penagwin

To be clear they said it “still works” and “ ok thinking more about this, there's no way this is the same vuln, this is another one.”

Meaning this specific bug the parent comment is talking about is not new in just these patched versions from today.

So delaying your updates today would mean you’re exposed to an actively exploited RCE and still exposed to the other memory corruption bug (which may or may not be an easily exploitable RCE) for an entire 2 or 3 days.

And from the sounds of it people are very actively exploiting this so waiting 24+ hours will practically ensure you’re hit.

I do agree that it’s generally matched my impressions of models for general SWE at least until we had Opus 5 and 5.6-Sol beating Fable 5. Which coincidentally is when they hit 74%.

Models like Deepseek 4.1 flash, Muse 1.3, and MiniMax 3.1 flash are certainly capable and very cost effective, but I’m not sure anyone would say they’re better than Fable 5/5.1 in capability

> [Space Bunny] was an early version of MiniMax M3.1 Flash.

They mention that space bunny was an earlier checkpoints, so it may have improved (or may it could be benchmaxxed)

I find it interesting that so many models seem to be at almost exactly 74% specifically which makes me suspicious of the benchmark. I want to say 5.6-sol, Astra, deepseek flash 4.1, opus 5 (and 5.5?), Gemini 3.8-flash and a few others all ended up there?

Would there be any point in 2.5gbps beyond installing system updates a few seconds faster?

AFAIK you’d be hard pressed to hit 150mbps with your own local media.

And any streaming services will use way less - Netflix and Hulu recommend a mere 16Mbps at 4k for example.

I assume it’s largely a side effect from the final RL in post training?

That’s the step that causes the most significant gains in agentic performance.

But the RL doesn’t care about anything except maximizing the score, so if you only score based on coding benchmarks, anything can happen to the writing style (as long as it doesn’t hurt the coding performance).

That’s why it often gets worse on models that simply had more RL post training from the same base.

Reasoning models are the same speed. They’re just post trained with RL to do CoT inside tags like <thinking></thinking> before a tag like <response></response>

There’s no difference in the inference implementation, parameter count, or speed.

Yes, as well as applying for jobs (or at minimum when hired), renting an apartment, and lots of financial things including any type of KYC crypto exchange or investment accounts. I've also had utility companies ask for it.

These are in no way secret, I have no idea how people are okay with this. You can easily social engineer so many critical services if you know somebody's SSN.

Apparently 3G is being phased out by AT&T (and I think most other providers have already left 3G) in the US.

Which sucks because I can get a 3G module for my iot projects for 5$ and I think 4G modules start at 50$?

Are we just at the mercy of 4G now for services that need large coverage but low throughput? I know 5G technically has a spec for IOT type stuff but I haven't seen anything about it, and I haven't seen modules for it.

To me it's fair to say it's not a likely scenerio, it's just that they continously say "this would be difficult" at every step. I appreciate the breakdown, but it comes off as trying to convince you it was a near-complete impossibility. I understand it was unlikely, but it was possible, and that makes it severe either way.

We currently self host a version from ~2019, I haven't looked into upgrading the version but just the old version is working great.