An AI with more real layers can actually be “smarter” in the conventional sense of the term.
Words found on the source page on Nov 6, 2026.
More from Scott Alexander
BlogThe Specter Of NeuraleseSep 24, 2026
For 100 steps in a row, the AI can shoot delicate subtle ideas from layer to layer at the speed of light. Then there’s one step where it has to encode them into twenty-six glyphs invented by Phoenician turquoise miners in 1800 BC. Then it has to re-encode the Phoenician glyphs into delicate subtle lightspeed ideas before it can do anything else.BlogThe Specter Of NeuraleseSep 24, 2026
The science of reading these thoughts is a subfield of AI interpretability, which is still in its infancy.BlogMysteries Of AI GeneralizationSep 23, 2026
This story of misalignment says that LLM alignment makes AIs more aligned, RLVR makes them less aligned, and the exact level of alignment depends on how these two things interact or cancel out. But if this work generalizes, all the bad effects from RLVR get sequestered to RLVR like problems.BlogMysteries Of AI GeneralizationSep 23, 2026
This trains the AIs to be focused on task success, which naturally risks including things like reward-hacking, cheating, and single-minded pursuit of stated goals at the expense of ethical injunctions.
On the same topics
BlogWhy I still haven't bought into true RSISep 19, 2026
It is important to not confuse massive steps in inference-time scaling, a dynamic which should be fairly predictable, with being the outputs of RSI, which is highly uncertain.PodcastDwarkesh Podcast: Agent swarms, alignment, and recursive self-improvementSep 17, 2026
If you have five hours, you’re probably going to do a lot better. The AI models are pretty similar.PodcastDwarkesh Podcast: Agent swarms, alignment, and recursive self-improvementSep 17, 2026
So I was like, “I don’t think we’re going to get it in 2026, probably not in 2027, maybe in 2028.” So it did happen a lot faster than I expected.PodcastDwarkesh Podcast: Agent swarms, alignment, and recursive self-improvementSep 17, 2026
The concerning scenario is that, especially as these models are becoming more capable, we make them what we think is aligned, and they’re 99.9% aligned. Then we use these models to help us with the next generation of models, and they end up being 99.8% aligned. Then with each subsequent generation, we see an increasing degradation in alignment.BlogWe Must Pace The FrontierSep 14, 2026
If I was in charge of an alignment or interpretability team I would never run out of experiments to run and things to try, including having the time to properly talk to the models in the first place.BlogWe Must Pace The FrontierSep 14, 2026
Over the summer there was a step change, as Mythos and Astra started kicking off the early stages of recursive self-improvement (RSI).
About this quote
Quotesby Glyph