[ExI] ai in education

[Adrian Tymes]

On Fri, Mar 6, 2026 at 8:21 PM spike jones via extropy-chat
<extropy-chat at lists.extropy.org> wrote:
> LLMs must have some kind of source code, or it would do nothing, ja?

While technically LLMs have source code - to compile weights and use
them to generate responses - most of the strength comes from those
weights, which are technically data.  This is the basis for the "AI
copies the weights" scenarios of an AI exfiltrating itself from a data
center: the weights are, by far, the vast majority of the data that
would need to be exfiltrated - although if a relatively simple program
would also need to be installed and run on the target system, to turn
these weights into a functioning AI rather than mere potential.

On Fri, Mar 6, 2026 at 8:06 PM Keith Henson via extropy-chat
<extropy-chat at lists.extropy.org> wrote:
> There is no source code for any AI that I know about.  There is
> training code with which an AI is trained on a vast corpus of text,
> but nothing a programmer would recognize as code. As far as I know,
> the inside of an AI is a mystery to all the companies.
>
> If I am wrong here, please let me know,

No, no, the companies know the basics of how the AI works.  They might
not know every single detail, but the details are all there if anyone
wants to dig into it.

Just...digging into it would take unfeasibly long.  Longer than a US
presidential administration, to thoroughly examine a single generation
of AI that will be obsoleted in a substantially shorter time.

Minor differences in format, multiplied by the sheer raw volume of
data, also makes it infeasible for one AI to simply "examine" another
AI's weights: even AIs would take unfeasibly long to do it.  (This is
likely to remain the case for a while: a more complex AI could examine
an earlier generation AI faster, but an earlier generation AI is
usually no longer of interest.  In the time that AI took to get more
complex, its competitors got more complex too.)  They can improve
themselves faster by sticking to their current methods.

This is a case of scale mistake - like noticing it is physically
theoretically possible to travel to Alpha Centauri in one lifetime,
concluding it must therefore be easy to head there, and being confused
why no one has set up a colony over there yet.