AI Top-of-Mind foe 6.14.24–3D Me
Today: Unique3D, AI-driven UX design, OpenAI’s Whisper hallucinating, McKinsey on Tech Services and AI, and can AI deliver good feedback?
Top-of-mind for today is creative. Jim Clyde Monge writing in ‘Generative AI’ describes great new tool from Tsinghua University, Unique3D, that converts simple 2D images into 3D animations. He includes a link for access and describes the architecture. To dive deeper, a link to the original paper is here. An example with my two dogs, showing pretty good background removal:
More on the creative front, but this time looking at the evolution of the user experience, or UX and some compelling statistics. Mark Seefelder in ‘UX Collective’ outlines the different capabilities of a ‘GenAI-first UX:’
- Content: Personalized text or media, like using a user’s name in greetings or adjusting articles based on age or cultural background.
- Tone: Adopting a formal tone for professional users while switching to a casual tone for younger users.
- User Journeys: Customizing the sequence or pathway a user follows based on their objectives and familiarity with the platform.
- Layouts: Adapting visual arrangements to user preferences or devices, such as grid-based or list-based layouts, dark mode, or light mode.
And he goes on to describe a typical future user journey as well as the differences in adding citations to a research paper:
Imagine this: you need new running shoes, and your phone recommends a pair with the perfect level of support based on your recent workouts. No more wading through generic options — just the perfect fit for you at that specific moment.
And the latest on hallucinations. Some analysis by Cornell University on OpenAI’s ‘Whisper’ speech-to-text feature and its tendency to insert words and even sentences, some of which may be harmful. There were larger issues with those who had speech difficulties. The author does note that it may have improved since the study was conducted, but a good heads-up for any of these platforms. From the post:
In their analysis, researchers found that roughly 1% of Whisper’s audio transcriptions contained entire hallucinated phrases — including references to websites, real and fake, that could be reappropriated for cyberattacks. For instance, in one sound bite, Whisper correctly transcribed a single, simple sentence, but then hallucinated five additional sentences that contained the words “terror,” “knife” and “killed,” none of which were in the original audio.
Another bit of research from ‘McKinsey,’ this time looking at the future of tech services in the face of increasing enterprise AI spend with a goal of streamlining some of these services. The future for SIs/MSPs is to offer new services. From the research:
The next 12 to 18 months will be pivotal. Enterprise customers are already exploring new ways to manage some core IT work themselves while ramping up all manner of gen AI pilots and initiatives, and a wide range of other tech players (from hyperscalers to hardware and software companies) are beginning to make or contemplate moves into the burgeoning AI services market. Traditional services providers that don’t begin to reimagine their value proposition in this arena risk losing some relevance — and potentially more than 15 percent in revenue and profit.
Next to last, what type of feedback does AI offer, and can it replace humans? ‘The Horizons Tracker’ looks at this question, with a survey showing that employees are more likely to trust initial feedback from AI rather than peers, but that they still need advice from humans. Some observations:
· “The study finds that such purely objective evaluation of failure provided by machine, whether they are GenAI or more simple software, can help someone learn better from failure based on human evaluation as well, so that’s an important finding,” the authors continue.
· The study also shows that when machines give feedback on mistakes, people are more likely to learn from their peers’ feedback too.
· People were happy with an AI boss so long as everything was running smoothly. If there were disagreements or things the employees wanted to change, then it resulted in a rapid deterioration in the relationship.
· This can have implications in areas like negotiations, where research from the University of Southern California found that virtual agents can often be more effective at negotiating than we are.
Finally, on the back of the Apple Intelligence launch, how does their approach to AI capabilities compare with Samsung? ‘CNET’ looks at the two, with a view that Apple is more focused on privacy, while Samsung’s current capabilities are more aligned to communications and productivity. But there is of course a lot of overlap.
