Read the abstract ➤ https://www.conf42.com/Python_2024_Boluwatife_BenAdeola_python_cutting_edge_language_model_research Other sessions at this event ➤ https://www.conf42.com/python2024 Join Discord ➤ https://discord.gg/DnyHgrC7jC Support our mission ➤ https://www.conf42.com/support Chapters 0:00 intro 0:20 preamble 0:40 intro to mechanistic interpretability 2:06 mech interp 2:57 mech interp toolkit: causal interventions 5:10 example mech interp research 16:59 interpretability libraries and packages 17:44 nnsight architecture 20:42 example intervention 23:30 anatomy of an intervention 27:43 model internal i/o are nodes on the intervention graph 28:33 information flow 29:11 average-out information vectors 31:06 getting the average activation vector 34:45 adding the average to one-shot activation vector 39:15 resources