Tony is a Geek

 ever wondered what goes in INSIDE a Large Language Model (LLM)? Back in the "day" LOL okay, a couple of years ago, this was mapped out and us early adopters had this worked out mostly on paper - but this is so cool to see / zoom in & out and trace through what is happening at each step of the way through an LLM to get from initial prompt through to output: https://bbycroft.net/llm If you press [SPACE] you can step through in a guided tour: This is really useful & potentially helpful for people to understand what happens behind the Wizard of Oz curtain

Well, why not?! This seems to be the current (okay, it's been out for a while) trend, where non-coding people have "an idea" and then just want to describe that to a GPT and then get that to autom-magically create an application for apple / android & web.  Then sit back and rack in the money.  Sounds idyllic & all very cyberpunk.  But, how close is the reality to the expectation? I thought I'd give it an investigation look using RORK.com and BASE44.com Here is what I found out from an initial attempt at just using a prompt description. Now, you may (or may not) notice that I started this at 09:41, it seemed to make a good and quick start, however, I think either I picked a busy time or it was a bit more of a complex tasks than I was led to believe. I got a little bored and started to click around to see what the "logs" were saying: I genuinely thought I had crashed it. I thought I'd have a nosy around at the code it generated. Although it is Reac...

During the past, well, let's think, 10+ years, I've been involved in Machine Learning projects (okay, they call it "AI" now, but that's an argument for a different time) that require lots of data to be ingested / used to train the Machine Learning model. For instance, if it is text, we need lots of snippets of text that can then be categorised, so that the model "knows" that the 10 words grouped together in that context have a "meaning" and that meaning is labelled - this helps with questioning later on. It also helped to extract out the entities and relationships between the wording to give more context.  Here's a simple example using the spaCy tooling to give you an example: https://www.labellerr.com/blog/image-annotation-services-and-data-labeling-for-ai-models/ If it is imagery, we need lots of images with the segmented parts identified, usually a bounding box(?) to identify the elements / objects inside the image, so, again, this can b...

Fancy trying out the new Nano Banana image manipulation tool? Well, originally I wasn't too bothered and then I tried it out with some images as starters and then got more creative with the prompts and was then actually very surprised at just how good it was at making fake imagery look really well done. You can literally just fake and make any imagery / scene that you can think of just from a base image. This is good, but, yes, can also be quite bad for obvious reasons - yes, there is a T&Cs violation restriction, however, it won't stop some weirdo getting hold of an online photo image of you and then including things in the image that were never there, like other people, etc... I'll show a few examples later, maybe. However, as this guy on YouTube shows, it can really be useful if you're working with your own imagery and want to be more creative: If you don't want to watch the video, I can give you a quick walk through of how to test out Nano Banana for free! Y...