<div dir="ltr"><div dir="ltr"><br></div><br><div class="gmail_quote gmail_quote_container"><div dir="ltr" class="gmail_attr">On Mon, Mar 16, 2026 at 11:49 AM Adrian Tymes via extropy-chat <<a href="mailto:extropy-chat@lists.extropy.org">extropy-chat@lists.extropy.org</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">On Mon, Mar 16, 2026 at 11:16 AM Jason Resch <<a href="mailto:jasonresch@gmail.com" target="_blank">jasonresch@gmail.com</a>> wrote:<br>
> On Mon, Mar 16, 2026 at 9:49 AM Adrian Tymes via extropy-chat <<a href="mailto:extropy-chat@lists.extropy.org" target="_blank">extropy-chat@lists.extropy.org</a>> wrote:<br>
>> On Mon, Mar 16, 2026 at 9:45 AM Jason Resch via extropy-chat<br>
>> <<a href="mailto:extropy-chat@lists.extropy.org" target="_blank">extropy-chat@lists.extropy.org</a>> wrote:<br>
>> > On Mon, Mar 16, 2026, 6:55 AM John Clark <<a href="mailto:johnkclark@gmail.com" target="_blank">johnkclark@gmail.com</a>> wrote:<br>
>> >> On Sun, Mar 15, 2026 at 2:38 PM Jason Resch via extropy-chat <<a href="mailto:extropy-chat@lists.extropy.org" target="_blank">extropy-chat@lists.extropy.org</a>> wrote:<br>
>> >>> > So long as the exact same input and context are provided, their output is the same.<br>
>> >><br>
>> >> Yes but the context is never the same. Regardless of if it's biological or electronic, a thinking brain is in a constant state of flux, it is never the same from one moment to the next.<br>
>> ><br>
>> > We were talking about the context window for an LLM. This is input, and it can be represented in binary 1s and 0s. It is easy to copy.<br>
>><br>
>> Just like, in theory, the input to a thinking brain can be.<br>
>><br>
>> You keep ignoring the problems with precisely capturing literally all<br>
>> of the input. You just assume that can happen.<br>
><br>
> We're both talking about LLMs here, are we not?<br>
><br>
> Are you and John thinking of a robot/android brain that walks and talks and sees live video?<br>
<br>
I can't speak for John, but I'm talking about LLMs.<br>
<br>
Perhaps an analogy. You know basic electrical engineering, or you can<br>
look it up, right? Say, calculating the current for a given circuit<br>
that connects a battery and a resistor, where the voltage level of the<br>
battery and the resistance of the resistor are known, is something<br>
that you could do if you had to, right? (Again: this includes being<br>
able to look it up.)<br></blockquote><div><br></div><div>But note those examples of currents are analog, imprecise things, not exact, digital, things like the binary strings of digital input that go into the context window supplied to a LLM.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
<br>
So...just because you can calculate that, you therefore can<br>
immediately calculate every current level et al going on in a typical<br>
CPU while it is running a complex operation, right? Just because the<br>
former is something you can do by hand, the latter is not only<br>
possible but practical for you to do by the exact same means (by<br>
hand), then?<br></blockquote><div><br></div><div>At the level of software (for digital computers) what matters are the 1's and 0's, not the exact values of voltages, which I admit are never exactly the same between two CPUs. But we don't use analog computers for this very reason: they're inexact.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
<br>
The answer is, of course, "no". But that's basically what you're<br>
suggesting here. The answer is "no" for basically the same reason.<br></blockquote><div><br></div><div>I think you're mistaken if you think CPUs are so unreliable that we can't perform trillions of operations without a random bit flip corrupting the result. Computers are for more reliable than you seem to be supposing here, especially when technologies like ECC RAM are used.</div><div><br></div><div>For example, when overclocking a CPU, it is common to run a stress test for as long as 24 hours straight, running a tool like Prime95 ( <a href="https://en.wikipedia.org/wiki/Prime95#Use_for_stress_testing">https://en.wikipedia.org/wiki/Prime95#Use_for_stress_testing</a> ) which can detect and report even single-bit errors introduced by any step of the computation.</div><div><br></div><div>Jason</div><div><br></div></div></div>