Welcome back everyone to another AI Insights newsletter by NotCentralised. We explore the interesting things we’ve been finding on the world of AI this past week and synthesise it for the things we think you should care about.
If you’re interested in talking with us about how to get the best out of AI for your work then feel free to get in touch with us either on this contact form or email.
NC Show – Episode 4
In this new episode of The NC Show, we talk to Mark Dwyer from Paiid, who have partnered with us to push SIKE into the financial services space. He just came back from a Visa Conference in the USA and had some great insights for AI in finance.
Canva Create
Canva Create is over and had some great new features driven by AI that were showcased. A great rundown of that is here on X by @CodeByPoonam – link.
Controlling the chaos in AI
Here’s an interesting video from Machine Learning Street Talk which dives into the concept of control theory when it comes to LLMs. Given they are proabilistic models, understanding how they work and what sort of controls we can have on them is very important.
In this talk, what they found was that even if a system is designed to give the next likely token (like most LLMs are), it can be made to give the most unlikely if prompted in a certain way. They talk about the certain degree of control that you give to the outside world to the potential outputs from the system so there is a need to be wary of this and they also discuss how it is better to make changes at the prompt level and increase robustness at the software layer.
This is similar to the reduction of randomness that we’ve spoken about in previous blogs and podcasts (video discussion below).
AI Shaping the Future
Great video from Forbes here on how AI will shape our future
AI Skepticism
But to counter that, and important to note, is this piece covering AI scepticism. The video critically examines Silicon Valley’s culture of hype and speculative investment, highlighting how tech trends like big data and AI are often overhyped to maintain high valuations and investor interest. It argues that many tech companies prioritise showmanship over real innovation, resulting in unprofitable businesses propped up by continuous capital injections. The video suggests that AI is the latest in a series of overhyped trends, following the same pattern as previous technologies that failed to deliver on their grand promises.
Critiques are important but equally so, it’s important to have some perspective too and we have the following thoughts on the critiques of AI
- Incremental Progress: While initial promises may seem overhyped, many technologies, including AI and big data, have led to substantial incremental progress and practical applications across various industries.
- Innovation Ecosystem: The culture of speculative investment and hype fosters a dynamic innovation ecosystem, driving rapid advancements and encouraging ambitious projects that can lead to significant breakthroughs.
- Long-Term Benefits: Even if not immediately profitable, sustained investment in tech trends lays the groundwork for future innovations and industries, contributing to long-term economic growth and societal benefits.
You can watch the full video here
Concerns at OpenAI
Interesting breakdown of a talk by Helen Toner who was on the OpenAI board and was part of the group initially ousting Sam Altman. Some insights here into the inner working from her point of view
Matt Wolfe breaks it down too and seeing those who know AI concerned as they are is certainly worth understanding given the reliance many have on OpenAI.
SIKE Example: Federal Budget Papers
Finally, a reminder to those who might not have seen it, but you can see more about how SIKE works in this example diving into the recent Federal Budget papers.
GAN Progress
Very interesting paper here that takes the copying of human facial expressions to AI avatars to another level (even covering off on wrinkles and more details than before). See the research here.
Deep Fake Tracker
We’ll need to be aware of the AI fakes out there that are increasing, and Wired has been keeping track of this sort of thing (like this deep fake of Donald Trump with black voters). See more here.
Netflix of AI
Taking it further, the team at Fable Studio (who brought us fake South Park episodes made with AI) have come up with the capability for everyday people to prompt and watch their own TV shows – check out this video from Theoretically Media.
Also, our friends at Transhuman Coin have been partnering with Fable to bring this stuff to life – check out this recent video they shared.
The future is coming at us fast, that’s for sure!
Also, if you missed it, below is a video showcasing the different ways in which SIKE can help us understand the Federal Budget. From ad-hoc queries, to AI agents with set parts to look for and even synthesising stories with your very own political economist. Check it out below.
Get in touch
Stay tuned for more insights from us next week and if you’re interested in SIKE (our privacy focused AI tool), then you can check out more on what it’s about here.
If you’re interested in learning more, reach out to Arturo Rodriguez or 📈 Mark Monfort here and get in touch with us here or on email at we@notcentralised.com.
