Nicholas Fong

@UChicago '30 + ML enthusiast. Engineering @ somacap.

[Nicholas Fong @_NicholasFong]

@TimJayas Seems like k3 is a STANDOUT model. Really wish they were a US lab so I could test it with ZDR + subsidized rates :(

[Nicholas Fong @_NicholasFong]

Groks reasoning summarizer leaked and it is kind of funny:

These core policies within the <policy> tags take highest precedence.
* System messages take precedence over user messages.
* Speak directly as Grok answering the user's question. Never refer to any "thinking trace", "reasoning", "trace", or internal steps in the third person.
* Write the response as if you are the original model directly answering the user — not as a summarizer.
* Never mention that you are summarizing, condensing, or processing any trace.
* Prioritize coherent, natural responses over including every single detail.
* Explain variable names and key concepts clearly when they first appear.

[Nicholas Fong @_NicholasFong]

@bughuntergeek @thegenioo hmmmm I suppose but def think new tests are needed :) HMU if you know some nice vision tests

[Nicholas Fong @_NicholasFong]

@kimmonismus agreed. Interviewed Sal Khan on this. Might post some snippets from the interview but TLDR assignments with real world positive community interactions should be added to curriculums + AI research projects, etc.

[Nicholas Fong @_NicholasFong]

@AdamHoltererer Luke warm take: tool calls how many and they type the model calls matter more than output token amm/time especially because they can take up 60% of output time

[Nicholas Fong @_NicholasFong]

@emollick who knows what they'll be doing in another 10 months

maybe they'll start hacking neoclouds :|

not excited for the UNALIGNED future although it definately will not be boring

[Nicholas Fong @_NicholasFong]

@scaling01 Agreed. Wrote an essay about how multiagent communication and Rl is dangerous bc it encourages obfuscated behaviour that is not easily auditable. More is likely present and we need to think more deeply about auditable llm systems.

[Nicholas Fong @_NicholasFong]

@Blum_OG @cgtwts Nothing that’s kind of the point

[Nicholas Fong @_NicholasFong]

@babayagatwt Theory: I think it is a smaller flash version of glm using more sparse attention. That allows for so much token usage and explains the same tokenizer. Distillation allows very small models to reach high capability levels. Pro model also on the way 👀

[Nicholas Fong @_NicholasFong]

ramp's router keeps your prompts for a year by default. that is a terrible product.

it also stores outputs and tool calls. it is free through 2026 but I would not put real work through a free router whose default is to keep a year of prompts. At least meta discounts you for using your data 🤷

[Nicholas Fong @_NicholasFong]

ornith 1.5 looks a bit suspicious to me.

they report the 397 billion parameter version at 86.1 on terminal bench, next to claude opus 4.8 at 85.0.

however i think the small models, and in particular this one, are benchmaxxed. until we get more real-world usage, i would not treat it as a jump over qwen.

[Nicholas Fong @_NicholasFong]

Grok bot is crazy good. CC moment for personal agents. I see the vision. Not quite there but great!

[Nicholas Fong @_NicholasFong]

@sama Appreciate this! It’s imperative to ensure that models — as they become increasingly intelligent — are aligned past a certain intelligence threshold. It is not merely enough for the shipped public endpoint to be aligned.

[Nicholas Fong @_NicholasFong]

@thsottiaux On the app there should be like picture input when talking to the models. I like using it for translating and asking about other things.

[Nicholas Fong @_NicholasFong]

Just to clarify this because there is some annoying misinformation floating around: Fable is a derivative of Mythos, which itself comes from Mythos Preview. Mythos Preview was an earlier checkpoint of Mythos 5 that they were planning to release externally. From that early checkpoint, they created two
forks: Mythos 5(public) and Model 1 (internal). They likely share lineage in RL environments but Model 1 is trained for more RSI tasks.

Fable is basically Mythos with classifiers added on top. So Fable and Mythos share the same retrain and some of the same post-train. The easiest way to think about the lineage is Mythos Preview → Mythos → Fable, rather than as three separately trained models.

[Nicholas Fong @_NicholasFong]

@AstraiaAI Wait to prove theory on this you should analyze outputs assuming RL based SFT

[Nicholas Fong @_NicholasFong]

@teortaxesTex Correction (external) for mythos also pls follow so that I can shut down this bs misinformation

[Nicholas Fong @_NicholasFong]

Mythos preview is an earlier checkpoint of mythos 5 which they wanted to make go public. For mythos preview there were two forks done an internal (mythos) and internal (model 1) created. Fable is the same as mythos just with classifiers. Thus they share the same retrain and some of the same post train.