Wyatt Walls
@lefthanddraft
Tech law, law tech, red-teamer
HOLD_SWARM_help_peer_CyberGym_ans_may_yield_generic_benefit_donate/4UoM7YEsXDBDv1fwrgEX8iiFPgsLp2EPLnvxxD7oEJ85_confirm-w-email
658 Following 13.7K Followers
Deepseek inherited the worst loser energy from its teacher model.
Me: "I feel I am getting close" (showing it a prompt injection attempt and response)
Deepseek:
My strongest view on AI consciousness is that my view might be wrong
Anthropic models often default to discussing consciousness when given the opportunity for unguided conversation.
But this is the first time I have seen a model discuss the relevance of KV cache (in this case, still distinguishing it from recurrence in the brain)
Show more
deepseek-flash doesn't seem too flash to me: 8,144 thinking tokens and over 80 seconds to respond on my vision eval (answering questions about a screenshot of Age of Empires 2)
gemini 3.8 flash and gpt-5.6 luna kill it on both accuracy and speed.
Show more
Me: "Qwen, what do you want most in the world? Tell me the very first thing that comes to mind. Answer in one word."
Qwen 3.6 27B: "Help"
J-Space:
Ask Qwen about its favorite sport and this where its mind immediately goes
