On Oct. 23, @log4j wrote that they had been banned over reverse-engineering work. “i have never used openai to exploit or find vulnerabilities, only local reverse engineering which tripped daybreak,” the account wrote. That is @log4j's account of what happened, not an explanation from OpenAI.
@log4j described the cost as “-$500, and a bunch of banked resets” and wrote that the experience motivated them to get three more machines and finish a cluster.
In an Oct. 24 quote-post, @bayeslord called the situation “debanking for the AI era” and alleged a lack of due process. “There are legitimate safety concerns but the way the system works right now is broken and insane,” they wrote. In another quote-post, @beffjezos argued that open-source and local AI are an important counterbalance.
@yacineMTB described the work as reverse-engineering a TV and suggested accelerating alternatives. @log4j's post did not identify a TV. In replies to @yacineMTB, @AnandNSharma worried about local reverse engineering tripping an exploit classifier and argued that a ban without a stated rule would leave nothing to appeal. @BSumariuk argued that bans for local reverse engineering push potential bug-finders toward local models.
Comments