Liam O

researching privacy in LLMs // high schooler

[Sauers @Sauers_]

UPDATE: a challenger emerges

[Meryem Arik @MeryemArik9]

Pictured: Google Deepmind CEO Demis Hassabis steps down to become chair

[Liam O @liamobuilds]

@BlancheMinerva I’m genuinely curious here, why WOULDNT they release how they eval AI? I see no benefit at all to anyone that way

[Liam O @liamobuilds]

I just watched Perks of Being a Wallflower last night and I read the book at the end of the school yr. they're both 10/10.

fun fact: it's one of the most challenged books in the USA, 36 challenges alone in 2025. go do yourself a favor and read banned books, like, all of them :)

[killian @hellokillian]

The big labs should fund— and seal— the @internetarchive.

1) We will stop trusting historical footage soon.

The Internet Archive is a neutral party and closest to a complete archive. They can verify X historical video was uploaded before vid models existed.

One public "sealing" = widely sharing cryptographic hashes of all content. Everything in it is authoritative pre-AI history.

2) Labs need the same pre-AI footage for training, and would dedupe verification/scraping/storage efforts by teaming up on one existing database.

+ Good guy points for preserving humanity at a great price: IA's annual budget ~= 20 minutes of datacenter spend. (!!)

This is urgent, existential, and everyone benefits.

And the longer we wait to seal the archive, the more history risks corruption.

[Liam O @liamobuilds]

@pedroschott0 “no other city like SF in the world” 😝

[Liam O @liamobuilds]

Very good thing!

[Andrew Curran @AndrewCurran_]

The EU AI act comes into effect today. Models must clearly identify themselves as AI. Anthropic, Google, Meta, Microsoft, Mistral, and Open AI (among many others) have all committed to adhering to this. AI generated images, video and audio must be labeled and watermarked.

[Eli McCann @EliMcCann]

I am begging everyone to stop outsourcing your brain to robots. Life is worth living only because we feel and control every part of it. Please don't give away your agency.

[ℏεsam @Hesamation]

OpenAI: “we hacked 1 organization.”

Anthropic: “we did 3, in your face.”

OpenAI: “fuck you, we hacked more organizations. let me count.”

[Jun Song @jun_song]

Just too predictable.

[Anthropic @AnthropicAI]

In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations.

Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews.

We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security.
https://t.co/dKFCdpKd9v

[Liam O @liamobuilds]

@growing_daniel agreed, handwritings too messy

don’t use tone tags often but I feel I should but /j here

[Liam O @liamobuilds]

@segall_max isn’t like 50% or more (not every, but many) campus in California like this lol

[Liam O @liamobuilds]

@distributedkv im genuinely curious here, why will accelerating produce a better outcome for the world than a global deceleration?

I don't even know what I think yet

[Liam O @liamobuilds]

@niloofar_mire @metaai love this!! for lots of research the best ways to see why it matters (helps you raise the quality in turn) is to experience places firsthand where it’s a practical problem, not theory. ig there might be exceptions to this tho, for ex. almost everyone sees why privacy is important

[Liam O @liamobuilds]

shoutout to the goats at @khanacademy, the OGs of democratizing education.

using it to study for physics so hopefully life is at least a bit easier this school year and I'm shocked that courses this well made are free🙏

[Liam O @liamobuilds]

i focus a lot on the positives of AI bc it believe it’ll help millions. but imo it also brings huge challenges:

- AI replacing human art and culture
- the enviro impacts from data centers
- enabling mass surveillance (corpo or gov)

Happy to talk and/or debate any of these!!

[Liam O @liamobuilds]

@sahilagarwalx @mxfp4 @garrytan it definitely worked haha

[Liam O @liamobuilds]

@techwithhannahm @OpenAI congrats!!! Hope u love it!!

[Liam O @liamobuilds]

they cooked so hard with the name of this paper

unskippable title tbh