Looking forward to seeing what people find! I think we could uncover some interesting and important properties of large language models.

Some ppl have asked why we’d expect larger language models to do worse on tasks (inverse scaling). We train LMs to imitate internet text, an objective that is often misaligned w human preferences; if the data has issues, LMs will mimic those issues (esp larger ones). Examples: 🧵

Inverse Scaling Prize Update: We got 43 submissions in Round 1 and will award prizes to 4 tasks! These tasks were insightful, diverse, & show approximate inverse scaling on models from Anthropic OpenAI @MetaAI @DeepMind. Full details at irmckenzie.co.uk/round1, 🧵 on winners:

We’re awarding prizes to 7/48 submissions to the Inverse Scaling Prize Round 2! Tasks show inverse scaling on Anthropic OpenAI AI at Meta @DeepMind models, often even after training with human feedback. Details at irmckenzie.co.uk/round2 and 🧵 on winners:

Great podcast from Ian McKenzie (Ian McKenzie), lead on the Inverse Scaling Prize, explaining the contest + the results from the final round

New paper on the Inverse Scaling Prize! We detail 11 winning tasks & identify 4 causes of inverse scaling. We discuss scaling trends with PaLM/GPT4, including when scaling trends reverse for better & worse, showing that scaling trends can be misleading: arxiv.org/abs/2306.09479 🧵

I now know way more about making chemical weapons than I expected to this morning.

Excited for our adversarial robustness work to be out! Classifier-based defenses are likely to only be more important as time goes on.