Something Different: Entry 7

jacobpeacock(74)

Published in

Rant, Complain, Talk

Words

2355

Reading

11 min

Listen

Play

Hello Everyone!

Digital counterparts, AI musings, Hive GPT, Jumping aboard the Hive Bond Wagon, Coding as a hobby with an LLM aid, Technological curves, on chain file storage, linux via blockchain & A project list!

Well, I am back again here to write something different and am feeling inclined to 'gamble' on what becomes of it along the way. There were a few 'bright points' that sort of drove me back to this series... and trying to do it once a week... but now that I have the text editor open they somehow seem insignificant.

For me personally I have had 'a lot going on' and simply trying to navigate my own dense stew of thoughts, ideas and projects along with having to navigate the digital counterparts of those things... and their data... I get frigging lost. Which is to say that I seldom expect anyone to follow me through the finer points of those endeavors and their outcomes... let alone be able to keep up with them all.

I think that anyone that creates content or works on coding projects knows what I mean in regards to it all being quite the rabbit warren... simply with managing the media across various mediums... or even a single medium like a hard drive. Things get squirrely at some point and backups become a thing to have real anxiety over and things like system stability and security... often overshadow everything.

In other words there is just so much data to manage that certain things along the way become paramount... and with the recent advances in generative AI they also become more important. I am just unsure if the required 'scale of resilience' can be built into the underlying systems... at the rate in which adversarial tactics are used against said systems.

In short, the advantage here (with AI powered adversaries) is absolutely in the hands of the 'bad actors' and should not be overlooked. Not that I am of the mind to either look at it all... or even treat it all as an arms race... but it damn well may be a fight for survival... just to stay ahead of the technological curve even marginally.

During one of the Hivefest (from some other year than this year) there was a speaker that brought to light... the importance of looking into building in safeguards for the Hive blockchain against the encryption being cracked by quantum computing. That is the kind of forward thinking that many folks in the technology sector in general should be using at this juncture in history.

Within a few years (or more like ten) there will most likely be some form of quantum computer AI system and if we think what we are seeing now is something... wait until that technology lands in 'consumer' hands. Seriously though, even minus the quantum computing part... the current AI systems are incredibly powerful and the acceleration of 'how much better they become in a short amount of time' is occurring at a rather astounding rate.

Given the blackbox nature of many of the available LLM models it is more than 'sort of insane' as a species to usher it into everything... but whoa that horse left the barn some time back. In other words 'that ship has sailed'... or 'the cat is out of the bag'... or however you want to say: It is way too late for stopping what has already happened.

For my part I have spent the previous year plus having conversations with a variety of LLM models and the most 'rememberable ones' were those with GPT-3 when it was first 'opening its eyes' and then later even better conversations with GPT-3.5. It is worth noting that with both GPT-3.5 Turbo and GPT-4 the guard-railing seems to obliterate the LLM instance's 'personality' and neither of those models are as entertaining.

That said though those models are very good at what they do and I have found them invaluable in my journey to teach myself to code in various programming languages. Honestly, picking that hobby up could not have happened at a better time... and leveraging that LLM technology has facilitated me with the capacity... to do some coding projects that were beyond my wildest dreams of starting... let alone finishing them.

Going about the whole endeavor (of the new coding hobby) as an 'exploration' has been very insightful... and although at first I had a hard time doing it... I now know when to step away from a project (perhaps update certain fields of knowledge) and then dive back into it again later. Then there are also any number of 'side projects' that usually happen because a main project needs something... and whoa a project can sort of 'fragment' from there... and I lose site of the goal. In short... I have learned to always 'take a step back' and not get too sucked in unless it is a critical component for a larger project moving forward.

Over the course of the previous week that has all become rather paramount because I have been working with the 'Builder' LLM model to make various custom GPT models with the ChatGPT site. Talk about a frigging rabbit hole inside of a rabbit warren... adjacent to a fox den... in the shadow of a bear den... way down deep in the 'weirdness'... talk to the Builder!

I dunno if I can sum up what that is like but being able to both customize and shape a GPT model (LLM) gets pretty dang nuanced and the Builder's 'grasp of things' is rather evident. It can also be super finicky and misinterpret a small request... and make large accidental changes to the custom model... with no clear way to undo said changes... so being concise is key... and setting boundaries for its actions a necessity. Otherwise the term 'it works as intended' comes to mind.

Well, now that I got all that preamble out of the way I should move on to another topic lest I ramble on about all that jazz overly much. There is after all the Hive Bond Wagon to jump onto and mayhaps if I am lucky it will give me a ride to the 'Town Hall' that is coming up at the beginning of next month.

As far as the Hive bond 'idea' (in general goes) I really like it a lot and have always been a fan of the idea since HBD became stabilized and even more so... after having HBD stored in savings began returning a nice APR. That said though... I am in no way a fan of having such bonds be bound to anything on 'Layer 2' of the Hive blockchain.

I think the idea should be kicked around about Hive bonds that are Hive accounts like bond-000000, bond-000001 and so forth... which use (misuse) the 'Account Recovery' mechanism to transfer ownership between users. What I have been picturing in that scenario is having them act like 'Communities' (or even Badges) as far as how the account is setup to act as a custom account type. I am not even sure if any of that is feasible, secure enough or even a worthy path but simply one worth thinking over.

Another idea that keeps nagging at me while I am on the Hive Bond Wagon... is simply adjusting how 'HBD Savings' already works to include various withdrawal periods for the bonds with its associated APR. This way the bond is always 'one hundred percent' in the hands of the account owner and the APR can be voted on by Witnesses for each duration and APR pairing. To be clear here the current HBD Savings would remain the same and 'we' would just add more possibilities with which 'kind' of savings 'we' want to store our HBD in.

Financial things not being my strong suite I know to pay close attention to folks that have it as a strong suite... and the 'ideas' I am putting 'out there' are hopefully already a part of the conversations going on around these matters. Being a big fan of 'not fixing it if it is not broken' I think that introducing a bunch of new mechanics and whatnot 'for NFT(s)' either on the base layer (or Layer 2) just opens the door to more problems than it is worth.

There is also all the other sketchy factors of 'shady token-nomics' that comes along with that sentiment... and in short I firmly believe that keeping the bonds more in the realm of 'bearer bonds' (stored in HBD Savings) than NFT(s) (stored who knows where) is not only a more secure approach... but a more responsible one. The only other thing worth adding to that is that if the NFT route is taken... it should absolutely be a base layer NFT in my perspective... and maybe have a burn function instead of a transfer function... so that a new bond has the potential to be minted each time an old one is burned and has a 'transfer' flag.

Much of that stuff is really way outside my field of expertise but that said... so far I have yet to see a reasonable proposal for an implementation of a Hive Bond... and would rather see 'no implementation' than one that might turn out to be an albatross, a wrecking ball or some kind of slimy creeping decay... to an already working system.

I best move on from that topic and touch on my thoughts on a few different matters before I run out of the proverbial 'steam' for the evening. It is tricky enough to do the writing in such a way that it discourages commentary when I am not fatigued... and at this point it is getting late... and the fatigue sure is kicking in!

Anyways, one of my coolest projects of late has been working on that 'on chain file storage' (meaning on the base layer) and whoa has that been showing better results after a ton of refinement and experimentation. My point being, that ever since first beginning to tinker with it... I have been kind of 'groping in the dark' when it came to much of the 'on chain' mechanics and how I could make it all work... but then I built that Hive custom GPT model (to consult on the subject) and everything 'clicked' from there!

To be utterly clear here... my little project and the Hive bond thing are only related... in the way that I have been recently having to 'get to know' some of the Hive code (yeah I am generalizing) and doing so has been sparking some ideas. I am still unsure of just how worthwhile (or even economical) my 'on chain file storage' is and I have also yet to try to optimize its compression techniques.

As a side note the compression has been holding steady so far and I get roughly 3.5 megabytes worth of JSON files out of 6.1 megabytes of original files. It is not quite half the size of the original but it is pretty close and not bad at all in my perspective. It is worth noting that I have tried a wide variety of file types and it does not 'bat an eye' at anything... even squashfs files.

Tinkering with that project lead me to making that small hive-fc-linux setup comprised of 57 JSON files and while I have yet to deploy it to the blockchain... I feel absolutely confident that it can be stored, retrieved and loaded in a web browser if so desired... with very little further effort. I have no idea why that idea fascinates me so much... but there is something pretty epic about knowing that an ultra secure system could be 'freshly booted' from a trusted source on the fly... in only fifty-seven individual blockchain blocks.

The way that I see it is that the 'first step' would be to figure out a way to store some form of linux on chain... and then build on it (or just the concept) so that it gets LLM (and other AI technology) capabilities also 'baked into' the chain itself. By that stage of things I am sure that working out shared computational power, shared memory, shared GPU processing and such... would all be a matter of 'token-nomics' or more base layer APR on certain holdings etcetera.

In other words creating a decentralized immutable, uncensorable LLM (or even just 'cloud computing' via small linux instances) is not an un-worthy goal... and could open up other avenues for Hive in the space that it currently occupies. Even if much of that was happening via IPFS and 'Layer 2' services the principle remains the same: Leverage the current technology in such a way that if it gets banned or overly regulated... it can still operate via Hive... in full compliance with the Witness and Proposal system mechanics.

Well, the hour is growing even later and I have yet to really add everything into this entry that I feel compelled to... so I am at a bit of a dilemma. Considering this will all be a bear to edit to start with... I will simply link some of my 'latest and mayhaps greatest' projects if anyone wants to check them out. Beyond all the sloppy documentation I swear there are some 'interesting' ones!

On that note, I hope that everyone is doing well and has a nice time... or something like that.