It's summer and everything is BLOOM-ing...

Okay, so this isn't about flowers, although I did just do some gardening and turned a sad looking area into an area of small beauty thanks to some hard-work and some considerate flower purchases - and yes, this was done in the week of 35+degrees of heat - but it'll be fine it's the UK, it'll rain next week and also I'm rigging up a rainfall driven automated watering system that will allow for trickle feeding the plants with water - with no technology, well, an Arduino might get in on the equation soon.


The BLOOM that I am referring to, is in relation to A.I.  It's not your normal library of pre-trained models and it has stayed pretty hush-hush, considering it is the equivalent of a $5m system, available for FREE.

Yes, I said FREE and I meant FREE.

What's it all about then?

https://towardsdatascience.com/bloom-is-the-most-important-ai-model-of-the-decade-97f0f861e29f

(I'll snippet some text from that article below, if you're too lazy to follow the link)

Hang on, doesn't Google, META and OpenAI already do everything we "need" from an AI? Well, let's take a little look at what they give you:

Oh, they "keep" the good stuff for themselves, y'know, 'cos $$$$$$ - why offer this stuff out for "free" to people who can then use it & do stuff with it that would probably never have been thought of, to you know, "innovate" and create some things that might take us to the next level?  Well.... calm down there, you can't just spend millions making these models and then just give them away, y'know, what kind of world do you think we live it? (oh, that is a title for a whole new article on its own!)

Here's Alberto's description of what BLOOM is - yes, you'll have to click on the image below to read it (or, as I said earlier, go read the original article)


Unfortunately, Alberto is hooked up on the "Open-source-ness" of BLOOM and what that represents rather than telling you what BLOOM actually is.

However,. if you have some patience & time, please watch this YoutTube video, the guy does a really good job of giving a walk-through of BLOOM and explains the difference and how to use it:

You'll probably not watch the video (and I highly recommend you do!), here's a summary of the power & usage of BLOOM.


Basically, it is not your usual NLP engine, it is not your usual Chatbot engine, it is more of a PROMPT engine.  What does that mean?

"Why is that of any interest to me?", I hear you say, well, it's more about understanding what you are currently saying / doing / typing and then "predicting" what is going to come next.

Well, I can do that with NLP and a bit of coding logic, can't I?  well... yes...and...no...and...maybe, but this is not the point, read what BLOOM actually can do / try it out and then see how to fit it into your Use Case and then go and change the future.


--------------------------------------------------------------------------------------------------------------------------

Oh, and of course my prediction about the (ab)use of GitHub and other repo's seems to have come true - you can even type a natural language question about "code" and get some code snippets returned.  This is the ultimate in lazy programming, but, in reality it is no more than someone just typing it into a Google / DuckDuckGo / Bing search and instead of having to click through several links to assess the code snippets and determine which is good / bad, it'll just do that for you.  Again, I see this as yet another negative to the true software programmer (who wouldn't be using this anyway), but it will be used by unskilled people who have absolutely no skill / training (therefore lowest of the low paid) to write code for big corporations who outsource their development and want it built in next to no time.  Okay, harsh and a bit unfair (what's new from me?), but ultimately true.  This is getting close to the final nails in the coffin of coding.  Which is a real shame.  I'm putting away the soap-box as I've covered my thoughts on this issue before and this article is about BLOOM and all of its greatness.

--------------------------------------------------------------------------------------------------------------------------


https://huggingface.co/bigscience/bloom



SHOW ME SOME EXAMPLES, DAMN YOU!

Okay, if you didn't watch the VIDEO from above (that has some great examples) I'm not going to duplicate them here, but point you at someone else who does a really good deep-dive




You can DOWNLOAD and run BLOOM on your own hardware - which needs to have about 3-400GB of RAM and needs 48-64 cores.... (okay and a TON of GPUs!) now, that is interesting, I have a server/box with 1TB of RAM and 128vCPUs available to use in my current work environment.... looks like I might be adding this to my R&D checklist to see what we can do with this.... if anything...


Because I love a bit of JavaScript: here's a repo to download.......


Okay, if you've got this far - you may be wondering? why is this a gamechanger?  well, when you SEARCH (Google / DuckDuckGo / Bing / etc...) you are typing in some text and nowadays the "engine" behind the scenes actually does some smartness to tokenize the text and "understand" what it is you are typing, in the old days it just used to do "keyword bingo" against the content of web-sites and provide you with a list of those web-sites.

With BLOOM you can now start to ask questions about ANYTHING and get some reasonably intelligent response - think of Star Trek when they say, "Hey Computer....XYZ", that's what this is now enabling you to do.  You may argue that Alexa / Google Home does that for me already...well, yes & no (again), using BLOOM you are not providing data to big corporations to make more $$$$ off of you.

Okay, okay - I promised EXAMPLES, here they are: 

--------------------------------------------------------------------------------------------------------------------------

UPDATE: and here's what happens when things go wrong with this type of technology.  oh dear.

--------------------------------------------------------------------------------------------------------------------------


Comments