Yeah, I think the biggest cause was actually not groundbreaking stuff, but our understanding of microorganisms and our changes in hygiene (all summarized in better education).
The fact we treat many diseases now helps, but I think it makes for a much smaller percentage.
It is unpredictable because there are so many permutations
Actually LLMs are unpredictable not only because the space of possible outputs (combinatorics) is huge, though that also doesn’t help us understand them.
Like there might be an astronomical number of different proteins but biophysics might be able to make somewhat accurate predictions based on the properties we know (even if it requires careful testing in the real thing).
For example, it might be tempting to calculate the tokens associations somehow and kinda create a function mapping what happens when you add this or that value in the input to at least estimate what the result would be.
But what happens with LLMs is changing one token in a prompt produces a sometimes disproportionate or unintuitive change in the result, because it can be amplified or dampened depending on the organization of the internal layers.
And even if the model’s internal probability distribution were perfectly understood, its sampling step (top-k, nucleus sampling, temperature scaling) adds another layer of unpredictability.
So while the process is deterministic in principle, it’s not calculable in a tractable sense—like weather prediction.
Not just that, LLMs behavior is unpredictable. Maybe it answers correctly to a phrase. Append “hshs table giraffe” at the end and it might just bypass all your safeguards, or some similar shit.
I doubt that LLM queries honestly leak all that much more information.
I don’t doubt it. Search engine queries are just like a one time short query and then you just click on what seems like what you’re looking for and continue from there.
LLMs conversations, on the other hand, go much deeper, exactly to what you’re interested in, clarifications etc.
It’s very possible for someone to appear fine in public while struggling privately. The family can’t be blamed for not realizing what was happening.
The bigger issue is that LLMs were released without sufficient safeguards. They were rushed to market to attract investment before their risks were understood.
It’s worth remembering that Google and Facebook already had systems comparable to ChatGPT, but they kept them as research tools because the outputs were unpredictable and the societal impact was unknown.
Only after OpenAI pushed theirs into the public sphere (framing it as a step toward AGI) Google and Facebook did follow, not out of readiness, but out of fear of being left behind.
Folks, we have the most INCREDIBLE announcement today. Maybe the best in the history of announcements, many people are saying it. We’re launching the beautiful Maxest app — and by the way, it was my idea, nobody else could have thought of this. It will come pre-installed on every phone. The fake news media will try to tell you it’s some kind of “spy app” - WRONG! Totally wrong!
The Democrats don’t want you to have this amazing technology. They want you to stay uninformed! But we’re giving you the most beautiful, most secure app ever created. Our cyber experts — the best in the world, probably the best who have ever lived — they built walls around this app.
Other countries will be so jealous of our app, they’ll probably try to steal it. But they can’t, because it’s AMERICAN MADE!
It’s funny that the exact same thing happened in Brazil during Bolsonaro. When the pandemic peaked, the Federal government simply sit down while the States themselves (literally all States) joined and negotiated the vaccines for Brazil bypassing the Federal role.
It’s crazy to me that workers in the US need permission to organize and coordinate actions.