Viewing profile — 5555watch

Even for consumers.. Every enthusiast now wants those specced out rigs to play with LLMs. Make a nice chip for that, and it will reduce some pressure on consumer RAM demand

Is it still true? I'd assume you should be able to freeze the matrix and unfreeze an expert block, before feeding particularly chosen training data. Or that doesn't work?

In my understanding the first Deep Think / Pro models were already very good as they were doing some kind of parallel repeated reasoning, thus were slow and expensive. So if chatji…

I mean in their code the simulation is incorrect. Someone found the source: https://github.com/pem725/Dunning-Kruger They're sampling slope and bias for the line from U[0,1] and U[…

It's an interesting read. Curiously, it doesn't really debunk anything. The fact that given X and Y random and independent, that Y-X is correlated with X doesn't disprove the Dunni…

The extrapolation can also be a learned skill, especially in math. How many papers took result X, extended it to Y using known building blocks, and applied to Z. By the way, convex…

They're simulating randomness incorrectly: relationship between true and perceived will average 0.5, not 0; and bias will average 50%, not 0%. That's why their "random data" is slo…

Thanks for finding the code! Now it's much more clear. The simulated data tries generating the true relationship between actual and perceived scores from 0.0 to 1.0, and bias in se…

So what does that change? If there's no error bars on the graphs you can't discuss significant differences easily. And if they do differ, plotting differences will yield the U shap…

Very hard to understand the meat behind all the fluff of the article, especially as the simulation code is not available, and as the presented simulated and original graphs are eff…

I like the illustration that the models are working on a convex hull of known information. Filling gaps with linear combinations of known facts and results. They can't exit the hul…

For reference, Safari is doing around 53, and Chrome is around 60 for M5, of which they were bragging about [0]. So curious to see whether there's any gains from your custom browse…

Aha, I'm getting 97/100 on Safari

What's the Speedometer 3.1 [0] rating for that browser paired with an M5? [0]: https://browserbench.org/Speedometer3.1/

But it's not proactive then? There's just a hidden loop or some cron-like signal that feeds data to a reactive loop.. I'd imagine proactive as something like, hmm, no signal from X…

I wonder into what weird research niche would all such AI schools converge to after enough time.

I'm curious, how hard/expensive it is to burn a really large model into silicon, and why aren't we doing this already? Or, when we will start doing this, who's going to be able to …

It's a division by zero. We can change infinity to 'undefined'

I used to chat with "the paper" until both me and the AI were roughly on the same page at what's happening. It's important, as a "full read" of the "full paper" is not always used …

I'm probably a weird outlier. Coming from academia, it's ranging from 1x to infinity-x (as in, certain tasks wouldn't even be touched if not for AI). For stuff that I'm used to (R)…

I love that they let you switch to a more common q'k notation!

A silver lining to think about.. Currently there's a big issue (at least in some countries) that mathematicians must publish from 5 to 7-8 papers within a 5-year window to keep the…

So with this release do they kill the 5.5-Pro model with super long thinking and reasoning? 5.6-Sol-Ultra is not the equivalent, right?

Will it also have hardcoded self-lobotomy if asked about cutting edge ML or LLM solutions? (Looking at Fable here)