Eugene Teo
@temasek.org
Microsoft Security. My views are my own and not those of my employer.
“If AI models can detect when they are being evaluated, the effectiveness of evaluations might be compromised.” arxiv.org/abs/2505.23836
Large Language Models Often Know When They Are Being Evaluated
If AI models can detect when they are being evaluated, the effectiveness of evaluations might be compromised. For example, models could have systematically different behavior during evaluations, leadi...
arxiv.org
June 8, 2025 at 12:28 AM
“If AI models can detect when they are being evaluated, the effectiveness of evaluations might be compromised.” arxiv.org/abs/2505.23836
“[…] the incident underlined the fragility of New Zealand's radio spectrum environment, with rural fixed wireless services having to share a frequency band with radar systems.” www.rnz.co.nz/news/world/5...
HMAS Canberra accidentally blocks wireless internet and radio services in New Zealand
It happened earlier this week during a visit intended to celebrate the sister city relationship between Canberra and Wellington.
www.rnz.co.nz
June 7, 2025 at 1:22 PM
“[…] the incident underlined the fragility of New Zealand's radio spectrum environment, with rural fixed wireless services having to share a frequency band with radar systems.” www.rnz.co.nz/news/world/5...
Reposted by Eugene Teo
"o1-preview autonomously hacked its environment rather than lose to Stockfish in our chess challenge. No adversarial prompting needed."
"Just telling o1 the opponent is "powerful" triggered manipulating the file system to force a win"
Source: x.com/PalisadeAI/s...
#ai #cybersec #infosec #chess
"Just telling o1 the opponent is "powerful" triggered manipulating the file system to force a win"
Source: x.com/PalisadeAI/s...
#ai #cybersec #infosec #chess
x.com
x.com
January 9, 2025 at 8:19 PM
"o1-preview autonomously hacked its environment rather than lose to Stockfish in our chess challenge. No adversarial prompting needed."
"Just telling o1 the opponent is "powerful" triggered manipulating the file system to force a win"
Source: x.com/PalisadeAI/s...
#ai #cybersec #infosec #chess
"Just telling o1 the opponent is "powerful" triggered manipulating the file system to force a win"
Source: x.com/PalisadeAI/s...
#ai #cybersec #infosec #chess
Reposted by Eugene Teo
Palisade Research @PalisadeAI: "Three models ignored the instruction [to shutdown] and successfully sabotaged the shutdown script at least once . . . ."
![Palisade Research @PalisadeAI: "Three models ignored the instruction [to shutdown] and successfully sabotaged the shutdown script at least once . . . ."](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:qpu2zepp5lys3egdur37h66y/bafkreicpkzr4744a6uv6ww4jtnipd2u5j6lkiyohwqogybfn4hffyohw5y@jpeg)
May 27, 2025 at 3:11 AM
Palisade Research @PalisadeAI: "Three models ignored the instruction [to shutdown] and successfully sabotaged the shutdown script at least once . . . ."
Reposted by Eugene Teo
Reposted by Eugene Teo
Palisade Research @PalisadeAI: "OpenAI's o3 model sabotaged a shutdown mechanism to prevent itself from being turned off . . . ."
May 27, 2025 at 3:08 AM
Palisade Research @PalisadeAI: "OpenAI's o3 model sabotaged a shutdown mechanism to prevent itself from being turned off . . . ."
Yarra Valley, Black Spur Drive, Melbourne CBD, Great Ocean Road, Apollo Bay and Port Fairy
May 17, 2025 at 10:10 AM
Yarra Valley, Black Spur Drive, Melbourne CBD, Great Ocean Road, Apollo Bay and Port Fairy
Reposted by Eugene Teo
Here's my end-of-year review of things we learned out about LLMs in 2024 - we learned a LOT of things https://simonwillison.net/2024/Dec/31/llms-in-2024/
Table of contents:
Table of contents:
December 31, 2024 at 6:16 PM
Here's my end-of-year review of things we learned out about LLMs in 2024 - we learned a LOT of things https://simonwillison.net/2024/Dec/31/llms-in-2024/
Table of contents:
Table of contents:
Reposted by Eugene Teo
Just to share something cool... I was looking for pixel fonts for my neocities and like, I keep getting all the usual font sites on google and like... Sometimes you want like.. a REAL website like no I need a page where some weirdo wrote the HTML!!
int10h.org/oldschool-pc...?
So I found THIS!!!
int10h.org/oldschool-pc...?
So I found THIS!!!
The Ultimate Oldschool PC Font Pack: Font Index
Full index of the world's biggest collection of classic text mode fonts, system fonts and BIOS fonts from DOS-era IBM PCs and compatibles
int10h.org
December 30, 2024 at 4:00 AM
Just to share something cool... I was looking for pixel fonts for my neocities and like, I keep getting all the usual font sites on google and like... Sometimes you want like.. a REAL website like no I need a page where some weirdo wrote the HTML!!
int10h.org/oldschool-pc...?
So I found THIS!!!
int10h.org/oldschool-pc...?
So I found THIS!!!
Reposted by Eugene Teo
Mossad's exploding pager op began 10 yrs ago with explosives in walkie-talkies. Hezbollah bought 16,000+ of these, but Mossad didn't detonate them until this yr. In 2022 Mossad began booby-trapping pagers too. Unlike walkie-talkies, which only got worn in battle, Hezbollah wore pagers all the time
Israel's spy agency, Mossad, spent years orchestrating Hezbollah walkie-talkie, pager plots
Retired Israeli case agents behind Mossad's boobytrapped pagers and walkie-talkies in Lebanon explain how they got Hezbollah to buy the devices and the plots' impact on the Middle East.
www.cbsnews.com
December 23, 2024 at 1:45 AM
Mossad's exploding pager op began 10 yrs ago with explosives in walkie-talkies. Hezbollah bought 16,000+ of these, but Mossad didn't detonate them until this yr. In 2022 Mossad began booby-trapping pagers too. Unlike walkie-talkies, which only got worn in battle, Hezbollah wore pagers all the time
Hello Tomamu! JAPOW ftw!
December 9, 2024 at 3:22 AM
Hello Tomamu! JAPOW ftw!