ChatGPT's Knowledge Gaps Exposed: Purdue Study Finds 52% Wrong Answers
A recent study by Purdue University analysed ChatGPT's responses to 517 Stack Overflow software engineering questions. The study assessed the correctness, consistency, comprehensiveness, and conciseness of ChatGPT's answers. However, the study found that ChatGPT produced incorrect answers to more than half (52%) of the questions.
This highlights the limitations of ChatGPT and how it can provide inaccurate or unreliable information, partly due to being trained on outdated and potentially misleading text conversations.
Anjan Kundavaram, Chief Product Officer at Precisely, comments that it's essential to examine the integrity of the data ChatGPT is trained on. He explains that the accuracy and reliability of AI models depend heavily on using high-quality, unbiased, and current training data. By prioritising data integrity and monitoring for biases, businesses can ensure AI generates trustworthy insights. The study reinforces the need for oversight when utilising AI like ChatGPT which may have flaws.
Hackers Take Aim at AI Systems in Public Red Team Challenge
Thousands of security experts and hackers recently competed in the Generative Red Team (GRT) Challenge at the Defcon conference in Las Vegas. The contest involved trying to expose flaws and biases in chatbots and text generation systems from major AI companies like Google, Meta, OpenAI, and startups Anthropic and Cohere.
Participants attempted challenges designed to overcome safety features and coax harmful text from the AI systems. The public red team exercise allowed leading firms to test their AI models with a wide range of hackers and students. Organisers say the scale helped identify vulnerabilities that can now be patched.
The GRT Challenge builds on earlier public AI testing events and signals growing concern about generative AI's risks. Groups like Microsoft see value in democratising red teaming to improve AI security. However, regulation around testing AI systems before deployment remains minimal in the US. The challenge highlights the need for accountable AI development as the technology advances.
ABBYY Leads Inaugural Intelligent Automation Month
As intelligent automation becomes more prevalent, ABBYY is spearheading the first-ever Intelligent Automation Month this September. ABBYY will partner with leading automation platforms like Doculabs, FormTran, Pipefy, and others for digital events exploring the value of intelligent automation across industries.
According to ABBYY CMO Gabrielle Lukianchuk, the goal is to "help businesses cut through the hype and hear first-hand how [they] are combining experience and innovation to enable maximum ROI." With involvement from major players in automation, Intelligent Automation Month aims to promote an understanding of how intelligent automation can drive efficiency.
领英推荐
Ollie Sulley features as a guest on The Digital Workspace Works Podcast
Our very own Ollie Sulley Sulley was a guest on The Digital Workspace Works Podcast, and the episode is live!
Dive into the impact of AI on roles, from prompt engineering to the deep fake dilemma, while unravelling the nuances of the technology hype cycle.
Join Ollie and Ryan Purvis for a revealing conversation about AI's influence on job evolution and the future of tech careers.
Episode 3 of The Edge of Transformation Podcast is now LIVE.
In episode 3 of The Edge of Transformation, Harrison Goode is joined by Ivo Knejp , COO and co-founder at Pointee .?
Ivo shares with Harrison some of the pulse-racing ways he likes to spend his downtime and how these activities translate to how he runs Pointee, his love of travel, and his affinity for tech from a young age.?
We hope that you find these resources useful and enjoy hearing about the latest Edge Tech updates. If you’d like to receive more news like this, don’t forget to subscribe to The Edge!?
Edge Tech
???? +44 1908 382 398
???? +1 628 254 5056