[tsukikage]

This process should not require a human in the loop.

Consider:

* spammers have access to large amounts of compute via their botnets

* the effectiveness of any particular spam message can easily be measured - it is simply the quantity of funds arriving at the cryptocurrency wallet tied to that message within some time window

So, just complete the cycle: LLM to generate prompts, another to generate messages, send out a large batch, wait some hours, kick off a round of training based on the feedback signals received; rinse, lather, repeat, entirely unattended.

This is how we /really/ get AI foom: not in a FAANG lab but in a spammer's basement.

[FeepingCreature]

At least one sci-fi novel iirc had an AI spam filter achieve sentience, because the task basically amounted to contrastive-learning a model of humanity.

[russnewcomer]

That’s one of Peter Watt’s Rifters trilogy, I think maybe the second one? Been a few years since I read them. I think it’s a biological neural net, not an Ai per se. Lots of big ideas in those books, but not a lot of optimism and some rough stuff.

[biztos]

Having worked in the field, I think you’re more likely to achieve AGI by intelligently watering tomatoes in a hothouse.

[Bluestein]

Very well could be. Seconded. After all, it could very well become one of the largest vehicles for "mass training", ever ...

PS. Howerver, see comments downthread about "survivorship bias". Not everybody will reply, so biases will exist.-