Samsung Restricts Use of AI After Engineer Leaked Proprietary Source Code on ChatGPT
Now in addition to using personal email, Chat GPT is now the latest threat companies have to be aware of where their corporate secrets could be inadvertently leaked by employees:
Samsung Electronics has experienced continued data leaks, amid the intensifying global competition for supremacy in the semiconductor sector, according to industry officials, Wednesday.
Korea Times
Samsung Electronics said its Device Solutions (DS) division in charge of chip production dismissed an engineer recently who was found last month to have sent dozens of emails containing proprietary data to private email accounts.
The company also asked for police to investigate the case.
“Through disciplinary measures and legal actions, we will be tough on coping with this issue,” a Samsung Electronics spokesman said.
In March, another Samsung Electronics engineer mishandled confidential company data by uploading source code to ChatGPT. This case led the company to restrict its employees from using the artificial intelligence-based chatbot during work.
You can read more at the link.
Something is not right about this story.
Uploading source code to ChatGPT is… pointless.
There is no reason to do that.
You know, what’s sad is that the US was once THE most innovative country on the planet. But now they are competing with China in a race to steal and coerce technology from other countries. It’s not exactly great, being thought of an equivalent to China in terms of bad behavior.
https://www.ft.com/content/9e72a96f-5d92-460f-a154-0715c343e7c9
“You are an old man who thinks in terms of nations and peoples. There are no nations. There are no peoples. There are no Russians. There are no Arabs. There are no third worlds. There is no West. There is only one holistic system of systems, one vast and immane, interwoven, interacting, multivariate, multinational dominion of dollars. Petro-dollars, electro-dollars, multi-dollars, reichmarks, rins, rubles, pounds, and shekels.”
ChatGPT can write code, albeit very basic code.
However the coder can take that code and improve on it, which means ChatGPT can improve that coder’s productivity because the coder doesn’t have to start from scratch.
OpenAI, the creators of ChatGPT, has a module, where for a fee, coders can play around with downgraded versions of their GPT model.
It seems like the Samsung employee was trying to train a GPT model to write code more to his liking and uploaded the proprietary code in question as a dataset to the above mentioned module.
As for the guy who sent information to his personal email. Seems like he was getting ready to move to a more high paying job in China.
Chinese chip makers are luring away Samsung and SK semiconductor workers by offering way higher pay, including paying the first year wages in one go, and offering to pay for the children’s education in International Schools, etc.
In return, the Chinese chip makers are know to suggest to the candidates, that they will get better treatment, if they can bring along information.
For unhappy Samsung and SK semiconductor workers especially those who are in their 40s and 50s, those offers are hard to resist.
“It seems like the Samsung employee was trying to train a GPT model to write code more to his liking and uploaded the proprietary code in question as a dataset to the above mentioned module.”
I thought of that… but couldn’t think of any scenario where I would upload my propriatary code (which must be complex and critical) and hope ChatGPT could improve it.
ChatGPT is amazing at basic code… perfectly Pythonic.
My coding has improved massively because I ask ChatGPT how to do something simple and it gives me perfect code… far better than the usual stuff I have to cut and paste from some boogerpicker’s website hosted in mom’s basement that has the right idea but need modified and cleaned up.
It fails at a lot of complex tasks. When working together, I find it best to break my problem down into understandable steps and then have ChatGPT write code for each step that I assemble. It has good memeory per session so it does well at remembering… “the input for this will be the output from the previous code you wrote. Please keep the data format the same.”
There is an LLM that was trained ONLY on Python from Github. I have more 4090s coming in and will set up a dedicated server running this in August and share it with a group. As soon as they become dependent, I will charge. This runs on a P40 but slooooow.
You can run ChatGPT ~3.5 on P40s but I got around 1.2 seconds per word. Deeeaath.
CH, that sounds just like John Lennon’s song Imagine.
And he got shot dead after singing that…
…in New York City…