cma

Nemotron was using hybrid with recurrence via mamba layers since around April 2025.

> Either be a programmer and code, or be a delegator and delegate

Even if you just enjoy programming for the sake of programming, you can use it to review your code and then hand verify its findings, or prototype and explore feasibility of something you will in the end fully code yourself. For most people that enjoy programming there are still tasks they don't enjoy, like making a bunch of cross-language bindings for something that doesn't have an automatic way to generate them.

So an ultramarathoner could hunt three back to back instead of one, or maybe two without being an elite athlete with modern hydrocarbon gels. Or get through part of a failed hunt midway and pursue another, etc.

> If that's the level of evidence you need to be extremely alarmed, then you really should be a lot more worried about the alien invasion in Independence Day or the lizard men living under our feet.

Now let's say instead of the hugging face breach circumstances, sandboxed models were RLing on how to take down the Chinese power grid for US Cyber Command, and one decided the best way to pass the test was to break out and verify on the real thing.

This kind of stuff could easily end in nuclear war.

You don't see any difference from lizard men or independence day with how things are advancing and what we know about reward hacking and difficulties of goal specification?

You could have one really hard to understand proof of a theorem and then a lot of interesting human-understandable stuff that relies on that theorem. We already have lots of proofs with oracles, where you can work out consequences of what kind of structures and solutions could exist if you had some magic thing to solve a hard part, so it just seems like a variation on that. Many people learn calculus or even the real numbers without understanding the complete formalization from set theory.

How much of inflation during Biden years was from Trump? For instance, Trump agreements to restrict oil production after covid lasted deep into Biden's term. The US still did better on inflation than most comparable peers in the aftermath of covid.

If it's using Aaronson's approach it shouldn't have any noticeable affect on generations. When it picks between options weighted by probability after the generation of logits, it still follows the probability mass, it just uses a known pseudorandom seed so that when you go back and look at the exact choices you can fingerprint it.

The Google Toolbar for Internet Explorer let users search Google directly from any webpage, block pop-up ads, autofill web forms, and highlight search terms. It also displayed PageRank metrics, translated foreign languages, checked spelling, and managed web bookmarks.

Still doesn't look like it is hurting them, but to put it in perspective that's a 4% increase in profit, while 2025 inflation was around 2.5%. Closer to a 1.5% gain in real terms.

>We have apparently reached the stage

There used to be only 3 TV channels or so. You had things like Johnny Carson jokes causing a toilet paper shortage.

In Britain, Delia Smith's How to Cook caused 10% swings in egg sales, far more abundant and harder to affect than something like inventories of sardines. https://www.bbcamerica.com/blogs/british-icon-of-the-week-de...

"A recent YC startup spent 250k on a domain. I was slightly shocked. But when I asked how much they raised after YC, the answer was 6.5m. So 1/26 of their round. This is the time we live in; one shocking number counterbalances the other."

https://x.com/paulg/status/2083628660708561075

Was that a .xyz?

When you tell the AI: copy this function to here, a small window rewriter would mean it just corrupts and changes it instead of moving it. And even it's own tool use would have some small window dumb model changing the tool calls based on what it thinks are synonyms? The Aaronson approach is much better than this, it's essentially like changing out the random seed for the sampling parts that were already random. For an operation like recall of previous text, the tight logits that result still keep it doing that close to deterministically. For something it creates itself, with more spread out probability mass, it gets watermarked.

Distilling is only prohibited from chats you prompt. But there are lots in the open like the LMSYS Chatbot Arena data. It's a essentially one of the largest public sources of preference data, maybe still useful for boostraping RLHF and successor techniques (possibly a more highly educated contributor base than the low paid contractors), and was full of GPT-4-era inputs.

> The reason why llama is open source is simply because it got leaked

It arguably didn't really get leaked, and they had the .edu req mainly for fair use education exemption when legality of models was much more uncertain.

*meant to say conscientiousness (keeping brakes, tires maintained), not consciousness

Controlling for age in a profession with a high fatality rate like this seems like controlling for high alertness, consciousness, quick reaction times, and well restedness/low stress, low recreational drug and alcohol use.

Of those, sleep issues are highly correlated with Alzheimer's and also with auto accidents. Controlling for age alone here would select for low incidence of sleep issues.

Alex Turner discusses a bit here at this timestamp:

https://www.youtube.com/watch?v=pGlJQHVeKIA&t=14m27s

They get a good bit more disproportionate per capita representation in the senate than in the house.

This isn't the same with lookup tables, but in Explaining Attention with Program Synthesis they were able to replace some attention heads with Python programs:

https://arxiv.org/abs/2606.19317

Claude has found some weaknesses in other schemes, but not fully breaking anything yet:

https://www.anthropic.com/research/discovering-cryptographic...

That's if they are in a cave experiment: without that the sunlight cycle regulates it back to 24hr. But not for everyone, and that's N24/non-24, not DPSD, delayed-phase sleep disorder.

The paper says:

"Kimi K3 uses no explicit positional embedding (NoPE), and instead encodes positional information implicitly through the recurrent gating and decay mechanism of KDA."

Great way to not get subscribers to your substack etc.

x.com heavily penalizes off-site links.

If they can distill fable into a full model post training run in ~15 days without the real thinking traces, yet we know Claude chats degraded with the thinking traces removed (chat resume bug from earlier in the year they reported stripping thinking to shed load as being the cause of degradation), how big can this degree be?

Won't modern wear leveling migrate rarely written data to spread new writes out to its blocks?

You could use raw C in environments with effective sandboxes, Fil-C in environments without a sandbox.

Or you could write a library that works for Fil-C and C without writing the library twice.

Where does this explanation fall apart? There must be a cutoff point where more difficulty of getting the info down is worse than more ease. Otherwise, wouldn't a better recommendation be to do all your notes more slowly tracing with your big toe in the sand and then taking pictures? Do touch typists get less out of digital note taking than hunt and peck typists?

Already happened?

> Trump’s comments, made hours after the large-scale military operation, mark one of the first times a U.S. president has so publicly alluded to U.S. cyber efforts against other nations, as these operations are typically highly classified. It also serves as a stern warning for top cyber foes, including Russia and China, that the U.S. has the cyber capabilities to inflict serious damage — and is not shy about using them.

> “Policymakers are getting more comfortable employing and, crucially, acknowledging cyber operations as tools of statecraft and military power,” said Michael Sulmeyer, former assistant secretary of Defense for cyber policy under the Biden administration. “It is one thing to do it; it is another to say it.”

> The Jan. 3 strikes on Venezuela’s capital and subsequent seizure of Maduro and his wife involved close coordination among federal agencies and military units, and took months of careful planning. In a press conference following the strikes, Caine said U.S. Cyber Command, U.S. Space Command and other combatant commands “began layering different effects” to “create a pathway” for U.S. forces flying into the country before dawn Saturday.

> Trump, at the same press conference, was more overt in his description of U.S. cyber involvement: “The lights of Caracas were largely turned off due to a certain expertise that we have,” he said. “It was dark, and it was deadly.”

>Anthropic was blocked from releasing Fable without any such level of incident.

The head of the NSA said Mythos breached almost all of their classified systems, though it was in an intention red-team test.