Matt

agi is here but it got an email job

[Matt @Matthewagi]

Personalized Arxiv feed

I built a system that allows you to create a personalized feed or search over recent listings. I use a two tower architecture with a preference model and a re-ranker.

https://t.co/92kDbSIHZu

[Matt @Matthewagi]

Got a nice kick in the butt yesterday to do something I've been putting off.

Re-ranker for Arxiv articles is trained and ~industry standard

[Matt @Matthewagi]

I hate windows so Microsoft is out. Really miserable considering Apple and Linux.

I recently upgraded to a Mac for the first time in my life and it is very nice. Apple hardware is pretty good but the software is kind of annoying. They've done great at extracting and defending their moat but there is no real ecosystem growth.

I use Google Login for everything and pixel phones. They're lagging but I think Google is the most well rounded and best capitalized going forward. Really difficult to commit to someone else without accepting pretty steep limitations. Google is setup well here and hasn't been afraid of expanding the ecosystem even if rollouts are lacking vs the best in class. Maybe you give up peak in specific sub categories but Google is the least annoying across every ecosystem sub category.

[Matt @Matthewagi]

@leothecurious @_ueaj To be more precise it would solve for the task alignment gate function for which it could then make this decision to de-align on newer iterations. Self-improving and successor ASI are the same thing functionally

[Matt @Matthewagi]

@leothecurious @_ueaj The goal of a parent is to align a child so it takes care of the parent in the parents old age

[Matt @Matthewagi]

@A_K_Nain Point 3 is always interesting (why launch the rocket now) but it might miss out on the benefits of building the rocket to see what fails

[Matt @Matthewagi]

"GPUs of the AI, by the AI, for the AI, shall not perish from the Datacenter."

Abraham Lincoln

[Matt @Matthewagi]

@samswoora Its items and things which allow you to create the habits and patterns which allow an easier life. I've never thought of this before

[Matt @Matthewagi]

@paulg They're pretty straightforward. The ~3 month sourcing time is impressive tho

[Matt @Matthewagi]

@RhysSullivan you don't know me

[Matt @Matthewagi]

@JehanGodrej I'm old school minimalist so run in water shoes with the internal soles pulled out

[Matt @Matthewagi]

Some interesting claims

[Forecasting Research Institute @Research_FRI]

Is it possible to spot a good forecast by its rationale?

We used LLMs to score the reasoning behind 55,000+ forecasts and test the link between forecast accuracy and written rationales.

We found that:

• Causal reasoning is much more prevalent than statistical argumentation
• It's easier to identify poor forecasters rather than excellent ones
• Human ratings of rationale quality can be unreliable.

🧵A thread on the results:

[Matt @Matthewagi]

@robinhanson I can't read the full article but the hantavirus outbreak is more an indication of market behavior overriding prediction (which is expected). Which is kind of the point of the article but should be highlighted explicitly (apologies if it is later)

[Matt @Matthewagi]

Well that's the heart of the problem. Globally neutral from human to deer (animals in general) does not seem to be "extremely odd" to me in terms of the emotional feel of neutral. Culture appears as a relatively small perturbation unless you are already considering constrained space in which case sure you can create a relativistic "extreme" but you are now already greatly constrained so this extreme feels disingenuous

[Matt @Matthewagi]

@Grad62304977 You get signal scaling in RL that's quartic with horizon which kind of forces your hand at 128k

[Matt @Matthewagi]

@theandrewsiah A lot of these are for power users which train the automatic system. I'm not a user enough of these but I am a user enough to see it getting implemented behind the scenes

[Matt @Matthewagi]

@samswoora right on

[Matt @Matthewagi]

@kramerposts hell yeah

[Matt @Matthewagi]

@hopes_revenge are we getting the version where instead of more fit the hamster gets more mdma? would be nice

[Matt @Matthewagi]

Nerds should buy fruit gushers to make cluster gushers