[ExI] How AI understands the world
Jason Resch
jasonresch at gmail.com
Sat Feb 21 21:32:21 UTC 2026
On Sat, Feb 21, 2026 at 3:55 PM <spike at rainier66.com> wrote:
>
>
> *From:* extropy-chat <extropy-chat-bounces at lists.extropy.org> *On Behalf
> Of *Jason Resch via extropy-chat
>
> _______________________________________________
>
>
>
>
>
> Gemini says that these Pure Transformers were never released because they
> were unusable.
>
> Labelling and manual adjustments are essential to LLM development.
>
> BillK
>
>
>
> Useful to most people or not, they understand the world.
>
>
>
> You can give a pure model the phrase: "The capital of France is" and it
> will complete it "Paris." Jason
>
>
>
>
>
>
>
> All AI large language model reasoning still must go thru words. We aren’t
> to the point where we can build a robot which will go about watching and
> listening, then be able to solve this kind of puzzle:
>
Multi modal AIs (which are trained on both text and images, and therefore
are able to see) can do reasonably well on sequence series IQ test-style
questions. For example, see page 5 of this PDF:
https://arxiv.org/pdf/2502.00698v2 in which an LMM (large multi model)
solves a question not unlike the one you give below.
The results (now 1-year out of date) show that humans still do better than
the LMM, but I am sure the gap has closed considerably with the latest
models.
I found the latest results: https://arcprize.org/ (It looks like Gemini 3
deep think is now scoring 80% on this test).
Jason
>
>
>
>
> As I understand it, there is a standing prize of one miiilllllllion
> dollars for any AI which can solve these ARC puzzles (Abstract Reasoning
> Corpus.)
>
>
>
> Perhaps we can figure out a way to translate the images into words, hand
> the words to the AI, then see if it can find the answer.
>
>
>
> spike
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20260221/7b927479/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image002.jpg
Type: image/jpeg
Size: 19032 bytes
Desc: not available
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20260221/7b927479/attachment.jpg>
More information about the extropy-chat
mailing list