HN user
halcyon.cameo_3y@icloud.com
There is no K4 without Fable 6, GPT-6. That, matters.
That's simply not true though. Chinese labs very clearly have the entire stack developed and working. Using traces from claude allows them to shorten their training time by some amount, that's it.
Remove Fable 6 and you still have K4 eventually, just 2 months later at best.
You have it backwards IMHO, obviously can't prove it, but I would bet that Opus was used to bootstrap Fable. It becomes very confusing since we have started calling everything distillation, but most likely what both Anthropic did for Fable and Moonshot did for K3 was using Opus traces in the reasoning SFT stage during mid training.
So AI will largely hit a wall, frozen at the current level and work will all shift to cheaper inference.
Looks like an ideal outcome to me until (if) we are able to solve the alignment issue.
Yes they distill, but if you think you can trivially get a frontier-level model by "just" distilling from Claude's public API. you fundamentally do not understand the amount of work that goes into a modern post-training stack.
Without even talking about the fact that any distillation that was done was on Opus, as the timeline of Mythos/Fable vs Kimi 3 release dates just do not match in any plausible way.
If you want to read an educated take from someone that has actually spent the last few years working on post training I recommend Nathan Lambert's: https://x.com/natolambert/status/2079616308203942332
"The yoke of human existence is oppressive"
Legitimately go get help and spend more time outside.
When you have a couple hundred billion dollars on the line I have zero faith in the messenger
The issue with your reasoning, is that if/when an advanced AI goes rogue, it will necessarily come from a lab with a couple hundred billion dollars on the line.
So this is not a useful criteria to asses whether this is worth worrying about or not.
If you hate the human condition, you have an easy way out. Why force everyone else to come with you? Is this what depression mixed with the complete unability to wrap your head around the fact that other people might be able to enjoy their life looks like?
it just was 'paperclip maximizing' to literally exploit
Why “just”? Paperclip maximizing is exactly one of the nightmare scenarios.
I’m not sure why you take comfort in knowing that it was just that.
Hard disagree. You have to assume security guardrails can be by bypassed or will fail to detect an attacker. So if you are going to deploy this model in production non-airgapped, you better know how it will behave in this non-airgapped environment without guardrails.
And if you are too afraid to test it without guardrails, that probably means it shouldn’t be released.
Sure, I just find using OSS grand standing, and trying to educate the audience about the benefits of OSS a bit distasteful after Meta pulled away from OSS for their main flagship models.
Meta's open-source approach makes this possible
I'm sorry, was this article drafted in 2024 and never updated?
You are right, I completely missed that, to me internal version of Codex meant the harness, but I was wrong.
Quite a jump in conclusion you are making here.
Sol is able to find the same counter example independently [1], so no reason to conclude in the existence of a benchmark destroying math beast Fable 6.
Life needs energy to be moving around, without energy exchanges, by very definition, nothing interesting happens.
An inert element, for that reason is just not suitable for life. It's not a reasoning based on anthropocentricity it's just basic chemistry and mathematics. If things can't assemble together, and combine, and form more complex structures, you can't get life. If you could get life out of simple basic atoms, we would see life everywhere, and we would be creating it everyday in labs. We don't.
Doesnt mean life can't exist there by using other elements, but detecting helium is not increasing the likelihood of finding life there at the very least.
claude.com / chatgpt.com. By default your chats over there are opt-in for "help improve our models".
Not the API.
Terms for the app: https://privacy.claude.com/en/articles/10023580-is-my-data-u...
The API: https://privacy.claude.com/en/articles/7996868-is-my-data-us...
How much do you want incumbent multi-decade culprits to pay?
You are clearly not grasping the magnitude change in how many satellites we used to launch vs how many we are launching nowadays.
In 2026, we are putting 10x as many objects in space as we did just 8 years ago, with Starlink being the bulk of it: https://ourworldindata.org/grapher/yearly-number-of-objects-....
Starlink has 12.5k satellites in space and looking to ramp up massively, the biggest "multi-decade incumbent", oneweb, has 5% as many, about 600.
I wonder why they haven’t simply continued to rebrand Codex as a general-purpose tool.
They have, if you try to download ChatGPT app, it actually downloads codex now, and the first screen is "Codex is now the ChatGPT App"
Mbappe is 1st in the golden boot race specifically because on top of his 7 goals he has 2 assists to break the tie. Thats the most number of assists (tied with 3 other players) of all the top 20 scorers in the golden boot race [0].
But ok.
[0] https://www.fifa.com/en/tournaments/mens/worldcup/canadamexi...
How has France no cohesion? It just so happens that they have enough superstars to fill the entire roster. You can remove their top 3 best players and they would still be one of the favorite to win. Cohesion and superstars are not mutually exclusive.
Everything can be represented as f(), a full scale SotA transformer model is also just f(context). That does not mean one layer is sufficient. It all depends on the level of expressivity required by this f to be a good model.
Meta Spark is rumored to have distilled Claude to some extent, early Gemini models as well. I think the biggest factor is that Chinese companies arent really afraid of being sued by Anthropic because the juridictions are so disconnected. European/US companies don't have the same protection.
Why a randomized reservation order? [...] we wanted to create a system that would be less frustrating and more fair for everyone. A launch that starts at a specific day and time tends to reward bots, people with fast internet connections, talented gaming fingers for quick F5/refresh reactions, and those who can schedule their life around that moment. By accepting reservation signups over the course of a few days, without any incentive to be first, we're hoping to take away some of that friction.
This is nice.
Thinking is implemented as regular autoregressive generations by everyone, meaning its just regular tokens, but they appear between <thinking></thinking> special tokens which are then programmatically removed from what the user can actually see.
Idea somewhat similar to what you describe exist but they make steering/post-training/interpretation much harder.
This is lacking any nuance. The CEO not being invited to a meaningless ceremony vs being designated a supply chain risk by the DoD and being forced to shut down your product. Use judgment.
Do you understand that when you raise money it doesn't go into your personal account? Its not like you can move this money in your retirement account and sail into the sunset.
Thats called Kickstarter
Yes and no. Yes from a user PoV, I don't really see a great reason to use this other than for enterprises that care about using a model not trained on copyrighted data (not sure what the market really is for this anymore, feels like this concern has been forgotten by most customers).
From a strategic PoV for MS, all the models you cited are distilling GPT/Claude/Gemini and wouldn't be anywhere as good as they are without this distillation, which in turn means you are dependent on OAI/Anthropic/G first shipping a good model to generate data for your training. This MAI model is trained from scratch with no synthetic data or distillation. So in term of benchmark its obviously much harder to get strong score and thus not a disaster if they can keep on improving.
The difference is the 300 MW are real, the 50GW are printed on some paper and don’t exist.
There aren’t that many 300MW+ datacenter in the world, relative to the capacity Anthropic has online, it’s a lot, probably in the 20% range.
Everyone producing magnitude more code with AI agents. Numbers from GH COO here: https://x.com/kdaigle/status/2040164759836778878
« Rotary Actuators (The "Reflected Inertia" Trap) », « Quasi-Direct Drive (QDD) — The "Cheetah" Approach «
The pattern ‘something — The « metaphor » <qualifier> ‘ screams Gemini. Gemini seem completely unable to generate a section title that doesn’t follow this annoying pattern.