely resembling a "brain" evolved that > was able to extract information from the environment and use that > information to improve the animal's chances of getting its genes into the > next generation. * > > *Oh I suppose you could say that even a single celled creature can move > away from something that's too hot or too cold, but if you count that > as intelligence then you'd also have to say a thermostat is intelligent. * > > I do. Intelligence admits of degrees. > > > > *And if you do that then the word starts to lose its meeting. * > > Not at all. Does weight lose it's meaning because you're a lot heavier > than a bacterium? > > > *> First life evolved then eukaryotic life evolved then...* >> > > > *There is certainly no reason for modern software engineers to repeat all > of the dead ends, irrelevancies and downright silliness that Evolution > dreamed up during the last 3.5 billion years! Evolution is a > TERRIBLE engineer, it makes stupid designs (as expected for something > involving random mutation) , it's ridiculously slow, and requires > gargantuan resources. It's not strictly relevant but Natural Selection is > also hideously cruel. * > > But evolution developed the intelligence we have, a lot of silliness > succumbed to natural selection. The funny thing is that software > engineering seems to have fallen into a form of unnatural selection as they > train bigger and bigger LLMs so they no longer understand what they've > "engineered". > > Brent > > > John K Clark See what's on my new list at Extropolis > <https://groups.google.com/g/extropolis> > 5gn > > -- > You received this message because you are subscribed to the Google Groups > "Everything List" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > > To view this discussion on the web visit > https://groups.google.com/d/msgid/everything-list/CAJPayv39NXUgQhRcARcpyb5VL9vTRpHpy89rdLWGXhjthG3yRQ%40mail.gmail.com > > <https://groups.google.com/d/msgid/everything-list/CAJPayv39NXUgQhRcARcpyb5VL9vTRpHpy89rdLWGXhjthG3yRQ%40mail.gmail.com?utm_medium=email&utm_source=footer> > . > > > -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/everything-list/0a46007c-7caa-4019-bcde-41e81b8d0b60n%40googlegroups.com.
