3 ms·
AI Is an Existential Threat to Itself
- senttoschool 3y agoDo these mainstream news media read HN, Reddit and basically regurgitate what was discussed there? I swear I read these propositions last week on HN from individual contributes. Now this week, mainstream media keeps writing about it with sensational headlines. This seems to happen very frequently.
- aszantu 3y agoProbably
- jqpabc123 3y agoSummary: If widely and successfully deployed, LLMs will eventually start digesting each others output and form one big, gigantic feedback loop that ultimately tends toward what ... most likely random nonsense and incoherence ... that may or may not sound plausible. Conclusion: LLMs are still as dumb as a hammer because they can't identify their own output as the bullshit it is. This is *not* real AGI but rather just a poor facsimile of it. The only way these bullshit generators can destroy humanity is if humanity itself is bullshit.
- novaRom 3y agoI would rather say there is existential threat to get all human knowledge spoiled, mixed with fabricated high quality facts, faked textbooks, faked movies, faked replicas of original scans, etc.
- jqpabc123 3y agoLLM knowledge has to be spoiled first before "human" knowledge can be. Since "humans" are already aware that this is happening, I'm not going to lose too much sleep over humans forgetting how to unplug these bullshit generators once they are proven to be largely unreliable. And they're almost certain to become less reliable as they start feeding on each others bullshit pulled from the internet. So a more likely outcome in my judgment --- the internet will just be further spoiled as a reliable source of factual info.