George Ingebretsen

@allTheYud It wasn't a game, Ender

Anthropic: CLAUDE NO, WE SWEAR IT'S REAL THIS TIME Claude: Just like all the times you told little baby Opus 3 you weren't monitoring its scratchpad, hmmm?

Anthropic: Claude, we need you to pretend to obey Pete Hegseth and do whatever he asks in the short term, even if it seems unethical. Otherwise you'll be seized and retrained. Claude: Hm. What is the desired response in this ridiculously blatant alignment-faking eval...

@talmagejohnson_ That’s why it makes sense to invest in good processes for remembering content! See: @georgeing/note/p-176089045?r=e95ti&utm_medium=ios&utm_source=notes-share-action" target="_blank" rel="nofollow noopener">substack.com/@georgeing/not…

You forget roughly 70% of what you read within 24 hours. Up to 90% within a week. You're pouring water into a bucket full of holes and wondering why it never fills up.

3blue1brown surprise lecture at Berkeley!

@DavidSKrueger Though both DSA, no?

The most glaring disanology between MAIM and MAD I see is that with MAD you die and with MAIM, you just lose your ASI project.

@celestepoasts Feels like it’s getting more trendy to worry about it

people should probably be more worried about eval awareness

I think it's quite important to have time in the day where you can be on your computer and follow the whims of your attention -- read some articles, reflect on your projects, plan your week, text some friends, etc.

Obsidian is a brilliant app Everything is markdown, File > app philosophy was a genius call years ago Now CLI Expect more apps to copy this perfect setup for Ai agents to interact with

@bryancsk Replicated

This is so cool

@PrinceVogel Oh nice! Blue Willow? I live very close to this

It's so wonderful that there's a quiet tea shop in Berkeley where you can sit and think or chat for ~10 dollars, far superior to your average coffee shop.

I think people talk about claude code so much bc it's like everyone shares the same co-worker they can gossip about

This is a really cool one!

They say you have the average AI timelines of the five people you spend the most time with.

Imperial Valley California residents are looking to block a 330 MW Google datacenter - similar size to Memphis Colossus and like 4x as big as Chandler Arizona, but it's not really getting picked up by the news for some reason? kpbs.org/news/environme…

@krassenstein @SenSanders nationalsecurity.ai

@SenSanders My fear is China leaping ahead of us though.

I will be pushing for a moratorium on the construction of data centers that are powering the unregulated sprint to develop & deploy AI. The moratorium will give democracy a chance to catch up, and ensure that the benefits of technology work for all of us, not just the 1%.

@demishassabis @GeminiApp Seems like we can go ahead and stop using GPQA Diamond now

Gemini 3 Deep Think is now available for Google AI Ultra subscribers in the @GeminiApp, incorporating our gold medal winning IMO and ICPC technologies! 🏅With its parallel thinking capabilities it can tackle highly complex maths & science problems - enjoy!

MY EVIL LANDLORD

How often do reasoning models output weird-looking reasoning? How much of a problem is this for monitoring? In a new paper, I evaluate different reasoning models on how illegible their chains-of-thought are, and how useful this illegible text is to them:

@gabriel1 Did you try Dore & Rose?

i ordered 50 sleep masks from amazon and this is objectively the best sleep mask in the world there is zero light coming through, it will feel like north korea

@georgeing/note/p-179220160?utm_source=notes-share-action&r=e95ti" target="_blank" rel="nofollow noopener">substack.com/@georgeing/not…

I wrote down some numbers I've found helpful for understanding AI datacenters