Aardvark Daily

Please visit the sponsor!

Goodbye encyclopedia, hello local LLM

I come from a time when every household had an encyclopedia on the bookshelf.

Sometimes these were just a single bulky volume of facts and information, other times (in richer households) it could be a multi-volume, leather-bound array of books with a new addendum provided every year.

Many households went into debt to buy these encyclopedias because they were considered essential to ensuring that your kids had access to the educational resources that would help them attain good grades in school.

Names such as "The Encyclopedia Britanica", "Funk and Wagnalls" and the "World Book Encyclopedia" were often associated with such weighty tomes.

Many a fondue party in the 1970s saw arguments that were eventually resolved by resorting to the household's encyclopedia.

These were the pre-CDROM and pre-internet days and life was fun.

Of course, as hinted above, huge masses of pulped tree flesh with inky markings eventually gave way to computers and CDROMS. Now we had disk-based encyclopedias such as Microsoft Encarta to deliver a wealth of facts and we even got tiny postage-stamp sized movies as well as text and still images.

The lifespan of the disk-based encyclopedia was surprisingly short however, because less than a decade later the internet arrived.

Now, instead of relying on something that was likely already out of date before you even opened the package, people had access to a constantly updated wealth of knowledge and information -- even if there were sometimes concerns over the veracity of what was published online.

Sites such as Wikipedia popped up and attempted to become the online equivalent of those old printed volumes but once again, since they were "user created", there were often challenges to the accuracy and objectivity of some of the material posted there.

Google's search engine eventually became the gateway to all things internet so I guess it gradually eclipsed any one site for providing encyclopedic information.

Let's jump forward to the present day and things are changing again.

It's no longer Google's ability to provide us with a list of links to relevant information that's delivering the information we seek, it's artificial intelligence.

Even a Google search will spit out an AI summary ahead of any relevant links so now it's probably fair to say that AI has replaced all those printed volumes that cost us so much back in the 1960s.

However, what if you don't want to rely on Google? What if you want "off grid" access to an encyclopedic wealth of knowledge and information?

Simple, just run a local LLM and a significant amount of the internet's entire knowledge will become resident on your AI device, without so much as a Wifi connection in sight.

My Raspberry Pi is now my first port of call when I want to know something. I simply type in a query and almost every time I'm blown away that it comes back with a useful response. It's almost as if they really have condensed down the entire internet, and many books, papers, etc., into format so condensed that it fits on a single microSD card.

Even better, it often (but not always) provides a useful insight into the data being presented and allows me to interact in a conversational fashion that helps me drill right through to the most relevant information. In essence, it is my research assistant.

Unfortunately, just like those printed and disk-based encyclopedias of old, its information is fixed in time. It knows nothing of things or additional knowledge that has been acquired subsiquent to the creation of its token-set. I think this may soon change however.

I'm pretty sure we're going to see locally hosted LLMs that can either auto-update or update on demand, so as to remain current. In fact, there's probably a pretty good business model here for subscription-based local LLM updates for those who want/need that.

It still boggles my mind to think that I now have a device which is little bigger than a pack of cards but which gives me local access to such a wealth of knowledge and information.

My prediction is that within the near future, we'll see smartphones with their own built in local LLM rather than relying solely on cloud-based LLMs. The fact that this can be done on a paltry Raspberry Pi means it could also be done even better on flagship smartphones with their super-fast multi-core processors, inbuilt NPUs and relative abundance of RAM.

Isn't the future great -- in fact it's so good that I fully intend to spend the rest of my life there.

Carpe Diem folks!

Please visit the sponsor!

Here is a PERMANENT link to this column

Rank This Aardvark Page

Change Font