Reply to thread

Message: <blockquote data-quote="MemoryHacker" data-source="post: 29052050" data-attributes="member: 576605">A recent study by Stanford University and UC Berkeley compared the performance of GPT-4 and its predecessor GPT-3.5 over a few months since GPT-4's release. They found that GPT-4's accuracy in answering complex mathematical questions significantly dropped from 97.6% to 2.4% between March and June. Moreover, GPT-4 was less inclined to provide detailed explanations for answers, even when prompted. On the other hand, GPT-3.5 showed improvements in basic math but struggled with complex code generation. GPT-4's coding capabilities also regressed, with only 10% of generated code adhering to platform guidelines compared to 50% in March. OpenAI's update process remains somewhat mysterious, raising concerns about transparency amid growing global concerns and regulations surrounding AI ethics. Despite some shortcomings, GPT-4 showed improved resistance to certain types of attacks and avoided responding to harmful prompts.Ad:[MEDIA=youtube]kVmi1GKSc90[/MEDIA]</blockquote>

Top Bottom