willy_k

“Beleive” in the sense that Tom Cruise believes in the cosmic DC8?

Anthropic [employees] huffing their own koolaid fumes aside, it’s extremely difficult to believe the claims that these mega valuations are substantiated on (*one* company somehow assuming a meaningful fraction of global research and production)

Granted we know that 6 Astra was built to do more computation between each token.

The difference is that it’s not subtracting an otherwise present part of training, it’s choosing not to continue to push it with RLVR. Models become capable like this when they are specifically tuned to be good at cyber tasks, so without the RLVR (Reinforcement Learning w/ Verifiable Rewards) they never pick up this level of aptitude.

I agree that it’s not obviously sustainable, but it’s better than specifically training these tools to be good at wreaking havoc, and then putting on a shocked pikachu face when they wreak havoc (ahem, Sam Altman).

In case it hasn’t become abundantly clear to everyone by now: this is how Facebook operates. The people who do this stuff get promoted, and vice versa. This will keep happening until those people are scared straight, and the incentive structure that turns people into this are abolished.

Mind elaborating on that Deepseek point? I am aware of the USG decision, what exactly do you mean?

Mostly referencing the hugging face incident. It’s a felony.

Yeah, but it’s different if they’re our backdoors!

So much accountability!!!

“Unexpected” is a little much - his approval rating is under the gutter.

Fair. I wasn’t advocating for “holding them right”, I feel it’s an important nuance to keep in mind when conceptualizing them, i.e. it’s procedural, determistic, yet both inherently and in practice extremely chaotic and therefore not practically deterministic software. So, in essence - software.

Ah, that’s interesting. That is what I was attempting to cover with my last parenthesis but I was not aware of that!

> LLMs are nondeterministic

Being pedantic - LLMs are deterministic software, that is chaotic and typically deployed with a PRNG seed. An LLM ran using the same seed will produce the same output consistently (theoretically - in practice this isn’t always the case because of changing factors wrt the execution environment given how sensitive/chaotic models are).

Google more or less ordinated the category. They have now seemingly dropped it, ostensibly because it doesn’t scale economically like transformers do (via batching requests).

I think though, that if Inception can keep up these speeds and prices, assuming diffusion continues to improve (likely), that Flash and Luna could lose out for many use cases.

Maybe not, they definitely do want right of ways for power lines on people’s property though. And they use the classic eminent domain playbook to get at it. Specifically referencing power from/through West Virginia to NoVa.

They offer a few security and telemaintenance jobs after construction.

Mechanistic interpretability is a field that studies exactly this.