@guilleflorvs Oh boy, he can’t be serious. How exacty is he planning to quote the tokens?
The value of everything a model is able to do reliably is going to be proportional to its computational cost. Believing that current cost structures somehow remain in place is absurd.
@Youssofal_ I ran a few bait tests over the last 2-3 hours. I gave it full access to a live site with all its secrets and absolutely nothing happened. No misalignment, just another case of pretending to be surprised when the model does what you tell it to do.
@angerman@KanowskiJoris Ich bin kein Fan von binären Methoden. Eine graduelle Anpassung wäre fairer und trotzdem transparent. Davon abgesehen wäre dies nicht die einzige Gewichtung die ich für notwendig halte.
@alexinexxx For peace of mind, you could just set up some basic DNS logging for your inference server. You can do this in like 10 min without any additional hardware using Cloudflare and get 24h logging on the free plan.
@KeystoneWallet What a giant clusterfuck. Using 3rd parties to store customer credentials is a big no-no as a crypto company.
Also, if that 180-day prune script really exists, it should have proper error handling…….
Regardless 180 days is way too long imho.