LLMs are like nuclear weapons:
you wouldn’t really need them if no one else had them.
I wanted to try some LLMs for personal use — wouldn’t a (robotic) personal assistant (for the poor) be nice? — so, just to understand where those things may live, what they may eat, and where they may come from, oh, and also, what the hell they can do for me in the first place…
…before writing
…yeah, I won’t be talking much more about my reasoning for my personal dis/interest with LLMs resulting from my experiments, I’m just sharing some context and some technical details thereof.
Now, in an ideal world, where people talk to people — and their personal robots talk to each other — LLMs would indeed be rather niche, however, with human civilisation of our time, personal “memory-enabled” LLM instances will be enormously useful — to deal with the borderline-force-fed content, generated by by all the others’ LLMs, and MLMs, and BDSLMs.
Remember when your e-mail inbox used to contain messages, mostly generated by natural human intelligence?
Gross.
With all the content-generating platforms and services (which is all of IT industry nowadays, basically), competing for your precious attention, who is going to save your time and decency, while not letting you miss out on things? ghostbusters? what is that entity you trust with protecting your personal world of information — as well as leveraging your personal* information, not dissimilar to virus protection — coupled with file system analysis — for your PC? do you recognise the similarities of the new threat and the old threat? Do you have an idea on the new threat protection for you and your family, or in your mind, that is just another function of your imaginary friend, the government?
*information that you legally and physically own, and thus, which can be deliberately and securely used for fun and profit
Robots not only play chess better than best humans, they can now con average people better than average con men. And I see only feeble, impotent and ultimately pointless attempts to regulate this by the governments.
With self servers— and a little coöperation on personal automation — I believe essential content streams — and bureaucratic routines — can be tamed, and the systematic problems of
Too Much Information
as well as
Too Much Interaction
can successfully be tackled with the help of personal LLMs.
So, your friends against those evil corporate LLMs are those software developers who — instead of creating another semi-pseudo-functional service — for BIG profits (like VPN for dummies… give me a break… actually… wait for it… stay tuned) — passionately work on their digital environment, and I mean, not just following trends and trying some new shiny things, like email clients, every couple of months, not even just configuring and interconnecting tools they use, I mean, really work on it nerdy-garage-style, with ETL processes and rule engines. These developers, if monetarily motivated, of course, will polish their (shared) information management components, including personal LLMs, and having also organised into a monetarily motivated distributed community, they — through your trusted closest IT friend connected to them — can provide truly secure (lasting) solution for the aggravated information manipulation threat. Remember rumours of antivirus companies manufacturing viruses? With the new threat, it’s not even funny, how the problem is not suitable for regular commercial solutions (like services of companies). So, this matter will cost you… either way, but with a developer community solution as described (and within a self server network, preferably) the cost will be close to optimal.
Imagine any and all public or private information feeds of yours digested by your personal LLM instance into regular concise summaries, or your LLM instance filling in trivial interaction with your individual bureaucracy (probably using LLMs themselves) for you… I would like to work for a project like that…
So, anyway, having heard about OpenAI, and ChatGPT, and to be sure, tried to use it for work…
…now I wanted to try running an instance of a similar LLM for personal use (with all my private data, individual automation etc), and none the wiser, chose a (quite popular) self-proclaimed alternative to OpenAI, which is LocalAI.
In order to do that, after only a little bit of research, I used the following docker-compose configuration:
to play around (in my local browser, for instance, at http://localhost:16623), like…
(the cryptic self-server commands serve as a teaser for the convenience demo of the system)
If you want to try… I mean… just to dip your toes in LLMs, this is one of the simplest ways, but also, a quite powerful one — with a kind of self server integration (again, I’m teasing for the coming demo) — this is not very far from what you actually need to connect the power of LLMs to a personal or multi-user system of yours (see OpenAI/LocalAI API as well as model galleries etc).
Happy using them! Don’t let them use you…
p.s. A better — though less clickable — title for this would be:
“robots.txt” for… me, please
— so, you see, I need LLMs to reach different llama populations… *wink*
