Strikes me as way, way premature to be making such strident conclusions at this stage. We have barely even got to the point where orgs can run these models in-house on their own data, which is where the strongest perceived benefits have been proposed. I’m something of a skeptic myself. I’m not convinced LLMs are as disruptive as the hype says. But they’re not nothing , and it’s still early.
As someone studying AI and ML: they never were, this was pure hype cycle mania to juice the numbers, be it from existing incumbents trying to prop their stock prices (FB/META) after the massive losses they took from meta-verse, to Open AI teaming up with M$ in a deal that honestly probably let them have a longer life than they would have otherwise as these numbers confirm.
LLM's were the panacea that was going to help the VC world out of a post-COVID slump where massive layoffs, massive losses, down-rounds and overall worker discontent have been the norm since the bubble burst as the cheap money and absurd valuations were destroyed.
This is something that needs to be understood, LLMs are just relying on a form of training (which is really to say surveillance) scraping the internet and is trying to predict the next series of words based on weights on which the algorithm it's operation on deems more successful than the rest--this explains why hallucinations sound so confidently wrong, because someone, somewhere made that same mistake and it was written to sound convincing enough until it was proven otherwise. And nothing more, it cannot reason, it cannot think, it cannot intuit.
I hope this dispels the BS Musk drama about it being the cause for mankind's downfall, when in reality its just statistics; and often made on not very good percentage of accuracy mind you.
This speech by Meredith from Signal [0] underscores what the the real harm AI can pose right now and doesn't rely on some conjecture that somehow goes from a hallucinating LLM to Skynet; the further I go further into my studies (2nd year student) I'm finding out that AI/ML is just as hype-prone as NFT/crypto scams I'm also realizing that maybe cyber-security is far more valuable to what my interests within AI are than I had initially thought and I will likely pivot after this semester.
The last part is so poignant and if people want to face some harsh truths then look no further as it really speaks to what make this all possible.