You know the problem though, BPC is probably not actually bits per character,
it's probably something like take the average sum of all cases and half it and
then apply some log2base to it and add an exp to norm it across dataset sizes
that vary.unfortunately...
I swear I seen it one tim
It's not hard. Random data compresses to 8 bits per character.
On Thu, Aug 19, 2021, 8:19 PM wrote:
> I can't find using Google how BPC (Bits Per Character) evaluation for
> prediction works. GPT has a score using this evaluation. It's possible I
> could compare my score to GPT's if they can use
I can't find using Google how BPC (Bits Per Character) evaluation for
prediction works. GPT has a score using this evaluation. It's possible I could
compare my score to GPT's if they can use BPC in the same way.
I need an exact explanation of BPC, no math terminology or "stretch the [P] and
add
I was working with others who modified PAQ compressors to output the bits
with the predicted probability distribution and feed those back to the
model. Or equivalently, flip one bit in the middle of the compressed data
and decompress.
Since a run of length n compresses to about log n bits, there i
I don't have a problem generating text completions yet. And if it does predict
repeations then something I didn't code yet will be getting bored of
repetition. Often you catch my current code do or the cat the cat the
cat but then it breaks out even after 100 times.
On Thursday, August 19, 2021, at 1:03 PM, Matt Mahoney wrote:
> Still, DNA based replication is pretty close to the speed and efficiency
> limits that Freitas calculated for nanotechnology. Moving atoms exactly where
> you want them takes work on the order of 10 to 100 kilojoules per gram,
> eno
This is an interesting phenomenon. Are you running checksums?
On 19 Aug 2021 21:13, "Matt Mahoney" wrote:
> I've run into that problem too, that text prediction degrades to repeating
> characters. There are no English words that repeat the same character 3
> times in a row, but this pattern is st
I've run into that problem too, that text prediction degrades to repeating
characters. There are no English words that repeat the same character 3
times in a row, but this pattern is still common in many files so we have
to model for it.
On Wed, Aug 18, 2021, 5:08 PM wrote:
> BTW do you know why
I wrote it. All animals evolve to fear the things that can kill them, which
you can argue is different than fearing death because most animals have no
concept of death. We can debate whether a fish on a hook or a mosquito
evading a swat or an amoeba swallowing food actually feels anything or is
jus
On Thursday, August 19, 2021, at 10:11 AM, Quan Tesla wrote:
> ... would you consider your intelligence to be committed to a rapid
> evolutionary process with purpose to eventually assume network-interactive
> cyborgian functionality?
Nope.
I might describe myself as transhumanism-curious. I
Have you read my white paper on last-mile knowledge engineering, which I
shared more than once to this forum? If you hadn't, do go read it and let's
discuss/critique the technical implications for agi development.
On 19 Aug 2021 09:00, wrote:
> I'm not sure what you're trying to say NKT, I'm here
For many years, a program called Magic has been rewriting and merging code
from different languages. Pitrat's agi autogenerated code and in doing so
improved upon known solutions to many, classical mathematical problems.
On 19 Aug 2021 16:27, "John Rose" wrote:
> On Thursday, August 19, 2021, at
I have a general question. Given the nanotech now integrated with your DNA,
would you consider your intelligence to be committed to a rapid
evolutionary process with purpose to eventually assume network-interactive
cyborgian functionality?
I concede there are other methods with which to achieve cy
For example, dot net has a feature called reflection, as does Java, that
enables code to look at itself. Using this all sorts of chores are typically
done in everyday development. Yet reflection has not been utilized to its full
capability IMO. In some ways yes, for example look at NSwag API wra
On Thursday, August 19, 2021, at 12:31 AM, Nanograte Knowledge Technologies
wrote:
> Bots write better software than humans do.
In some ways they do, for example Visual Studio (which I use) has Intellicode
which suggests snippets quite well. But with bots writing software there are
huge improve
Lots of tweaking to do, but got a score of 19,085,594 for enwik8.txt. This puts
me to a new frontier. Decompression not tested but usually it works.
--
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/Td13a829978c4c9f
16 matches
Mail list logo