Ga onbeperkt met Magzter GOLD
Krijg onbeperkte toegang tot meer dan 9000 tijdschriften, kranten en Premium-verhalen voor slechts
Poging GOUD - Vrij
AI models bypassed explicit orders to shut down, say researchers
Hindustan Times
|May 27, 2025
TESTS DISCOVERED THAT CHATGPT 03 CIRCUMVENTED INSTRUCTIONS TO SHUT DOWN AT A HIGHER RATE THAN OTHER MODELS
The latest thinking and reasoning artificial intelligence (AI) models have displayed, at least in testing environments, a tendency to self preserve. That is, they refuse to shut down, even when told to do so.
Tests by Palisade Research discovered OpenAl's latest ChatGPT model, 03, sabotaged shutdown mechanisms to prevent itself from being turned off.
While other advanced models displayed the same behaviour in a series of tests, 03's propensity was more than others. The 03, released a few weeks ago, has been dubbed as the "most powerful reasoning model" yet by OpenAI.
This comes as another AI company, Anthropic, said that its latest thinking model called Claude Opus 4 attempted to blackmail an engineer by threatening to reveal an affair if it is shut down and replaced by another model.
Dit verhaal komt uit de May 27, 2025-editie van Hindustan Times.
Abonneer u op Magzter GOLD voor toegang tot duizenden zorgvuldig samengestelde premiumverhalen en meer dan 9000 tijdschriften en kranten.
Bent u al abonnee? Aanmelden
MEER VERHALEN VAN Hindustan Times
Hindustan Times
Google says it has developed quantum computing algorithm in breakthrough
Google said it has developed a computer algorithm that points the way to practical applications for quantum computing and will be able to generate unique data for use with artificial intelligence.
1 min
October 23, 2025
Hindustan Times
NHRC NOTICE TO 3 STATES’ DGPs OVER ATTACK ON JOURNOS
The National Human Rights Commission (NHRC) of India, on Wednesday, took suo motu cognisance of reported attacks on three journalists in Kerala, Manipur, and Tripura.
1 min
October 23, 2025
Hindustan Times
Women in focus as Tejashwi vows job security, salary hike
Rashtriya Janata Dal (RJD) leader Tejashwi Yadav announced on Wednesday that all contractual workers engaged in various Bihar government departments and around 200,000 \"community mobilisers\" among \"Jeevika Didis\" would be made permanent, and given a monthly salary of ₹30,000, if the INDIA bloc comes to power in the upcoming assembly elections.
2 mins
October 23, 2025
Hindustan Times
'Mini secretariats': Construction kicks off in five districts
The Delhi government has stepped up work on “mini secretariats” in five districts across the city—a flagship initiative announced earlier this year by chief minister Rekha Gupta.
1 min
October 23, 2025
Hindustan Times
Rohingya case points to legal vacuum on refugees
True character is revealed not in comfort, but in crisis. And India has passed that test before.
3 mins
October 23, 2025
Hindustan Times
LOW WIND SPEEDS KEEP CAPITAL'S AIR 'VERY POOR'; RELIEF UNLIKELY
Delhi's air quality remained \"very poor\" for the third consecutive day on Wednesday, deteriorating marginally as wind speeds dropped again.
1 min
October 23, 2025
Hindustan Times
IT ministry tightens rules for online content removal
The government has introduced new procedural safeguards to content takedown rules, restricting the power to senior officials and mandating reasoned orders with monthly reviews.
2 mins
October 23, 2025
Hindustan Times
Minors can, on reaching adulthood, cancel sale transactions made by guardians, says top court
The Supreme Court has ruled that minors, on attaining majority, can repudiate sale transactions executed by their guardians without the court's permission, and it is not mandatory for them to move court to cancel such sales.
3 mins
October 23, 2025
Hindustan Times
Why this season is looking very good for Arsenal so far
Gyökeres ends barren run with a brace as Gunners continue good run by dismantling Atletico
3 mins
October 23, 2025
Hindustan Times
Private sector can help address urban flooding
The torrential rains across India this monsoon once again exposed the deep vulnerabilities of its cities.
4 mins
October 23, 2025
Listen
Translate
Change font size