NotinReality

i think i get it now

[NotinReality @other__reality]

@KeyTryer I really like thebes short story on continual learning, something about it just feels so much like curent LLMs.

https://t.co/gYUGBaz4aR

[NotinReality @other__reality]

@AcerFur @voird33r Unless computer use is a native modality kind of like we do with robotics and VLA models, an LLM could identify and figure out the way to play a game but still be bottlenecked by the coarseness of text

[NotinReality @other__reality]

@SkylarDarkfox @thinbluesublime @LittleKeegs0 @HotOompaLoompa @LinkofSunshine No one has a strong enough definition of understanding to truly differentiate between the ability to analyze and real understanding. It still could easily be the case that LLMs don’t understand, but it’s not something we have evidence for or against.

[NotinReality @other__reality]

@AntonHand @KarstonKellar This site is only tolerable with an Adblock.

https://t.co/Omks1RVhDy

The blame manga is 10x better than the anime because the real draw of the story is seeing big sci-fi megastructures, and it just does it better. Not having an annoying dub voice is just a bonus

[NotinReality @other__reality]

@lukeprog

[NotinReality @other__reality]

@anotherpanacea @morallawwithin I feel like dungeon crawler Carl is bordering rationalism enough that the plot wouldn’t look that different, but maybe that’s a lack of imagination.

[NotinReality @other__reality]

@almaududiiqbal @andy_barcia_r @lthlnkso @jimstewartson This is the cot that’s been summarized to avoid distillation. This is the reasoning text after being promoted to prove the theorem that led to the creation of the counter example, disproving the theorem.

https://t.co/Wxx3Vsn7Ze

[NotinReality @other__reality]

@almaududiiqbal @andy_barcia_r @lthlnkso @jimstewartson The 125 page report is the reasoning chain, unless you think OpenAI is in the business of solving 80 year old math problems by hand then passing them off as AI.

[NotinReality @other__reality]

@tobiasly @1Brown_Charlie @baselessassert At a scale so much smaller compared to agriculture that completely eliminating them would have no overall impact on water usage, that it makes spending any time worrying about pointless when water is still being used at unsustainable levels on wasteful agriculture

[NotinReality @other__reality]

@willdepue Even if it was computationally cheap to serve, do we even have data for long context training? Modern models can barely stay coherent within a million tokens, and unless they have a really good synthetic setup, they definitely don’t have the data to keep it coherent over 100M.

[NotinReality @other__reality]

@HL_Alyx_NoVR I just obtained one with an autoclicker even though its out of stock. If its still in your cart, theres a chance you can obtain one, you just need to keep clicking continue on each step until you get through.

[NotinReality @other__reality]

@nanomysou @SSGNELL @Anton1699 @catgirlprostate It’s a transformer model generating two separate images. There’s a fine line between the two, and if the training was slightly different, you could likely apply this technique to noise to generate a completely new image.

[NotinReality @other__reality]

@AdmiralQuality @Empty_jr Oh wait yeah you’re right. For some reason I was convinced the book said differently. Yeah that’s a strange writing choice.

[NotinReality @other__reality]

@AdmiralQuality @Empty_jr I got a big surprise for you about Rocky

[NotinReality @other__reality]

@GjMcGowan @billyhumblebrag Does this matter if the focus is on accelerating ai capabilities?

[NotinReality @other__reality]

@ubuto23 @KelseyTuoc @lookingforgames Do you actually believe that the constitution is a moral framework that describes how an AI should interact with humans, and think that’s how Claude should be trained? Or do you just see the word constitution and freak out?

[NotinReality @other__reality]

@gabriel1 I love engagement bait

[NotinReality @other__reality]

@itsTeemux @mattyglesias Not too far off. In the book they make it much clearer that erosions have much better memory than humans. That’s why he picks up human language so fast, and why Rocky says “I can’t forget” in the movie

[NotinReality @other__reality]

@ChadNotChud @romlib_ @1nterchangeable The trillion dollar bet the labs are making is that increasing coding capabilities will allow for the automated discovery of new architectural improvements. Similar to AlphaEvolve, but applied to larger problems than just matrix multiplication and kernels

https://t.co/Y5VLj30Jy5

[NotinReality @other__reality]

@hugodeboss1 @TetraspaceWest So then the government shouldn’t buy this product, it’s not like this wasn’t clearly stated when they bought it

There are many other AI labs out there