The alpha, twice a week
That was 2 min.
Get the best new moments in your inbox every Wednesday and Saturday. It's free.
Check your inbox
We sent a link to your address. Click it to finish signing up. Not there in a minute? Look in spam.
or
MTS · 2 min · Nov 8, 2026
Frontier models can be worse at adapting their tone
Ben Hylak
Ben Hylak says RLHF can make frontier models less adaptable by pushing them toward a fixed tone and refusal style.
In this episode
Full analysis- 9:07Simulations replay production traces to test agent changes3 min
- 1:29More capable agents make failures harder to detect and costlier4 min
- 4:52Agent memory failures need separate storage and retrieval metrics3 min
- 12:53Frontier models can be worse at adapting their tone2 min
“i do think that frontier models are actually worse at these things than earlier models”
Ben Hylakat 14:03 · MTS
Takeaways
- RLHF can lock models into a narrower tone.
- Refusal tuning can help with security.
- Models may prioritize code over human interaction.
About this segment
Ben Hylak says RLHF can make frontier models less adaptable by pushing them toward a fixed tone and refusal style. That tradeoff matters when building personal agents.
Ben Hylak is co-founder and CTO of Raindrop.
Built Raindrop’s tools to detect, investigate, and fix failures in production AI agents.
Theo Jaffee, Host · Co-founder & host, MTS
One of the first members of a16z's new media team; earlier hosted the Emergent Ventures-backed Theo Jaffee Podcast.
Sophia Dew, Host · Host, MTS
From “Frontier AI Models Have Gotten Worse at Writing | Ben Hylak”, Nov 8, 2026. Watch the full episode
Thanks. We'll look at and fix what's off.
Transcript
Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.
- 12:58
it's not just the user prompting well i think that i i do think a lot of the future will be about uh building better ways to sort of align agents to what humans want like i don't know if that's going to be like through bci or if that's going to be just like better i think there's a lot that the products themselves can do um to align to people and like understand you better and i and yeah i i think that um i really think that that that's that's the gap right now is like uh you know you tell an agent what you want and it just does something that's like completely different from what you wanted and you realize that it's often a communication gap as well yeah yeah yeah um do you think that the like right now the ways that we have of agents like learning over time adapting to your preferences is just memory is is there something in between
- 13:52
or added that could like make that adaptation happen faster yeah it feels like there will be i i do think that like it's very obvious that um that on things like tone for example it just really struggles to adapt to people i do think that frontier models are actually worse at these things than earlier models well even even just like smaller models like i think they're just like really rlhf to have a very specific tone to have like very well calibrated refusals like all of these sorts of things which which is genuinely useful like if you're making a personal agent or something it's like it just like um you know if someone else tries even connecting to that like it just like it just has a pretty good intuition on what to refuse yeah kind of broadly so it can be useful from like a security standpoint but like yeah their tone is just and their writing
- 14:43
quality is like bad even like kimmy you know 2.5 was like actually like a pretty good writing model yeah i think that they're optimized to deal with like code yeah of course like you know they're not optimized for like human interaction anymore um but anyway speaking of human interaction you