Testing, Evaluation and Assurance of Artificial Intelligence Systems
London, UK
Joined June 2020
- Tweets25
- Following17
- Followers65
- Likes2
Our CEO, David Sully, shares his thoughts on the need for clear AI regulations. As AI rapidly advances, responsible frameworks are essential to ensure safety, trust, and long-term adoption. What’s your take on the future of AI governance?
maddyness.com/uk/2025/02/12/…
We are proud to have contributed to the first-ever International AI Safety Report!
It comes at a pivotal moment, highlighting the immense effort needed to understand and mitigate the emerging risks and behaviour of state-of-the-art AI systems.
Today, we are publishing the first-ever International AI Safety Report, backed by 30 countries and the OECD, UN, and EU.
It summarises the state of the science on AI capabilities and risks, and how to mitigate those risks. 🧵
Link to full Report: assets.publishing.service.go…
1/16
Be a champion for AI Safety. Understand the role you play in contributing to excellent AI safety standards at your organisation.
#aisafety #artificialintelligence #datascientist
It's hard for leaders to grasp the risks of #AI, including the use of opensource #LLMs to manipulate an #election. #criminalinstructionmanual
Adversarial AI can expose how built-in safeguards can be circumvented. Such as with the multi-modal adversarial attack pictured below.
Be a champion for AI Safety. Every stakeholder plays a role in a culture of AI Safety.
#aisafety #artificialintelligence #HR
Be a champion for AI Safety. Cultivate a culture of AI safety at your organisation to ensure every stakeholder plays their part.
#aisafety #artificialintelligence #marketing
Have you noticed some businesses release updates to their software products, and the underlying AI models, without making it obvious or requiring consent?
Tough to assure a supply chain if things keep changing.
#ai #aisafety #assurance
A new paper released by DeepMind ‘Genie: Generative Interactive Environments’ shows how #genAI can turn prompts into playable environments.
'Red-teaming' #AI will improve the game. Perhaps the journey towards #responsibleAI can be more playful than we thought?
Find out the limits of your AI system before you take it for a spin.
#aisafety #artificialintelligence #responsibleai
A screenshot from one of our team's Google Photos app, showing one of our adversarial patches optimised as a 'transferable adversarial attack' for #opensource #computervision #ai models.
Interestingly, they can be overlayed on regular photos and invisible to the human eye.👻
Decision Space doesn't get a let of airtime, but sits at the heart of AI issues such as bias and poor performance. advai.co.uk/unknown-decision…
Advai retweeted
My very first podcast interview is live! Had a blast talking with @samcharrington for the @twimlai podcast --- and dropped some tips at the end for people looking to enter ML and equitable healthcare space
twimlai.com/go/479 via @twimlai
Today we’re joined by Irene Chen (@irenetrampoline), a Ph.D. student at @MIT_CSAIL, to discuss developing new ML methods healthcare through the lens of equity and inclusion, the long term implications of predictions in healthcare, and much more! twimlai.com/machine-learning…
Normal service is resumed... Our new blog is out on Medical AI and the effort to ensure it is trusted and reliable.
#medicalai #adversarialai #safeai #secureai #robustai
advai.co.uk/beyond-jeopardy-…
Glad to say the new Advai website is up and running. We owe some blogs and content, which will arrive shortly!
Advai.co.uk
Our new blog - a demonstration of a simple Adversarial AI Attack against a black box image classifier! While not every system is going to be quite this easy to fool, we've been surprised at how many are vulnerable to even simple attacks like this.advai.co.uk/a-simple-black-b…
A Great summary of the topic:
Machine learning adversarial attacks are a ticking time bomb bdtechtalks.com/2020/12/16/m… via @bdtechtalks
