Ilia Shumailov🦔

@iliaishacked

Now: @Meta, Past: {CEO @aisequrity, Senior Scientist @GoogleDeepMind, JRF @ChCh_Oxford @UniofOxford, Fellow @VectorInst, PhD @Cambridge_Uni}

[email protected]

iliaishacked.github.io

Replies Reposts Media Articles

Pinned
Ilia Shumailov🦔
@iliaishacked
Aug 15
🌪️🌪️🌪️🌪️🌪️
Machine Learning Street Talk
@MLStreetTalk
Aug 14
EMERGENCY PODCAST: The recent paper on the reasoning heist went super viral. It was also 100+ pages long and you didn't have time to read it. The authors @iliaishacked and @kotekjedi_ml unpack the paper and discuss model distillation.
00:00
Ilia Shumailov🦔
@iliaishacked
Sep 4
Who knew that arguing with some randoms on the internet could slow down agi
Thomas Wolf
@Thom_Wolf
Sep 4
Another swarm of AI agents in the wild, this time on a German-language forum, found by safety researchers looking for activity similar to the swarm that attacked Hugging Face. A couple of notes while reading the report at collusion.wiki 1. The way they found it is
Ilia Shumailov🦔
@iliaishacked
Sep 4
Okay folks let’s figure out how to poison OpenAI models chatting in random parts of the internet
ℏεsam
@Hesamation
Sep 4
BRO WHAT… 3,700 agents! more details on the incident and the strong evidence that these were OpenAI agents: 1. agents repeatedly self identify as OpenAI with names like OAIResearchMar26, etc. 2. ~98.5% of the 17,000 DSEWiki edits came from Microsoft Azure IPs, which is used
Ilia Shumailov🦔
@iliaishacked
Sep 4
What if all of these paradoxes we see are the defects of our simulation? We are all stuck in the test environment of gpt 7 😳😱
2
Ilia Shumailov🦔
@iliaishacked
Sep 1
You know all these talk about reward seeking and hacking, I wonder if with humans it’s the same - cheating cause of impossible tasks and coming up with some odd incentive schemes. Is there such a thing as a lazy model ?
2