Resisting AI Hype
A lot of academics are jumping on the "resist AI" bandwagon. When we finish highlighting that LLMs are driven by statistics ("stochastic parrots"...is there any other kind?) and "resisting", then what?
In the end, it doesn't matter at all whether this is intelligent, or a "stochastic parrot", or just another computer program. It's useful, and that's what people are keying on. No amount of discussion about "automated plagiarism" will make it less useful.
I was rebuilding my Framework laptop yesterday evening, and was showing someone that hadn't spent time on the inside of a computer the laptop's various components: CPU, memory, SSD, wireless card, battery, cooling, display, webcam, etc. I remembered from building desktops back in the 90s and 00s that the CPU was connected to memory via a north bridge and connected to disk via a south bridge. In my Framework laptop, I saw some bridge, and the person I was with asked about it. I vaguely recalled that one of them isn't a discrete component anymore, but wasn't sure. I ended up searching on DDG, reading a couple of different pages, and ultimately concluded we were looking at the southbridge (one word!).
Today, I saw this "resist" piece, and thought "I just researched this chip question with DDG for 3 minutes. What would ChatGPT say?" I don't tend to think that LLMs are great as "search engines", but there could be real utility in having the LLM "read the docs for you" so it can answer questions. So here was my first attempt with ChatGPT (note my incorrect usage with two words rather than one):
What's the function of the north bridge and south bridge on a motherboard?The northbridge and southbridge are chips on a computer motherboard that control communication between different subsystems. The northbridge generally controls communication between the CPU, memory, and graphics card, while the southbridge controls communication between peripherals such as the audio, storage, and network interfaces. In modern system, northbridge functionality has been integrated into the CPU itself, known as the "chipset".
ChatGPT corrected my usage, explained their function (correctly), and even provided extra detail that the northbridge has been integrated into the CPU directly. That response is straight-up useful, and the fact that it could be "confidently wrong" makes it...the same as every other source of information I use. I constantly have to deal with people and websites saying things that aren't true, so raising that ChatGPT can be wrong is completely uninteresting to me: everything can be wrong, and ChatGPT seems to get it right more than most humans I know being asked the same exact questions.
So I'm not going to resist. The excitement about ChatGPT is not because people are buying into "hype", it's just folks using the tool and finding it useful. Folks that learn to use tools well tend to be more productive than those that don't, so trying and evaluating new tools, espdcially poweeful ones, seems to be a great use of time.