@fm2279@social.coop @jsbarretto@social.coopJoshua Barretto Just throwing in my 2 cents that I don't think that data poisoning is as effective as forcing companies to pay an acquisition cost for good text. Doubtless they have a QA pipeline, one that includes low-paid workers overseas, that can filter out generated text pretty effectively and at low cost. They'll have a much harder time avoiding tarpits, which could really slow down scraping if deployed at a large enough scale. As @jsbarretto@social.coopJoshua Barretto suggested, having the tarpits be as fast as possible and as diverse as possible furthers that aim.

I was also thinking a model like what signature-based malware or adware detectors use would be cool too. Whenever one tarpit discovers a new technique for snaring bots or keeping bots snared longer, it pushes that innovation to a database where other tarpits can download it and adjust their own strategy, allowing tarpits to collectively adapt faster than the bots can.

Anyway thanks for the good work!