#大佬观点

NVIDIA AVO continuously inspects, plans, implements, and evaluates, using memory, tools, and execution feedback to build on what it learns along the way. This allows the system to sustain progress across long-running tasks rather than starting over with each model context. Read about AVO and how we built it for long-horizon autonomous agents: https://nvda.ws/3S5qdS1

An excellent history of scaling laws from @jietang. In 2020, we explored the limits of sparsity in Switch Transformers by routing each token to only 1 out of 2048 experts (in retrospect, a bold choice). The model had fewer than 3B activated parameters, but 1.6T total parameters (comparable to today's frontier models). The 1.6T model achieved better C4 perplexities than the T5 models using far less compute, set a new SOTA on TriviaQA, but was dumb as bricks on reasoning tasks like SuperGLUE. The lesson was that the optimal tokens-per-parameter ratio is highly task-dependent. Or as @NShazeer had already intuited: FLOPs were intelligence; parameters were knowledge!

tfw your model is so bad at instruction following, you have to remind it to shut up every single turn (including tool-calls)

I keep hearing people say this, but I think it's to the credit of Altman & Amodei that they told us what leading researchers in this field actually think. We should worry more about Mark Zuckerberg and his politically savvy AI reassurances!

It's underrated how much you can learn about AI with what's public and out there already. People overestimate the degree to which you need to be in a lab to understand these systems and the direction of travel; don't let the status arbitrage discourage your independent research!

organic architecture good codebases have always had strong foundations and constraints. these files go here. this kind of code goes there. the codebase is held together either by catering to the lowest common denominator developer using conventional frameworks and languages, or by strict adherence to a style guide enforced by code review. the arrival of an infinite number of monkeys armed with keyboards have obsoleted the latter. codebases designed for agents need hard constraints so that even the dumbest agent (and pilot) can contribute meaningfully. the chaos must be tamed. refactoring your codebase and choosing a tech stack with strong type systems/compiler diagnostics has far greater impact than AGENTS.md and tweaking your skills. hard constraints and well thought out architecture guide agents to do the right thing by default. the best rules are in your code. invest accordingly.

Multi-agent alignment is basically just political philosophy, right?

defenders can see the future, and have a narrow window to uplevel their cybersecurity practices now. key is to uplevel fundamentals and apply the best AI tools. what we’re doing at OpenAI, and where other organizations can start: https://blog.gregbrockman.com/the-defenders-window

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.

Artificial intelligence is reshaping work across the economy, writes Peter McCrory. So far, it hasn’t raised unemployment, but is that about to change? https://www.thefp.com/p/ai-jobs-unemployment-economy?taid=6a862eb02f77dd0001a59ae9&utm_campaign=trueanthem&utm_medium=social&utm_source=twitter

BREAKING: Stripe CEO @patrickc: "There's a reasonable chance that 2026 Q1 will be looked back upon as the first quarter of the singularity."

Stripe sent a letter to investors this morning saying they believe The Singularity has begun, and that we passed the threshold on January 1st 2026.