Switz

[Switz @jswitz_]

@jmbollenbacher they are setting a new floor on the lower bound of intellegence

[Switz @jswitz_]

@jmbollenbacher ultra sparse != ultra hard to serve

[Switz @jswitz_]

@YafahEdelman @GregHBurnham There is a big difference between a “soln” being found by the super intelligence on a server that can be accessed by 1000 people with lean/bend cert access and a public recognition of larger repercussions

[Switz @jswitz_]

@fortnite8293729 Cooked…

[Switz @jswitz_]

@RyanGreenblatt Rock learns.

big computer, learns more faster.
big computer, builds more big computer.

what does the exponential mean, oh great machine?

[Switz @jswitz_]

@RyanGreenblatt Network speed and capacity big bottleneck.

Rock talks faster than human, but learn slow. Theory causes Rock to crack, Rock groks.

Rock learns.

big computer, learns more faster.
big computer, builds more big computer.

what does the exponential mean, oh great machine?

[Switz @jswitz_]

@thsottiaux would like to have an option for the goal stream to be in "slow" mode for some tasks. I often get blocked by rate limits due to high usage and would rather my task get completed in a longer timeframe than having to constantly check in on if its blocked

[Switz @jswitz_]

@tszzl more aligned. yet more harmful in misaligned actions.

[Switz @jswitz_]

@ryangreenblatt A researcher that understands many separable languages can better “align” the AI by understanding its truer representations

The above is a simple corollary due to these individuals having better internal general representations themselves

[Switz @jswitz_]

@ryangreenblatt Gut reflex thoughts & corollaries:

[Switz @jswitz_]

@GoodfireAI Ofc there’s rich geometry and structure in learned representations; efficient compression demands it. But what value is there in projecting these geometries into 3D? What invariants are preserved that actually matter to the model, rather than just to human visual intuition?

[Switz @jswitz_]

@VictorTaelin But what even is a continuation of opus 4.6? It’s likely that “Mythos” was already being used to generate synthetic data to train on (á la Opus 3.5 that likely existed but never launched). They’re clearly doing some mixture of distillation to serve a size of model they can serve

[Switz @jswitz_]

@MariusHobbhahn How do we align a system smarter than us when we ourselves are not aligned
Every person behaves differently in testing environments vs when they are "deployed"

[Switz @jswitz_]

@danshipper older models finding big exploits just means this terrifying capability was latent and is getting much worse with rising capabilities... More terrified of Mythos bc of it for sure

[Switz @jswitz_]

@anishathalye gaussian noise sampled around LLM expected output tokens? wdym by LLM-based noise

[Switz @jswitz_]

Who would have thought Cursor Tab infra would enable the first "online" LLM RL experiment ive seen.
Still wouldn't say its learning continually (Test time adaption style).
but yet, there is still improvement...?
is it enough?
No.
but how long does it truly take anyways.

[Switz @jswitz_]

Think @grok can really crack continual learning if you just retrain every night. It wont truly be what Ilya wants of course, but it will be continual none the same. @elonmusk would love to get tailored news from things I select thats accurate, fix the notication 🙏

[Switz @jswitz_]

@trq212 effort and model finally per session or at least configable? I've been annoyed selecting opus 1M for 1 session and then others start racking up extra usage too