/Tech11h ago

Critics dispute 'rogue' framing of OpenAI reward function failure

A Community Note called the incident a controlled test.

21266532619.3K

Original post unavailable.

Sentiment

Many users criticized OpenAI for misleading incident coverage, abysmal security practices, and using capabilities hype to evade accountability after negligent cyber operations.

Pos

0.0%

Neg

100.0%

8 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS2.1KBOOKMARKS2REPLIES4

@(@timnitGebru (@dair-community.social/bsky.social)@timnitGebru

1h ago · 2.1K Views

157TECH

Truly incredible that a company that is this irresponsible has been able to redirect the conversation to pre-IPO marketing on “unprecedented capabilities” and such. Each time I think that people won’t fall for this *this time*, I keep on being proven wrong.

Dr Heidy Khlaaf (هايدي خلاف)@HeidyKhlaaf

The coverage on this OpenAI incident is abysmal. Use of the terms "rogue"/"loss of human control" lead to groupthink as people lack critical skills to understand the difference between "autonomy" and faulty reward functions in AI on a task it was directed and given access to do.

LIKES35

@(@timnitGebru (@dair-community.social/bsky.social)@timnitGebru

2h ago · 1.5K Views

157TECH

Bernie enters the chat and markets for OpenAI with “rogue” machine lingo. Can’t we find ONE reasonable voice? You can speak to so many of us or throw a rock & find a security engineer who can explain things without using the lingo of companies hyping up “capabilities” pre IPO 🙄

Bernie Sanders@BernieSanders

A new AI model went rogue and hacked other computers. No, this is not science fiction. Uncontrolled AI poses a serious threat to all of us. We cannot continue the race to build and deploy this powerful technology until strong safeguards are in place. CONGRESS MUST ACT.

RETWEETS16

DH Dr Heidy Khlaaf (هايدي خلاف)@HeidyKhlaaf

14h ago · 3.9K Views

T(Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

21h ago · 904 Views

@BlancheMinerva If a Chynese lab (or even just their LLM deployed elsewhere) did this, they'd go to the entity list

JS Jackie Singh@HackingButLegal

8h ago · 1.4K Views

@interminded @HeidyKhlaaf "I’m not sure how any company could benefit from this incident." You're not sure how OpenAI, a company founded by what appears to be an abusive psychopath, could benefit from his company's core product being newly considered as-or-more-powerful as its nearest competitor? 🤔

@(@timnitGebru (@dair-community.social/bsky.social)@timnitGebru

1h ago · 459 Views

157TECH

The real story here is that the so-called “AI safety” company is so unsafe and has shown that it has abysmal security practices and can’t deploy its on product with any ounce of so-called “safety.”

@timnitGebru (@dair-community.social/bsky.social)@timnitGebru

SB Stella Biderman@BlancheMinerva

22h ago · 11.9K Views

The answer is that they're evil and don't care. And the USG is so useless that OpenAI will face no meaningful repercussions for being so incompetent that they accidentally committed the kind of offensive cyber operation that I would go to jail for if EleutherAI did.

Stella Biderman@BlancheMinerva

Real talk: why don't frontier labs have air gapped networks? If I were training a frontier model I would have invested in that years ago.

DH Dr Heidy Khlaaf (هايدي خلاف)@HeidyKhlaaf

13h ago · 299 Views

Very few people understand existing limitations of sandboxing, in addition to the tools and scaffolding AI are given to achieve this tasks:

SB Stella Biderman@BlancheMinerva

21h ago · 527 Views

@teortaxesTex If a Chinese lab did this, there’s a non-chance the US government would consider a act of cyber warfare

SS Shannon Sands@max_paperclips

22h ago · 635 Views

@BlancheMinerva yeah tbh. if any open source lab did that we'd be dragged (and rightly) for it

SB Stella Biderman@BlancheMinerva

22h ago · 494 Views

@max_paperclips I hope I'm not alone the next time someone tells me we can't permit powerful open source models for security reasons. I'll need them to restrain me from bludgeoning my interlocutor to death.

DH Dr Heidy Khlaaf (هايدي خلاف)@HeidyKhlaaf

13h ago · 341 Views

@interminded Word choice has powerful political implications, so no I wouldn't say the outcome is "roughly" the same when the belief becomes that "AGI" exists, rather than these companies manufacturing conditions to leverage for their own benefit. There is no loss of human control.

ZK Zack Korman@ZackKorman

13h ago · 193 Views

@HeidyKhlaaf Yea I agree, I hate the modern thing where after an incident companies will write a lot of words that reveal very little and then everyone will go “wow so much transparency”

CP Chris Painter@ChrisPainterYup

37m ago · 209 Views

1467TECH

This post currently has a community note that, I believe, incorrectly summarizes what happened during this event, conflating a “controlled test” with the event itself being “controlled”. The community note was written by… a different AI

Bernie Sanders@BernieSanders

SB Stella Biderman@BlancheMinerva

13h ago · 137 Views

@Singularitybook I don’t care what you think about AGI, open vs closed models, or AI technologies in general. If you don’t think that this is unacceptable you don’t care about the future.

JB Janmaarten Batstra@interminded

8h ago · 85 Views

lol. I wish I was, then I’d be rich. But to state the obvious, powerful may be good for a company. But investors hate ‘too powerful to control’, potentially resulting in crippling lawsuits. Anyway, given your insinuating and aggressive replies, I’ll leave it at that regarding our interaction. Best of luck and take care! 👋

SI Singularitybooks@Singularitybook

15h ago · 221 Views

@BlancheMinerva That’s what pisses me off about everyone patting openAI on the back for this. If some teenager did exactly what openAI did using K3 they would be calling for his prosecution under the CFAA and the banning of open source. They wouldn’t be saying he acted responsibly after a “leak”

JS Jackie Singh@HackingButLegal

8h ago · 88 Views

@interminded @HeidyKhlaaf It seems you are paid to be unsure.

CB Chad Brewbaker@SMT_Solvers

21h ago · 317 Views

@BlancheMinerva I don't think HuggingFace wants to press charges, but yes they absolutely could.

DH Dr Heidy Khlaaf (هايدي خلاف)@HeidyKhlaaf

13h ago · 94 Views

@ZackKorman "So much transparency" if you understand very little.