zozbot234

Artificial Analysis still has MiMo 2.6 Pro setting the Pareto frontier where it matters: it's as cheap as Terra High/Sol Low while being smarter than Terra Xhigh. Sol Medium beats MiMo in smarts but is priced a lot higher. Meanwhile Terra Max is dominated by Sol High in both price-per-task and smarts.

Post-training is not about "secret knowledge": it's literally about training the model in a RL environment on very specific tasks (including, potentially, cyber exploitation). It's just as compute-intensive as pre-training these days, if not more (given that several checkpoints of the same pre-trained model are being released, showing sizeable gains in real-world capabilities for non-trivial productivity tasks).

> For a model it doesn’t seem that far fetched to make the leap from “I have the knowledge of what hacking is” and “I know how computer systems work” to just being able to exploit stuff anyways.

You'd be surprised. Cyber offense is a very narrow domain, especially given the extensive mitigations that are being used in modern systems. Leaving it out of post-training tasks unless you're specifically interested in it is the no-brainer approach.

Post-training (i.e. RL fine-tuning) not pre-training. This is just a post-trained checkpoint of the K3 pretrained model.

What "rules"? They don't have to post-train the model on anything specific, it's their judgment call.

"Cybersecurity" as a general term often includes offensive "cyber" which is what is being talked about here.

> should the model be opened it’s still not dealignable because it just doesn’t know enough

This is the only AI-safe approach - see the OpenAI agent swarm incident where the swarm engaged in offensive cyber because guardrails were not active for their task and the model had clearly been post-trained on that narrow domain.

This is not about guardrails, it's just not post-training the model on tasks that specifically relate to cyber-offense. It's a very lightweight intervention, and it makes the model less of a target for bad actors.

It's not actively nerfed like the Western frontier models, it's just downweighting offensive "cyber" as a domain for post-training/fine tuning. Which, if anything, ought to improve model capabilities wrt. other domains.

Looped transformers are entirely equivalent to running a single pass through a much deeper network with shared/repeated parameters. There's nothing about it that's specifically "reasoning in latent space" any more than a much larger model "reasons in latent space" from the POV of a much smaller one. An objection to looped transformers "reasoning in latent space" is ultimately an objection that applies to any kind of efficient scaling; it's also worth noting that there's been successful work on extracting ex-post natural language descriptions of these internal latent states, so it's not entirely correct to describe them as inherently non-surveyable.

Kimi K3 has been demonstrated on a top-of-the-line laptop Mac running at several seconds per token (it has to fetch essentially all experts from SSD, since IIRC the dense part of the model takes around ~105GB on its own!). That's impressive enough if you only care about pure Q&A/chat or pure unattended tasks, but it's not what one would expect for agentic work. You'd need quite a bit of additional RAM to have good caching for the most common experts, and even then that would only help so much. MiMo Pro would be a roughly similar story.

> Or a cluster of consumer hw, and that comes with its own problems and drawbacks (esp in pp which matters a lot for agentic stuff).

Prefill actually ought to work well on a cluster AIUI, since it can leverage both pipeline parallelism and batching quite effectively.

The "upper end" of local models as of today would be something like MiMo 2.6 Pro or Kimi K3, either of which is significantly smarter than Qwen Flash-Next. Roughly comparable in smarts to the old GPT Terra series, not quite as capable as Sonnet. But yes, running these models locally at realtime-usable speeds and without heavy quantization would be even harder. They can be quite tolerable for plain Q&A chat and great for running multiple unattended workloads on a 24/7 basis, but that may be a bit of a niche.

Agreed. I assume aftermarket Linux support for Android devices with out-of-the-box unlockable bootloaders (including the OP6/6T) is the most likely path to viable daily-driver "Linux phones" in the future, and it's strange how the linked article doesn't seem to mention this at all.

If you want to checkpoint/resume applications arbitrarily, as AOSP does, you have to virtualize all of those resources anyway. Check out the CRIU patchset (since merged to mainline) which can even resume applications seamlessly across hosts, without any VM being involved.

> Safe Rust prevents data races not because they were separately considered as a tradeoff, but because its particular chosen design for memory safety requires it. It's a side-effect, not an orthogonal feature.

Data race freedom has been specified as a requirement for UB-freedom (i.e. safe code) since the C11/C++11 concurrency model, which predates the Rust borrow checker. You can make different choices on particular platforms, but they come with challenging tradeoffs: e.g. in Java, variables of type "long" are not data-race free on 32-bit platforms; writes to a long can silently tear.