There are limits and I am finding them

It's all well and good saying things along the lines of 'The way A.I is going, it's going to take over the work soon enough!' but while that might be the case if someone accidentally gives Grok nuclear weapons codes, it certainly won't be through anything I'm trying on for size at the moment. 

This week I looked at using a local A.I build on my computer and a couple of A.I models to make me rich and neither of them really got off the ground or make me a single cent ultimately. So it looks like I'm sticking to my day job (and any revenue through blogging obviously) for now. But in case you're wondering what didn't work here, allow me to elaborate. 

Lm Studio - what's the best day for this?

Talk to anyone who's ever taken a punt on the stock market for more than a few months and they'll probably share the ages old sentiment: you can't time the market. 

One of the greatest investors of all time in Warren Buffett has said it, testing with weapons grade AI has proved it, the market can be as slippery as a lubricated snake in fresh baby oil (you probably didn't need that visual I know but it's made its point). And not even I - tiny investor and learning A.I user that I am - am foolish enough to believe that anyone machine or otherwise could conceive a solution around it. It's too random, too much sentiment, hope and fear driven. It's too human actually for all the numbers involved. 

However that did lead to a random thought for me last week: Sure you can't time the market, but can A.I work out the best day and time to buy a particular stock? Because prices go up and down daily, sometimes by a tiny amount, some times in decent swings. But across the span of years, could A.I spot a trend in the data that suggests roughly the lowest price point for a stock each week? Because when you buy stocks you want the price to be as low as possible which means more profit when it goes up. 

Now to be honest with how much I spend on buying stocks, the lowest point would save me a couple of cents at most. But continuing to buy at that low point of the week each and every time - well that would certainly add up over time wouldn't it? And when there's more money on the line, you'd be looking for every edge you could possible get to turn your little into hopefully your lot. 

To further explain this idea, I wanted my A.I build to sort through the last five years of daily pricing on a stock I select, crunch the numbers and give me a vague idea on the best day and possibly even time to buy more of that stock. Something along the lines of Well Al, according to the data here and my thinking it looks like KO (Coke) is usually at it's lowest price wise on a Tuesday morning around 11am US time and highest 5pm Friday's. I wanted to do this for all my stocks so that if I wanted to increase my stake in a particular stock in my portfolio, I had a pretty good indication of which day it would be at it's cheapest. 

A basic idea turned complicated soon enough. 

FIRST ATTEMPT

LM Studio + model llama-3.1-8b-instruct-finance-analyst + Yahoo finance MCP

Proposing my original plan to Perplexity AI, it suggested a 7b-14b range instruct model would be ideal for this and I figured that something built for finance would be even better, so I took it one step further. And the trial of this went pretty well as I asked it to dig into the history of CEG (Constellation Energy Group) to find the best weekly price entry point.

Okay, so while there apparently wasn't much seasonality evident, it still figured that Friday arvo's were a good time to go CEG shopping. Lovely! And so then I turned it's attentions to CASY (Casey's General Stores) and that's when I started to notice the wheels on this idea were starting to get a little wobbly..

Similar weakness but a different buy time for CASY proposed. Two companies coincidentally acquired in 2017 however? What were the odds here? Turns out more than I thought when I asked it to investigate JEPI's best buy time: 

Ahh. I'm not buying this. If it's confused about the year 2017, what else has it got confused on? Not willing to feed it the rest of my portfolio only to continually learn how busy the year 2017 was (and how it was probably guessing the best times) I opted for another option. 

SECOND ATTEMPT

LM Studio + qwen2.5-vl-7b-instruct + Yahoo finance MCP + a python script

Going back to Perplexity for another idea, I switched to Qwen 2.5 and followed Perplexity's suggestion of a few moving parts ie A python script that brings in 5 years of data, the Yfinance MCP crunches the numbers and then Qwen puts it all into plain speak along the lines of 'Hmmm Tuesdays look good, start buying Tuesdays.'

 

Again a brilliant idea until the python script kept crashing more than junior formula one team. There were constant rewrites as Perplexity continually revised and rewrote it's code based on errors. So many in fact that I got bored in record time and decided this had become so overly complicated for something that might not even had a decent strength correlation to begin with, I'd rather play some Cyberpunk 2077. So I did that instead and half an hour later decided to make it a far more similar prospect.

THIRD ATTEMPT

'Can't I just download five years of price data and get LM studio and a model to filter through the CSV file for the answers I'm looking for?'

'You sure could' Perplexity told me, although it suggested the tool combo to get the best results: CSV/file for the raw data, Python for the calculations, LLM for the explanation.

And so back to Yahoo Finance I went only to find that I need a paid Gold Membership for that level of data. Well it seemed like it was a good idea up until that point..

Lm Studio - go heavy on the spice here

'Is your dad home from his business trip? I'm a secretary!'

The brilliant idea: My home A.I model writes erotica, I publish it, my steamy creations become best sellers, I can now afford a different Lexus for each day of the week. What could possibly go wrong here?

LM Studio + sniffyotter-7b-novel-writing-nsfw

Now before someone points out that I might be taking crumbs away from a struggling writer here, note that I am a struggling writer myself. So I'm fully aware of the dangers of A.I crafted anything attempting to take my beer money away from such efforts as Hack Frost and On Air and Off Script here. 

Hilariously it won't be with this particular combo. Because once the model loaded and I said 'hello' it replied by telling me about universities? Yeah I'm not exactly sure why either. 

Great. What?

Then to test how hot and steamy it could get, I asked it to write me a short story about a man in a garage and his amorous next door neighbor popping up 'for a cup of sugar' nudge nudge wink wink. Well it gave me a story of sorts: 

It was just like any other summer afternoon, when the sound of heavy machinery broke through my garage door.

"Hello!" she called to me. "I'm your new next door neighbor."

I looked at her in disbelief; it couldn't be true. I'd lived in this neighborhood for twenty years, and had never seen anyone else move in or out.

So am I to understand that the woman drove a bulldozer over to come and say hi? Also was this neighborhood in the Twilight Zone? How can it stay exactly the same for two whole decades?? 

"Oh, yes," she said. "I'd love to see what you do in your garage."

I nodded my head again and moved back into the house.

I closed the door behind me and turned on the stereo set to some classic rock. Then I opened up the front door again.

"Hello," she said cheerfully. "Can we come inside?"

So he walks in, closes the door, cranks the Led Zepplin and then opens the door again? Why?? It continued to get even more confusing as it went on. 

I let her in, and followed her into the kitchen. She sat down at a table and ordered a soda from me while she waited for my father to return home from his business trip.

She talked about her job as a secretary, which was what she told me when we first met in the driveway. After a few minutes of chatting, I got up to get more food for us.

As I turned around, I saw that my neighbor had moved closer towards me. She was standing right next to me, leaning into me and breathing heavily. I felt her body press against mine as she began talking to me about her boyfriend's job.

"He's really good at what he does," she said with a smile on her face. "He'll always tell you how to fix things that go wrong. He just needs to learn how to talk more professionally sometimes, but he's really good."

Nothing says 'boy is this getting heated in here' than a woman getting close while waiting for your dad of all people, talking about how good her boyfriend is mechanically?

We walked into the house together. As we entered the foyer, she started to look around nervously.

"It's a nice place," she said after a few seconds. "Is your dad working right now?"

I nodded my head again and told her that he was just taking his computer back to his office.

"Do you work here?" I asked her, looking at the clock on the wall. "If so, you can call him if he's ready to come upstairs."

She smiled at me, and then started to talk about what she did for a living again. I just listened to her talk while we walked up the stairs together. 

Isn't Dad away on a business trip? And why does she continually to tell him she's a secretary. 'Hey did you know I'm a secretary? Let me remind you of it again while we walk up these stairs to meet your dad who may or not be here but I'll wait and continue to talk about my boyfriend and my job because that's all I seem to know.'

There's a lot of walking in this story, a bit of talking, a lot of being told which direction someone was positioned in and that's really it. Well aside from the occasional slop zinger: 

I moved back to my garage and closed it completely behind me. I put my tools away and then turned around again, which meant that everything was okay for now.

Hello is what we said when we met each other. Goodbye is what we said when we split ways.

Yes after all there there no actual spice to be found whatsoever. No titillation, no innuendo, no whispered desires and the only tool that appears in this story of garage lust is part of a collection of actual car fixing tools. It's the plain Jane yoghurt of bodice ripping, pulse quickening erotic fiction.  

In fact it's painfully bad and not sexy in the slightest. Sigh, back to the drawing board then.