[skalil]

It’s happening!  The next step of Saving Wisdom.

This breathes new enthusiasm into creating and organizing my training data.

I feel like I need to be prepared to train new models as they become available – keeping my training data hygienic and organized…and ready to go.

Thank you Brian et al for all you do.

[skalil]

Your article helped me better understand the difference between LLM and diffusion models, so thank you for that.  I also use Grok to clarify/elaborate my own specific understanding  – Grok as my research assistant/teacher!

There are so many interesting things happening now that it’s impossible for me to stay current – not to mention having downtime to just think.
<div>Here is my idea:  It is said that LLMs are a mirror to the human condition – that is to say LLMs are trained on human cognitive output (language) and therefore model human “thinking”.  But can the same be said about dLLMs?</div>
<div></div>
<div>dLLMs use the same statistical framework to “learn” but the diffusion scheme is not limited to only language like LLMs are.  You can in principle use any image, sound…I suppose any analog data as input to the model.  Doesn’t this free the dLLM to tap virtually infinite data from our cosmos to “learn”?</div>
<div></div>
<div>Sure, you’d need a transducer to capture the data, but imagine training a dLLM on data inaccessible or muted to our 5 senses.  From ultraviolet light or radio frequency…or gravitational fields…or EMF…or ionizing radiation, chemical cues, pressure/vibration…the list goes on.  Imagine asking a LLM/dLLM complex to describe the Cosmos after learning these things?</div>
<div></div>
<div>What do you think?</div>

[skalil]

Hello @Brian,

I’m trying to keep this thread alive…curious here about the status of the Wisdom Keeper project.  I can’t be the only one who is enthusiastic to start training using my own data…but things seem to have gone quiet.

Following you on X I learned that the work you’re doing on “YOUR AI” is related to the Wisdom Keeper.  This is fantastically awesome.

Anyway, I’m hoping the more private Multiplex domain might be a place to discuss this more openly with less potential sewage than X?

Thanks,

Steve

[skalil]

Hello @Brian,

It’s me again.

Please know this comes out of me with love…through my fingers…into the keyboard and onto this forum…  I am seeing a lot of interest in Wisdom Keeper but I’m not seeing how I can help get the WK off the ground (beyond my own 1000Q and data organization).  As I have noticed with some observations of X, perhaps you need more active outreach/communication related to the WK?  I’m thinking something like a vague high-level rollout schedule with vague dates that reflect current “staffing” and milestone uncertainties would be – well, triggering and attention-getting.

I say this with humility knowing literally nothing about how much time you personally allocate nor who else may be closely involved in the project.  I’m just responding to the “feeling” I get as an enthusiastic observer/participant.

I’m not an engineer but I am a tech program manager and technophile.  Maybe the scope of work for a WK prototype is so enormous that the current level of general awareness is deemed appropriate given that you would want to coordinate public interest with timing of the release of a functional WK.  I’m speculating.

But I am able and willing to lend a hand.  I can bring project management skill and judgement and beta testing to the table, if needed.  Heck, I’m even willing to buy some HW and devote some time to help digitize and prepare training data.  Do you have (for example) workflows documented that could be distributed as part of training a small pilot group of smart, dedicated enthusiasts that could be the start of scaling up?  Something like that.

Anyway, you’re obviously a busy man.  Perhaps WK is one of those ideas that can benefit from some organized crowd-sourcing…starting with us on Multiplex!

 

Best,

Steve Kalil

[skalil]

Hi,

Hello @Brian,

I’m a fan, watching from afar.  Love your work.

I want to become more active building my WisdomKeeper AI with my data but not finding any practical guidance I can bite my teeth into.  Are you aware of any documentation/tools to get started…or is this still incubating in the idea stage?

I’m thinking about things like recommended hardware configs, OS and application environment and especially workflows.

Many Thanks!

Steve Kalil

[skalil]

Hello Fellow Explorers!

Does anyone have information on the status of the AI toolkit that is associated with the SaveWisdom project?  I suspect like me many people have been recording answers to the 1000 questions over time (as well as preparing journal entries, photographs, other audio files, etc) but I’m just not noticing any info on actual AI training tools…or not looking in the right places!

Thanks!

Steve

 

[skalil]

I am s-l-o-w-l-y beginning to understand this idea of comingling of human and artificial intelligences…and what this might mean to what I expect to be captured by my Wisdom Keeper.

https://johnnosta.medium.com/total-recall-99db3ab86ee6

[skalil]

@dieregterjohan
I enjoy your posts!  Makes me think.  Until your post I didn’t seriously consider the magnitude of the additional information contained in the acoustic patterns of the WK recordings…nor how this could be abused if accessed by nefarious actors/AI.  Hell, I’ve already delivered gobs of audio into the digital domain…cell phone calls, web conferencing, etc!  Makes me – as you say – scared – and is further reason to always think about security.</div>
I do hope @Brian has the bandwidth to pick up your thread – would be good to read that.

Finally, your last paragraph.  I’m hoping you can expand your thinking here.  You show urgency to capture your wisdom quickly now, a concern that your WK answers (I think?) will be diluted or corrupted as more multi-modal AI models are released.  Why?  Perhaps I am misreading.

Here is where my head is at:  I think you are much younger than my 60 years.  My point is that I have built up many actual experiences over the years that are now being encoded into my answers of the 1000-Qs.  Experiences nearly void of AI influence.  You are in a different situation where your future experiences have the potential to be “contaminated” with ever-improving AI around us in everyday life, thereby distorting YOUR wisdom into something not entirely your own.  Do I have that right?</div>

I am not clear why you think evolving AI impacts the wisdom that you choose to encode.

Furthermore, I wonder (as you may too?) if it’s even possible for us humans to maintain/preserve/develop human-based wisdom in a world that is releasing and integrating more machine intelligences into everyday life?

Thanks.

• This reply was modified 2 years ago by skalil.
• This reply was modified 2 years ago by skalil.

[skalil]

@feisty

You’re right @feisty – it’s a good question to ask what the expected latency performance is for one’s Wisdom Keeper.  I can’t, however, evaluate that without having an operational WK to play with.  Ha!  So I suppose the best approach would be to just initially build something to support basic functionality with an eye to future upgradability in speed/memory/storage as needed.

• This reply was modified 2 years ago by skalil.
• This reply was modified 2 years ago by skalil.

[skalil]

Hi Again,  @dieregterjohan  (and others out there!)

Thanks for your insight.  AI Marketers is a domain that I personally haven’t given much thought to, though I’ve probably already inadvertantly bumped into one.  Thank you for highlighting this important idea!

Are you noticing Brian’s X posts regarding his success training GPT-4 Multimodal with user manuals that are primarily visual in nature?  I am hopeful we are closer to a workflow to train GPT4All with our personal visual “data”… though my cynical side says this capability in GPT4All will not be available until the paywalled monetization of this LLM capability is exhausted.

Regarding your commentary on the Amorai business case for use in human “relationship building”…  Incredible.  As you say:  We need to keep an eye on this – for being forwarned is being forearmed.

And then there is the porn industry driving technological inflection points…first VHS video recording in the 1980s and now moving toward AI companions.

I will say, however, I feel like the potential benefit to overall human mental health by these new technologies is huge.  I am imagining my personal-AI helping me make sense of life’s everyday stressors using talk therapy, cognitive behavioral therapy, etc.  Personalized mental health assistance with me at all times and at a moment’s notice.  I feel like many of our social problems stem from individual-level psychological distress.  Even toxic political and media influence might be put into perspective with a personal mental health counselor.

I’ve talked to professional therapists and psychiatrists about the role of AI in mental health and it’s fairly consistent:  How dare I consider non-human therapy into a purely human domain!  Almost like it’s only humans that can properly treat mental dysfunction.  To this I say “dysfunction” is a continuum of mild anxiety to severe illness.  There are some issues that can be treated by AI…and others only by a trained professional.  Our societal problem is so huge we absolutely need AI “therapists” to fill the gaps on the low-complexity end of dysfunction.

Anyway, please forgive me, sometimes I go off on tangents!

We live in amazing times.

 

 

[skalil]

Hi @dieregterjohan,

Yes!  A “second version” of myself.  It’s *that* important, though I don’t fully appreciate the West World reference, I grok it.  I never thought about being strictly offline while recording (cell phone “listening”), but it’s an excellent point and thank you for that.

I am impressed at your age you have the insight to start this project … as for me, I’m 60 and sense my mortality which I think is part of my motivation.  I will say, though, I have been keeping a journal for years and I expect this to become part of the WK though it has a different scope than the 1000-Q.  I also think about all the other data I’ve generated over the years (digital and/or analog) that I think will enrich the LLM if I/we can figure out how to prepare that data for LLM training (I’m thinking about pre-digital photographs/negatives for example).  This is all new and we are the pioneers – this is so much fun.

I’m interested in your reference to psychobiography and telemetry for psychometrics.  You may not have time to expand on this but I will poke around to learn.

Thanks for the exchange – Keep the Faith!

Steve

 

 

 

[skalil]

<div class=”bsp-mentions”>hello @dieregterjohan , yes, I’ve noticed changes in myself since pecking away at the 1000-Q list, but I can’t tell if the change is influenced more by the concept of the Wisdom Keeper itself.</div>
<div></div>
<div>By this I mean – the answers I find myself crafting to the 1000-Q’s seem super important, designed to ultimately transfer “whatever I am” into the noosphere.  It’s not just how I feel at this moment speaking as a stream of consciousness.  It’s more structured – I feel the pressure to organize my thoughts logically, in a manner that best answers the question for the purpose of the WK…knowing there will always be other unanswered questions to probe different angles/perspectives/lessons/etc of this entity I call “self”.</div>
<div></div>
<div>Progress is slow for me.  Maybe less than one answer a day on average.  Maybe multiple recordings for the same question.  I’m also picking questions that seem “easier” to craft a response.  I suspect over time I will repeat or be contradictory or be illogical in some places.  I’ve copied the questions into a single giant text note on my phone and make note of the audio filename next to the question so I can keep track of which questions I’ve recorded an answer and the filenames into the recording device.</div>
<div></div>
<div>That’s how I do this but I’d be interested in hearing how others are approaching the project in this sense.</div>
<div></div>
<div>I’m thinking (hoping!) that the GPT4all (or whatever LLM I eventually choose to deploy and train for the WK) will, after training with a good chunk of my 1000-Q answers, eventually have the capability to answer my questions to it that clarify/enrich/correct inconsistencies in the WK model.  Then again, maybe it’s exactly those illogical inconsistencies that give my WK model its “personality”.  See what I mean?  Thinking about this too deeply can be exhausting!  :)</div>
<div></div>
<div>All for Now,</div>
<div>Steve</div>
<div></div>
<div></div>
<div></div>

[skalil]

Hello @islaydragons247 , (and anyone else watching this thread) checking in here.

I agree w @Brian (our KeyMaster!) – your post is comprehensive and brimming with clarity of thought.  Thank you 10,000 times over, and thank you for the GPTforAll idea.  I will pursue to learn.

If Wisdom Keeper should be accessible to all, the “workflow” as you say, should eventually be super easy and intuitive (or maybe call it a user interface) .  As for me, I generally understand your workflow except for the more technical items like…

• Tokenize and preprocess text (stopwords, stemming).
• Vectorize text (TF-IDF, Word2Vec, etc.)

I’m sure this will become clearer over time, it’s just – as we say – relatively early in this project.  I’m thinking this thread folder is a good place to keep an eye on.  There are, however, other threads starting that intersect like the one about personal AI – following Brian’s idea of “own your own AI or AI will own you”.  Wouldn’t this Wisdom Keeper AI LLM were talking about here be similar (or maybe identical?) to our personal AI?  Should I expect to deploy function-specific and separate AI’s?  Except for the (temporary?) off-line requirement for the WK, I feel like we’re bumping up against high-level architectural and air-gapping considerations here for all the personal AI LLMs we as users decide to deploy.

Regarding the use of a GPU-based off-line computer for the WK.  I think the vision is that this is to be accessible to everyone everywhere with wildly different cultures, ages, backgrounds, educations, etc.   I am *assuming* the WK LLM will be been pre-trained on “other language” to some useable baseline capability such that we users then train the model on the wisdom-specific data we’ve accumulated.  I don’t see this scaling if you have to buy expensive hardware.  I remember hearing in a podcast between Brian and Jordan Peterson where Brian is conceptualizing a purpose-built handheld device (like a cellphone form factor with enough capability to support this).  Yes!  But while we wait for such a device, perhaps we target enthusiastic people (like us) to build/repurpose something affordable and commonly available.  The more I think of it, yes – you’re right – Linux is probably the best OS given the evil interconnected nature of Windows.  Holographic crystal memory, yes, sure!  Why not?  We need creative vision, right?

I would like to understand more regarding the types of data we users are preparing for WK LLM training.  I understand the basics of written word content (either text-converted 1000-Q audio files, or documents/email/journals already in text format) that we’ve already written in language that is then prepared with “tokenization” and “vectorization”.  But what about all the other data we generate, like

• Photographs and videos.  I have zillions of visual content that I’ve created over the decades that have captured people, places, etc that “tell a story” about what I’ve found important over the years.  Some photos have metadata, some are scanned chemical prints.  Surely this is rich data for my WK.  “Whisper API” to convert the audio files into text…what to use for photos and video?  Further, I would hope the WK LLM would consume these images then ask me what they mean …what is this face that keeps appearing in your photos?  Who is this?  Tell me more about this face…or these scenes that keep appearing (holidays, birthdays, etc).
• E-Mails and Documents.  Again, a rich domain for the WK though the LLM it might need my real-time clarification or context to properly integrate into the WK.
• Finances. Amazon purchase history (or any other e-commerce or credit card history or investments for that matter).  What and when I buy things tells a lot about me.  I manage a Quicken file for family finances – use this?!
• 3rd party held data that you can request.  For example, Google is collecting geo-location for most everyone (if you consent to this).  Wherever you travel with your mobile phone is recorded.  Another example is browsing history.  Rich data!
• It goes on and on.  If we’re talking “open kimono” here with the 1000-Q recordings, why not include everything?

Of course me saying “everything” implies we need separate workflows to properly prepare these differing data types for training.  Further, it’s going to take *years* to record answers to the 1000-Q, but this other data is available *now*.  I feel like a parallel effort may be wise given the state of this LLM technology combined with the immediate availability (and complementarity) of these other data types.

What an amazing thing to think about.  I really can’t wait to have a conversation with my Wisdom Keeper!  I’m trying to temper the feeling that, once deployed and allowed to grow organically across the population, this is a major inflection point for humanity.

I feel like I want to help move this rock uphill.  I’m not a trained engineer but I am a retired program manager from tech.  @Brian – Is there a project plan in the works?  😉

Warmly, Steve

[skalil]

Picking up on the dialog between @islaydragons247 and @jtkolarik…

I’m struggling with the practicality of establishing, maintaining and improving the depth, nuance and quality of the input I create to feed the LLM.  After all, our “wisdom” (as it is) is far more complex and multifaceted than any external technology can possibly capture.  I see this as a life-long pursuit that continuously build layers of complexity in the AI/LLM.  In “the end”, whatever the state of my LLM wisdom will have to be sufficient for the noosphere.

The idea of the LLM promoting me to “fill in the gaps” of its model of my wisdom is *very* appealing.  My Wisdom Keeper will protect me from unnecessary duplication and point out contradictions over time that need clarification – in essence the LLM will protect the time I have left and guide me to use it most efficiently.  After all, I am woefully imperfect and these problems are a certainty.

Further, what is the plan “at the end” when I die?  I will have built a piece of hardware that has all this data on it (plus the backup dataset!).  What happens to this?  Should it be part of my will?  Is someone working on a noosphere infrastructure?  Who can I trust to properly take care of this?  My children?  Wisdomkeeper.org? I could be hit by a bus and be in this situation at any time.  Maybe my WisdomKeeper remains sequestered and silent until one generation passes then it becomes part of the noosphere “public domain”?

Maybe we have an attorney on Multiplex that can illuminate some of these issues…I can’t be the first to think about these ideas…

Steve

[skalil]

Hi All,

I’m enjoying the conversation here.  I’d like to bring up some practical considerations that I suspect may be common to others and that isn’t yet addressed in this thread…

1) I’m planning to revive an old PC for the solitary purpose of running my personal AI, something specialized and stripped down (HW and SW) and isolated from the internet/cloud and generally protected from any function not related to the purpose of Wisdom Saving.  Are there any suggestions in this regard?  I’m thinking the computer will have the following capability:

1). I’m assuming Windows OS.  Can be any stable version.  Updates not necessary as there is no internet connection (note:  I must accept latent bugs in SW in favor of an isolated system, though current versions are relatively stable for this purpose).  Assumed with this are all the drivers that support peripherals (like printer and audio cards) and the voice recorder.

2). Personal AI LLM.  I’m *assuming* at some point in the near future this will become available to us either through Multiplex or SaveWisdom.org at which point we install via thumb drive.

3). Audio-to-Text transcription.  A program installed to convert 1000-question audio files into txt for AI consumption.

Is it *that* simple?

In the meantime…I am slowly starting to record answers to the 1000-questions and organizing my digital assets…photos, emails/letters, documents, etc for eventual AI consumption.  I also keep a digital journal that captures everyday impressions that I suspect will be part of this project.

Thanks All.

Steve

[skalil]

Hello,

My name is Steve Kalil and I recently left my corporate tech job and am working full time to reclaim my humanity, to expand my awareness and learn some new things along the way.

I admire people like Jordan B Peterson (when he’s in his academic mode) and Brian Roemmele for their clarity of thought.  I’m always looking for new pathways of thinking – I’m a bit frenetic but always openminded (perhaps to a fault).

Anyway, I recently subscribed to Read Multiplex and am crawling through the content.  Perhaps we will cross paths someday.  Happy to be here and learn.

Steve

[Author]

Posts