Connectionists: Fwd: Statistics versus “Understanding” in Generative AI.

[Keith Lambert]

Please just stay on task and consider the gist of my argument. Not
interested in this back n forth which seems pointless and if anything
detrimental. Have a nice day.😎

Sent from my iPhone


On Fri, Feb 16, 2024 at 1:58 PM Gary Marcus <gary.marcus at nyu.edu> wrote:

> Wow. If you going to back to podcasts (not written articles, book chatpers
> etc) from 2014, you are desperate. In fairness,
> a. You are going to have to be more specific about what goalposts you even
> mean there? (I don’t have time to listen to 10-year old podcasts). What
> *exactly *did I say there that you object to exactly?  I NEVER endorsed
> the Turing Test, so far as I can recall. Always thought it was a lousy test
> for intelligence.
> b. You are going to  have hold Hinton accountable for the 2016 prediction
> that we should stop training radiologists. (We can’t have double standards
> here.)
> c. You are going to have hold Lecun responsible for saying last week that
> “ We don't know how to do this properly. It doesn't work for video. What
> works for text doesn’t work for video. And the only technique so far that
> has a chance of working for video is a new architecture that I've called
> JEPA...”, with Sora beating that a week later. (Again, we can’t have double
> standards here.)
> d. Specifically  on the subject of alternatives to Turing, I proposed
> something I call the comprehension challenge in 2014
> <https://www.newyorker.com/tech/annals-of-technology/what-comes-after-the-turing-test>in
> the New Yorker and stand by it. As impressive as the newGemini is, the only
> video test they demo’d (Buster Keaton) did not have audio as part of the
> input. In their own new long-duration task (details not disclosed; maybe
> mostly be memorization rather than the kind of inferences they described)
> they were only at 60%. My prediction that test would be hard; goal posts on
> that haven’t changed.
>
> Gary
>
>
>
> On Feb 16, 2024, at 11:40, Keith Lambert <keithlambert75 at gmail.com> wrote:
>
> 
>
> Fair point and I’ll try to do better. That being said, this entire talk is
> filled with goal post shifting examples. I mean you literally talk about
> creating a sequel to the Turing test, which would be definitionally a goal
> post shift, but I don’t want to get mired down in this….I believe my points
> stand regardless and should be the point of the discourse. To a better
> understanding for us all.
>
>
> https://www.econtalk.org/gary-marcus-on-the-future-of-artificial-intelligence-and-the-brain/
> <https://urldefense.proofpoint.com/v2/url?u=https-3A__www.econtalk.org_gary-2Dmarcus-2Don-2Dthe-2Dfuture-2Dof-2Dartificial-2Dintelligence-2Dand-2Dthe-2Dbrain_&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=wQR1NePCSj6dOGDD0r6B5Kn1fcNaTMg7tARe7TdEDqQ&m=yuIE75RLOV4WpTZyPMpG1iqqm6fUiuekP0pq6pvYCBgMpRpDpjTVMVwHHx4cGlxK&s=KvMcF4lKIQLr2T4Irs4u77l569m2e2WMW5E9zBa6Sog&e=>
>
>
> Sent from my iPhone
>
>
> On Fri, Feb 16, 2024 at 11:15 AM Gary Marcus <gary.marcus at nyu.edu> wrote:
>
>> “ Too much negativity and vitriol around this IMHO” from a guy who opens
>> his missive with an accusation of goalpost shifting that is supported by
>> not a single quotation.
>>
>> The level of discourse here leaves a lot to be desired.
>>
>> > On Feb 16, 2024, at 8:56 AM, Keith Lambert <keithlambert75 at gmail.com>
>> wrote:
>> >
>> > Too much negativity and vitriol around this IMHO
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20240216/ca6bfdec/attachment-0001.html>