Something pretty major happened this week in the AI world, and I think it's worth sharing - especially if you run a small business or you're one of my Microsoft Access students.

OpenAI just released two new models, GPT-OSS-20B and GPT-OSS-120B. And unlike the usual ChatGPT stuff you use online, these models are open-weight. That means you can download them and run them entirely offline on your own computer. No internet connection required, no API token costs, and no sending your data off to some mystery server in the cloud.

For the first time since GPT-2 way back in 2019, OpenAI is giving developers something we can actually poke around in.

Now, the big 120B model is a beast. It takes high-end hardware to run (we're talking 128GB of RAM and beefy GPUs). But the 20B model? That one's much more accessible. It runs fine on a decent PC with about 16GB of memory. So yeah - realistic for a lot of small businesses.

What can you do with it? Well, just like ChatGPT, it can help you with tasks like:

- Spelling and grammar checks
- Writing customer service emails
- Summarizing (1)
- Generating basic code (like SQL or VBA)

I find ChatGPT is best at doing things I already know how to do - it just saves me the time. Take writing, for example. I know grammar, punctuation, structure... but instead of spending an hour nitpicking every sentence in a Captain's Log post, I can just dump my thoughts on the page and have it clean them up in minutes. Same with code. I know how to write recordsets - I've been doing that for decades - but if I want a quick loop that pulls a few fields and pops up a message box, it's faster to just say, "Give me the VBA for this." I'm not asking it to invent something new - I'm asking it to save me the typing. And it does.

Now, is the 20B version as smart as GPT-4-turbo or whatever the flavor-of-the-week is online? No. But I don't really use those "super advanced" features anyway. I stick to the basics. And for the kind of tasks I just mentioned, the 20B model is more than good enough.

The big deal here is privacy. If you're working with customer data, employee info, or anything remotely sensitive, you probably don't want to send that over the web to some third-party AI. Running a model locally solves that. It keeps everything in-house. Literally on your machine.

Now, to be clear - I haven't done this yet myself. I've read through the setup instructions and it looks simple enough. Tools are available to make it easy to run these models locally. But before I dive in and start experimenting - or even think about building a course around it - I want to know: Would you be interested in this?

If you're someone who uses Access or Excel at work, and you've been curious about AI but concerned about security, this might be a perfect middle ground. You get the benefits of ChatGPT without handing over your data. Plus, you wouldn't have to spend money on OpenAI tokens. I can say from experience that I ran all of my video course transcripts through OpenAI to get the polished handbook text, and that was not cheap. If I had a tool like this, I could have saved myself a couple of bucks.

Let me know. Drop me a comment below. If enough people are interested, I'll start working on a step-by-step guide for getting a local AI assistant set up in your office and hooking it in to your Access database, like I've shown in my OpenAI videos and AI Builder.

It could be a game-changer. Or it could just be a cool toy. Either way, I think it's worth exploring.

LLAP
RR

(1) One thing I'm definitely guilty of myself is using ChatGPT to summarize long-winded stuff I don't have the patience to get through. For example, I can't stand watching most other people's YouTube videos - even if the topic is something I care about. Bad audio, thick accents, poor delivery... I just don't have the tolerance. So I'll grab the transcript, dump it into ChatGPT, and say "Summarize this for me." Same thing with long, rambling emails. I'll get a wall of text from someone and just ask, "What is this person trying to say?" It's not that I can't do it myself - I just don't want to. It saves time and lets me stay focused on the stuff that actually matters.