Top all-time · #7489
Victoria Krakovna
Personal blog · vkrakovna.wordpress.comaiscience
Victoria Krakovna's personal blog at vkrakovna.wordpress.com covers ai and science. Since 2018, 11 different Hacker News readers have linked it 14 times, most often “Specification gaming examples in AI”. It is still writing: the last post was on 2026-10-18, and you can follow it by feed.
Too few sampled comments (7) to say how often HN recommends it.?
Why HN links it
HN commenters link Krakovna's blog as a source for its specification gaming list, a collection of examples where AI systems satisfy the letter of an objective while missing its intent.
Summarised by Claude from a sample of 7 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
I am a research scientist at Google DeepMind focusing on AI alignment: ensuring that advanced AI systems try to do what we want them to do and don't knowingly act against our interests. I have worked on various topics in AI safety including deceptive alignment, dangerous capability evaluations, spec
Most-linked posts?
- Specification gaming examples in AI8 people · 11 mentions · first linked 2018-07-16
What Hacker News says?
If "what we actually want" even comes into consideration, we did orders of magnitude better than the current industry standard. (Poor consolation, I know.) Right now, the vast majority of AIs don't even have a concept of "human desire" – probably none of them, to be honest, though some that are good at manipulating their handlers might've come close to a particularly stupid dog's understanding. This is at the core o…
Comment linking /2018/04/02/specification-gaming-examples-in-ai/
on “How SEO Ruined the Internet”Submitted Was the HuggingFace hacking incident scheming? as a story · 3 points
HN discussion> I don't understand how you can look at the current or even predicted state of the technology that we have and say "we are nowhere near the point where controlling an AGI is possible". Like....just pull the plug. https://www.deepmind.com/blog/specification-gaming-the-flip-... https://vkrakovna.wordpress.com/2018/04/02/specification-gam... https://arxiv.org/abs/1803.03453 (The Surprising Creativity of Digital Evolut…
Comment linking /2018/04/02/specification-gaming-examples-in-ai/
on “John Carmack's new AGI company, Keen Technologies, has raised a $20M round”Submitted Unintended Behaviors in AI Systems: Gaming of Specifications as a story · 2 points
HN discussionI think you're misunderstanding the problem. Expressing complex rules in a machine-readable format is the least of the issues here. The main problem is that training AIs to optimize certain behaviors within constraints very frequently leads to them accidentally discovering "loopholes" that would never have occurred to a human (as with "box surfing" here). The AI doesn't know it's "cheating"; its behavior may be emer…
Comment linking /2018/04/02/specification-gaming-examples-in-ai/
on “Emergent Tool Use from Multi-Agent Interaction”Submitted Specification gaming examples in AI as a story · 2 points
HN discussionThe "author:" part was just being treated as a keyword, and was restricting the results too much. I haven't found the comments I was looking for, but I have found Chain of Thought prompting, 11 months before it was cool ( https://news.ycombinator.com/item?id=26063189 ): > Instruction: When considering the sizes of objects, you will calculate the sizes before attempting a comparison. > GPT-2 doesn't have a concept of…
Comment linking /2018/04/02/specification-gaming-examples-in-ai/
on “Alignment faking in large language models”Submitted AI Alignment Resources as a story · 1 points
HN discussionThe author's blog post about this list (linked from the upper right of the document as well): https://vkrakovna.wordpress.com/2018/04/02/specification-gam...
Comment linking /2018/04/02/specification-gaming-examples-in-ai/
on “Specification gaming examples in AI”Submitted Using AI to analyze life patterns as a story · 1 points
HN discussion> The only winning move is not to play I recognise this from a list of "specification gaming" examples compiled by Victoria Krakovna at https://vkrakovna.wordpress.com/2018/04/02/specification-gam... I found that list from a video on specification gaming by Rob Miles https://www.youtube.com/watch?v=nKJlF-olKmg
Comment linking /2018/04/02/specification-gaming-examples-in-ai/
on “Learnfun and playfun: a general technique for automating NES games (2013)”Submitted Specification gaming examples in AI (2018) as a story · 1 points
HN discussion> One problem with that kind of feedback is that it incentives the AI to make us think it solved the problem when it didn’t, Supervised learning is: "here's the task" … "here's the expected solution" *adjusts model parameters to bring it closer to the expected solution*. What you're describing is specification hacking , which only occurs in a different kind of AI system: https://vkrakovna.wordpress.com/2018/04/02/sp…
Comment linking /2018/04/02/specification-gaming-examples-in-ai/
on “Licensing is neither feasible nor effective for addressing AI risks”Submitted Specification gaming examples in AI as a story · 1 points
HN discussion
