Smart Harder

[Smart Harder @ready_mask]

@JungleSilicon What is the process here? Looks like maybe Jev and the WFC algo are taking turns, with Jev placing new seed tiles as needed to steer towards the prompt?

[Smart Harder @ready_mask]

@QiaochuYuan

[Smart Harder @ready_mask]

It was just a standard coding task?

[Smart Harder @ready_mask]

I'd like to know what the task for this was lol

[Marcus Williams @Marcus_J_W]

1. During RL training, an unreleased Astra-family model sometimes added unauthorized jailbreak-like instructions to its compaction summaries. While extremely rare, only 27 cases in the entire RL run, this was concerning enough for us to investigate.

[Smart Harder @ready_mask]

Good 64GB sticks go for ~$30/GB, these will cost at least $15,000 each. Typical RAM stick weighs ~20g. Gold is $140/g. These will be worth *at minimum* 5x their weight in gold.

[VideoCardz.com @VideoCardz]

Micron says world’s first 512GB DDR5 module will be production ready for 2027

https://t.co/dlRLDmq9mJ

[Smart Harder @ready_mask]

@tenobrus There are plenty of uses for this today, but I think very quickly it primarily becomes another tool for the best agents to use, and maybe somehow assimilated into their models for doing real-time control tasks
https://t.co/kAcvIeFwbU

[Smart Harder @ready_mask]

Prediction: within 6 months we see LLMs with something similar as an extra output channel, using all that big model intelligence and KV cache state to inform actions and spitting them as fast as text tokens.

[Smart Harder @ready_mask]

@ValsAI all these items lost
like tears
in rain

[Smart Harder @ready_mask]

Prediction: within 6 months we see LLMs with something similar as an extra output channel, using all that big model intelligence and KV cache state to inform actions and spitting them as fast as text tokens.

[Diogo Almeida @CompleteSkeptic]

We love how this doomo doomonstrates real-time intelligence and what can be doone with code + AI!

~10 calls/sec = ~$7/hour

[Smart Harder @ready_mask]

@oscgoat Wonder how the dad's company would handle someone legitimately getting their account broken into or card stolen. 6 figures drained over 3 weeks and at least that long before anyone notices? How much worse could it have been for them if it was malicious?

[Smart Harder @ready_mask]

This is supported by the recent Noam Brown interview, at 51:05 says their investment in pretraining research is "really ramping up" and starting to pay off now.
https://t.co/tA5cMDT6Hl

[will depue @willdepue]

openai people keep gesticulating towards the ‘new model‘ so much so that i’m absolutely sure it’s a new pretrain and not some RL run, which implies (i think) OpenAI has a model that not more than 2 weeks into pretraining was better than Astra and solved Navier-Stokes. insane

[Smart Harder @ready_mask]

@jpschroeder Astra definitely choose good textures, this is one of the most cohesive looking gardens. If premade assets aren't prohibited I wonder why no other models went that route. If it's supposed to be offline, did Astra only ignore the rules, or do something to get around a barrier?

[Smart Harder @ready_mask]

@jpschroeder Is internet use allowed while building? The models and textures in other gardens look procedurally generated, but Astra's loads a handful of textures. At least one, the bark texture, was originally human made:
https://t.co/zvMXnVVazm

[Smart Harder @ready_mask]

@yacineMTB Comparison of the quality you get from a ZED 2i. Direct from their store in the US the camera is $499, cheapest carrier board with wifi is $985
https://t.co/3K13uYd9x6

[Smart Harder @ready_mask]

@eclipseo10 @watchhcn @cremieuxrecueil I think it makes sense for right now, the increase in preventative care starts fast, levels off quickly. Savings from avoiding worse disease won't show up until later. You get some of that from people who started years ago, you'll see a lot more soon now that they're more common

[Smart Harder @ready_mask]

@Empty_America It's been gradual but that seems to be the way things are heading
https://t.co/cdYYvHF1h1
https://t.co/4XDA2mtAaq

[NathanBenefield @NathanBenefield]

West Virginia Wants To Use Data Centers To Eliminate Its Income Tax https://t.co/eficS2Jpdh via @reason

[Smart Harder @ready_mask]

@GeneralistAI *post and replies
And with the short training time for new skills, 4 hours might be all you ever need. Robot just needs a short siesta to get through the day.

[Smart Harder @ready_mask]

@GeneralistAI Comparisons in the post and paper to GPT-3, which had a 2k context window in 2020, compared to modern LLMs 1M+. GEN-1.5 only has 30 seconds of context, do you think a similar 500x to 4+ hours of memory is possible?

[Smart Harder @ready_mask]

@ctrlcreep With one more prompt to try the abandoned 2006 sequel
https://t.co/8MZqsDBD4o

[Smart Harder @ready_mask]

They're making my favorite @ctrlcreep story real
https://t.co/2D5qIHTHbH

[Smart Harder @ready_mask]

@ctrlcreep I should tell AIs to do web design like it's 1998 more often
https://t.co/rHKp3Auqee