Owen

[Owen @marginference]

@alexmarin @nickreidx 😂😂

[Owen @marginference]

@Chonghan_Liu IMO, the separation of independently controlled read/write gates is a major architectural breakthrough too - https://t.co/kIWTT5Td2f . Interested to hear your thoughts on this

[Owen @marginference]

@Chonghan_Liu incredible write up.

[Danish Hussain @astrodanish]

the best way to understand any engineered system is to trace:

- 1 bit of information
- 1 joule of energy
- 1 newton of force

from source to destination. once you can do this, your engineering task becomes much simpler. you will probably solve 5 or 6 problems on accident as you try to trace these flows.

[Owen @marginference]

@bryanlytics @FloridaMaannn @Dropbox How big is the full dataset? Might be able to help host a public download mirror for a couple of months. This looks incredibly useful. DM me if that would help @bryanlytics .

[Owen @marginference]

@mistakenlyused @thebeautyofsaas +1 to Eugene the goat. Will make you dangerous.

Others too: Great Leads six ways to start any sales message
Scientific advertising
Cashvertising ok

[Owen @marginference]

@kalomaze Agree - feels like a very interpretable model too if checking internal activation paths etc, and huge dormant clusters of features you can train to fire more/connect bridges that has huge qualitative perf jumps

[Owen @marginference]

@bycloudai Contact their support ai asap and it’ll refund you likely

[Owen @marginference]

@aifilmmaker @My_Ai_Bi 😂

[Theo - t3.gg @theo]

In other words, there’s a ~90% chance we’re fine

[Nathan Godey @COLM 🌉 @nthngdy]

🧵New paper: Matryoshka LM Suites

Instead of training all models of a suite independently, we nest them within a single architecture and train the suite in a single run => much less compute, same performance

Also allows free distillation, better speculative decoding & more 👇

[jietang @jietang]

Thoughts About Scaling Law

Scaling, but not only of parameters. Every model release now ends with the same question: how many parameters? It isn't a question that can be answered on its own. Parameter count is only meaningful alongside three others — how much data you have, where you intend to spend your compute, and who will run the model, under what conditions.

The field learned this the hard way. Kaplan et al. (2020) fit an exponent that told everyone to grow parameters faster than data — roughly 2.7:1 — and the industry complied: GPT-3, Gopher, MT-NLG. Hoffmann et al. (2022) redid the experiment across four hundred models and found the compute-optimal split is closer to 20 tokens per parameter, and that with sufficient compute the two should grow at the same rate rather than drifting apart. The error in the earlier fit compounded with every order of magnitude of compute, which is why the largest models of that generation were the most misallocated. The trillion-parameter round was, in retrospect, a detour the whole field took together and then reversed.

Chinchilla wasn't the end either. It optimized training compute for models that would be trained once and evaluated. Today a model is called billions of times a day and inference dominates lifetime cost. Put inference into the objective and the optimum moves toward smaller models trained far longer — deliberate over-training, which is what Llama-2-7B and Gemma-2-9B were doing at roughly 290 and 889 tokens per parameter.

Sparsity moved the target again. In a MoE model two quantities have to be kept apart: total parameters govern roughly how much the model can hold — knowledge, facts, the long tail — while activated parameters and effective depth govern roughly how far it can think, how many steps of a causal chain it can carry before it comes apart. A dense 20:1 ratio does not transfer. And the ratio isn't a single number at all: Roberts et al. (2025) find the optimal tokens-per-parameter is task-dependent, with memorization favoring more parameters and reasoning favoring more data. Follow-up work on MoE observes that at fixed TPP, pushing total parameters higher actually degrades reasoning, while activating more experts reliably helps it.

This matters for what we are building toward. Finding a vulnerability is not a retrieval problem. It doesn't come from having memorized more CVEs; it comes from carrying a twenty-step chain of inference to the end without losing the thread. That capability does not live in total parameter count.

Which brings us to this release. Total parameters appear to matter up to a threshold — enough to hold the world — after which additional capability comes from scaling elsewhere: effective depth per forward pass, and above all post-training. GLM-5.3 is our controlled experiment on that claim. Same base, same architecture, same total and activated parameters as GLM-5.2. One month of scaling long-horizon environments and RL. The gains are not marginal. Well, scaling has more than one dial. We turned the post-training one this time because it had the most slack left in it — not because the others are finished. Base model size, pretraining data, compute spent per forward pass: all of them are still on the table, and we will come back to each. What this experiment taught us is that the dials do not have to be turned together, and that the one worth turning next is rarely the one that was worth turning last. We are not done scaling. Next time, maybe mid-training, pre-training, and even more.

[Owen @marginference]

@JPBrebner But not for the reason you think

[Owen @marginference]

@teortaxesTex @an_interstice @teortaxesTex check this out https://t.co/a5BXLEXWT4