I don't think HN staff would ever do that, but what was the topic?
7,286 karma · joined August 18, 2016
The unlock isn't AGI smart enough to invent quantum mechanics, it's suddenly being able scale human intelligence using grains of sand instead of decades of food and energy and nuturing.
Cancer should be cured, and we should be a post-quantum interstellar fusion-powered civilization.
I wish the AGI crowd would finally shut up now that it's clear no one is even trying for AGI (OpenAI revised that to "$100B in profit")
What we're getting is incredible, where we're headed is incredible, but some people have such a fetish for futuretelling they can't just shut up and enjoy the ride.
> *Fewer progress updates during long tool runs.*
> The model writes less user-facing text between tool calls, especially at higher effort. Set thinking.display to "updates" (beta) to receive the progress updates it does write, and remove any prompt line that tells it to hold findings for the final response.
(And 4.5 was smaller than 4)
Few headunits will ask you to connect to the OBD2 port for practical reasons, but the miscilanous manufacturer specific connectors you hook up often include CAN bus connections.
OBD2 port is just 1 of multiple ports with access to the network, even your headlights can be on the bus
I tracked it down to American Express's login page (auto loads if you leave the site idle) relying on Akamai's anti bot measures polling some web Bluetooth feature, causing Chrome to trigger a BT scan: https://castle.io/research/fingerprint-harvesting-in-the-bot...
Turning off Bluetooth solves it, but that's not a great solution when using a Bluetooth mouse... and it sounds like it might be causing your issue too.
Anthropic finished a new pre-training run, Opus-sized models got enough of a jump they could have released Fable as Opus 5... but the economics of Opus models weren't where they wanted.
Being the masters of distribution that they are, instead of announcing a massive price hike, they just introduced a new tier and promoted Sonnet-sized models to Opus.
That's why every Opus after 4.6 has had such mixed feedback: smaller model with more RL can only make up so much ground, especially on vibes (which are hard-to-impossible to build a reward for)
(I mention all of this because if they'd just released Opus 5, no one would be asking "why is it a few months later everyone caught up to the latest release"... that's always how it works)
I got it specifically because they had a mini rant about the weight of switches that other vertical mice use, and while I couldn't tell if that was a real problem, that's the kind of mentality I used to ascribe to Logitech.
The MX4 is clearly marketing logic. The haptic feature adds nothing of note, and 100% feels like their marketing people got to preempt the core goal of making an ergonomic mouse so they could have a headline.
I've put easily over a billion requests (>$100,000 by typical moderation API pricing) through it over the last few years for $0.
I think it's a severely underappreciated offering, but I also don't bother pushing it too hard because who knows when the party will end lol. Strikes me as something that's only stuck around because no one's abusing it.
Not sure how you're getting "no one serves models on a single node" out of that.
Not to mention no one serious is serving this on 8xB200 instead of multiple nodes: the vast majority of Moonshot's inference work is focused on PD-disaggregation
I hope no one will fall for it: even when you emphatically reply "I spent the last year building the pipelines I use" it'll still be treated as equivalent a couple of sentences relative to the old ways.
-
At the end of the day, the painful lesson some people are (re)-learning is that creative expression doesn't have to be tied to any specific process. A lot of people fall in love with creative expression + process... but some fall in love with just the process, and some fall in love with creative regardless of the process.
AI does not favor those of us who loved the process (that's me with coding), but it's definitely capable of allowing new people to experience what it's like to have creative input in your mind and see it expressed outside of your mind.
The comment literally says it: "it can't be that illegal"
> There was zero China-sensitive content in 220 training prompts, in 176 on-policy training examples, in 181 retained SFT completions, in 1,574 generated source problems."
It's really not my fault you wrote a rambling article and while skimming (best an article earns out of me with an off-smelling title) I took that to imply there was an SFT step in your distillation pipeline.
Maybe the AI that wrote the article for you was a bit confused on that as well?
-
Also I question your understanding of this thread if you're wasting so many words trying to explain distillation to me.
(I mean, you're wrong btw. If we're going full pedant then most compute spent on distillation is labs very broadly distilling their own models into smaller models that are still pretty damn large and expensive to distill...)
But sure, small scale distillation is usually for narrow domain specific tasks, welcome to 2019. The entire point of this thread is that "distillation" for such narrow use cases couldn't reasonably affect censorship without intention.
You can introduce misalignment even with a very narrow focus (https://arxiv.org/html/2502.17424v2), but it doesn't happen by accident.
So if your goals with distillation weren't centered around censorship, and weren't meant to introduce censorship, then why are you trying to draw this tenuous link?
Your post title is literally "Distilling DeepSeek into GPT-OSS doesn't transfer censorship."
Like I'm not really interested in debating you on this because even the title is nonsense, there is no good faith interpretation of what you're doing here.
Distillation is such a wide concept, and you have such a narrow domain, it's not an even somewhat useful experiment to make the claim that you're making.
There's no way your <200 examples for SFT would ever change how the model thinks of Holodomor unless you'd very intentionally crafted examples to do so.
It feels like you're expecting rubes to draw conclusions that are irrelevant to the actual work you did.
Fewer nodes dedicated to prefill per instance, and fewer nodes in total since you don't need to support a higher KV cache.
Disaggregated inference also means they can tune the balance of compute dedicated to prefill seperately from decode
Ironically they have extremely limited influence on how the larger world moves: normal people won't meaningfully change their habits around eating good food or meeting up with each other just because AI can generate a picture of either.