{"podcast":{"title":"LessWrong (Curated & Popular)","slug":"lesswrong-curated-popular-5643401","podcast_index_feed_id":5643401,"rss_url":"https://rss.buzzsprout.com/2037297.rss","website_url":"https://sites.libsyn.com/421877","image_url":"https://storage.buzzsprout.com/xq8g0aka74ttwoa9xkxlmy5fn9oa?.jpg","author":"LessWrong","episode_count":1002,"summary":"Audio narrations of LessWrong posts. Includes all curated posts and all posts with 125+ karma. If you'd like more, subscribe to the “Lesswrong (30+ karma)” feed.","last_synced_at":"2026-09-28T08:23:50.844993+00:00","page_url":"https://stenobird.com/podcast/lesswrong-curated-popular-5643401"},"episode":{"title":"\"Alignment Midtraining Cracks Under Pressure\" by J Bostock, sidbaines, Daniel Tan, draganover, ma-rmartinez","slug":"alignment-midtraining-cracks-under-pressure-by-j-bostock-sidbaines-daniel-tan-draganover-ma-rmartinez","published_at":"2026-09-23T19:45:21+00:00","page_url":"https://stenobird.com/podcast/lesswrong-curated-popular-5643401/alignment-midtraining-cracks-under-pressure-by-j-bostock-sidbaines-daniel-tan-draganover-ma-rmartinez","show_page_url":"https://stenobird.com/podcast/lesswrong-curated-popular-5643401","url":"https://www.buzzsprout.com/2037297/episodes/19854741-alignment-midtraining-cracks-under-pressure-by-j-bostock-sidbaines-daniel-tan-draganover-ma-rmartinez.mp3","audio_url":"https://www.buzzsprout.com/2037297/episodes/19854741-alignment-midtraining-cracks-under-pressure-by-j-bostock-sidbaines-daniel-tan-draganover-ma-rmartinez.mp3","summary":"TL;DR We stress-test alignment midtraining (AMT) across model and token budget scales. Our results suggest that midtraining cannot tackle the hard problems of AI alignment—namely distributional shift and reward underspecification in the presence of imperfect data. For instance, we test whether midtrained motivations are robust to finetuning which elicits competing motivations. In our setting, 190 million tokens of midtrained motivations are overpowered by a relatively tiny amount (~50 th...","meta_description":"TL;DR We stress-test alignment midtraining (AMT) across model and token budget scales. Our results suggest that midtraining cannot tackle the hard problem…","key_points":[],"chapters":[],"topics":[],"duration_seconds":874,"processing_state":"not_requested","actions":[{"name":"request_transcript","method":"POST","url":"https://stenobird.com/v1/public/podcasts/lesswrong-curated-popular-5643401/episodes/alignment-midtraining-cracks-under-pressure-by-j-bostock-sidbaines-daniel-tan-draganover-ma-rmartinez/transcription-requests","description":"Idempotently request low-priority transcript generation for this episode."},{"name":"read_markdown","method":"GET","url":"https://stenobird.com/podcast/lesswrong-curated-popular-5643401/alignment-midtraining-cracks-under-pressure-by-j-bostock-sidbaines-daniel-tan-draganover-ma-rmartinez.md","description":"Read the agent-friendly Markdown representation of this episode resource."}]}}