In a paper from the AI lab Anthropic, which produces the large language model (LLM) behind the ChatGPT rival Claude, researchers described an attack they called "many-shot jailbreaking". It is as simple as it is effective.
Claude, like most large commercial AI systems, contains safety features designed to encourage it to refuse certain requests, such as to generate violent or hateful speech, or produce instructions for illegal activities. A user who asks the system for instructions to build a bomb, for example, will receive a polite refusal to engage.
Diese Geschichte stammt aus der April 04, 2024-Ausgabe von The Guardian.
Starten Sie Ihre 7-tägige kostenlose Testversion von Magzter GOLD, um auf Tausende kuratierte Premium-Storys sowie über 8.000 Zeitschriften und Zeitungen zuzugreifen.
Bereits Abonnent ? Anmelden
Diese Geschichte stammt aus der April 04, 2024-Ausgabe von The Guardian.
Starten Sie Ihre 7-tägige kostenlose Testversion von Magzter GOLD, um auf Tausende kuratierte Premium-Storys sowie über 8.000 Zeitschriften und Zeitungen zuzugreifen.
Bereits Abonnent? Anmelden
At least 75% of universities join fossil fuel pledge, say activists
More than three-quarters of UK universities have pledged to exclude fossil fuel firms from their investment portfolios, according to campaigners.
Verstappen says he has 'lost all respect' for Russell
Max Verstappen issued a condemnation of his fellow driver George Russell stating he had \"lost all respect\" for him after the pair were involved in an incident during qualifying for the Qatar Grand Prix.
'Not the best part of UK' Ortega hits out at Liverpool over Guardiola 'sack' chants
Pep Guardiola said he expected more respect at Anfield after being taunted about the sack during Manchester City's defeat at Liverpool, with the chants prompting the goalkeeper Stefan Ortega to criticise the city as \"not the best part in the UK\".
Salah seals statement win as City continue to flounder
When times have been tough in the past for Manchester City under Pep Guardiola, there has always been the sense they will pull through; it will be OK.
Powerless Guardiola gawps as his empire falls at the hands of Slot's meritocracy
Pep Guardiola kept holding up six fingers. The Liverpool fans were in delirium and the Liverpool players were jigging and jiving across the turf, and his own players had gone over to applaud the travelling support, which is really the least anyone deserves after attempting to travel across northern England on a Sunday.
Rashford and Zirkzee double up to demolish sorry Everton
\"Amorim, Ruben Amorim, nananananananaaa,\" the jubilant Manchester United fans chorused after Joshua Zirkzee's second goal.
Palmer's showstopper adds to Emery's worries
Bad news for fans of slapstick comedy: Chelsea appear to be serious again.
Postecoglou tunes out injury 'violins' as Spurs are slowed by Cairney
It would be tempting to talk of Tottenham at least being predictable in their unpredictability, of the way they cannot but follow up a great result with a disappointing one, of the inevitability of them, having beaten Manchester City 4-0 the previous weekend, failing to beat Fulham at home.
Hayes says fans entitled to boo USA's Albert at Wembley
Emma Hayes United States head coach
'Raring to go' Stokes brushes off injury worry after pulling up
Ben Stokes moved to play down concerns over his fitness and declared himself \"raring to go\" for the second Test at Wellington after the England captain ended his efforts with the ball mid-over during the eight-wicket win at Hagley Oval.
