I AM NOT A BOT Episode 6: “RAM Space”
I AM NOT A BOT Episode 6: “RAM Space”
OPNSense is a firewall and routing software. And it’s a beast.
“Along with acting as a firewall, it has traffic shaping, load balancing, and virtual private network capabilities,” my gal from Nepal reads from the online manual.
She and I are in a breakout room. We are each waiting for the 4GB download to finish. I fear this one will break my poor Dell. I was pleasantly surprised last week when my laptop survived the installation of Ubuntu on top of my Oracle VirtualBox. But this could be it for this grand ol’ dame.
To be fair, when we applied to this class, we were told we needed a machine with over 500GB of free space and 16 GB of RAM. I had under 300GB of available space and my machine had only 8GB of RAM. Still –
“-- did you see what ZaeZae posted last night,” she asks with her distinct accent.
The one about us spending more time in breakout rooms than in actual class?
“I can’t believe he actually calculated it based on last week’s schedule.”
Just a few minutes earlier, she told me about Nepali’s 11 phonologically distinctive vowels, including 6 oral vowels and 5 nasal vowels.
A nasal vowel is a vowel that is produced with a lowering of the soft palate so that the air flow escapes through the nose and the mouth simultaneously. By contrast, oral vowels are produced without nasalization.
“What’s your download estimate now?”
I’m about halfway done, I lie to her. I don’t know why I didn’t simply say I have no idea. That I only see the spinning wheel of death above the OPNSense logo.
“I think I’m almost done. I’m past 3GB it says. But you never know.”
I change the topic to the major topic of late. ChatGPT.
I don’t tell her that I loaded the first few pages of my dad’s journal into the app just to see how it would talk back to me. I don’t tell her that I was hoping the app could help me talk to my dad again.
“I finally gave in and used it for Personal Development. Just like everyone else is doing. I used it for my resume cover letter assignment.”
And what did the AI write?
“Some very generic return on how I can’t wait to be an analyst so I could make use my creative problem solving skills … blahblahblah.”
I don’t know why, but hearing someone with a foreign accent say “blahblahblah” is funny to me.
She shows her screen. She shows her ChatGPT result.
Cover letters are meant to be generic, I say. There’s only a few ways you can say thank you for your time.
“Or please give me a job already.”
I see her write the following,
WRITE AN SQL QUERY TO REPORT THE CAPTIAL GAINS AND LOSSES FOR SELECT STOCKS
What are you doing?
“Seeing if this AI can do my finances.”
That’s right, I think to myself. She was an accountant back in Nepal. She hated it, but it was one of the few professional opportunities available to women at the time.
She writes the following query:
SELECT
stock_name,
SUM(
CASE
WHEN operation = "Buy" THEN -1*price
ELSE price
END
)
AS capital_gain_loss
FROM Stocks
GROUP BY stock_name
She compares her write up to ChatGPT’s.
“Ha. It failed. Mine wasn’t the greatest of codes, but it works.”
She explains how ChatGPT made the mistake of joining one buy with every future sell. Not just the corresponding one.
“It’s solution only works if there is just one pair of Buy/Sell operations.”
All that goes over my head, I don’t say to her. Instead of trying to understand all that goober talk, I tell her to simply ask the following:
DO YOU THINK YOU COULD REPLACE A DATA ANALYST?
“Ha. Check out its response.”
As a language model, I am able to provide information and assist with tasks such as generating SQL queries, but I do not have the ability to perform the role of a data analyst. A data analyst is a professional who is responsible for collecting, organizing, and analyzing data to inform business decisions. This role typically requires as strong understanding of statistical and analytical technique, as well as domain-specific knowledge related to the industry in which the data analyst is working. Language models like me can provide support and assistance, but we are not able to fully replace the skills and expertise of a trained and experienced data analyst.
“Well that’s a relief.”
For now, I say.
“For now?”
I remind her about ChatGPT’s Supervised Fine Tuning (SFT) Model. That its intelligence is just like our intelligence. It is based on interaction. It can grow if given a wealth of two-way conversation.
Alpha. Omega.
In my head, I see the daily quote I got earlier from my dad, “There's not a word yet for old friends who've just met.”
That quote and the daily quotes I get from my dad are actually generated by ChatGPT based on the few pages of his diary I inputted into the app.
Every morning I ask ChatGPT, What advice does my dad have for me today?
ChatGPT responds the way it thinks my dad would, replying back with wisdom quips such as, “However rare true love may be, it is less so than true friendship.”
And every time I respond with, Thank you. That sounds just about right.
My every time input provides what the SFT Model calls a Reward or a Scaler Value. The reward model is required in order to leverage Reinforcement Learning in which a model learns to produce outputs to maximize its reward.
“That’s fascinating. This is all fascinating, don’t you think? Two months ago I wouldn’t have imagined me ever talking about stuff like Reinforcement Learning from Human Feedback. I was sunk in a cubicle, using Windows 7 Blackcomb technology to reconcile financial and operational records. And today I’m teaching a computer how to think.”
Only in an Infosec starved world could a gal from Nepal and a bartender from –
-- just then my Zoom link crashes.
I look around and notice that OPNSense has finally uploaded and is now attempting to open.
All my other program worlds start to collapse as well, one by one.
Discord … Slack … ChatGPT …
Going … Going … Gone.
OPNSense is sapping all the available RAM.
I am left alone. Disconnected.
ZaeZae and his jokes … Edamame and his know-how … my father and his …. my father …
Going … going …
