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.
SIKE on the Digital Nexus podcast
First up, it was great to see SIKE on show in the new Digital Nexus podcast we run in conjunction with Digital VIllage. In the examples section, we showcased how SIKE can make content consistent with your writing style (see 37:45 of the video)
Awards time
In the local community, it’s time to give a big congratulations to Sarah Kaur for winning the award for Creative Industries award at the Women in AI Asia Pacific awards last weeek.
See the post here.
DSAI event recap with NVIDIA
Michael Wang. President of the Data Science and AI Association of Australia (DSAI) here with a post highlighting the awesome event DSAI did with NVIDIA last week. See his post here.
New Australian AI startup, Redactive
An AI platform helpinig software developers build AI applications and created by ex-Atlassian stuff certainly sounds interesting. See the AFR article on this here.
Sam Altman and his Aspen Ideas
Sam Altman, speaking at the Aspen ideas festival said that there was going to be a significant leap forward when it comes to GPT5. He also spoke about the need for continuous release of product and iteration because the alternative is that a model just shocks the public when it comes out, whereas iterative releases allow for a slow buildup. That’s a good point and contrasts the critiques of announcing things too early Or when they’re not ready for release. Certainly calls into question for product builders, do you just release and continue iterating or wait until something’s more complete. It depends we guess on whether or not you’re building something like OpenAI and gamechanging technology. See the Tom’s Guide post on this here.
Also, here’s the video replay on YouTube
Microsoft partner to further AI in education
Microsoft announce their partnership with HEDx to further what’s being done in the world of education. More on this in Katie Ford’s Linkedin post.
Cohere and the efficiencies of AI
Various topics discussed here from Coheres CEO on podcast talking about their learnings and perseverance training models on the face of others doubting AI capabilities. At 16.30 the SVIC guys talk about something we’ve seen is a basic truth that keeps getting downplayed by some and that is the ability for these tools to improve processes. They describe where a doctor has very little time to go through a patients history so will use AI to give a summary of that just before seeing them and saving time doing so. Perfect use case.
New Attack vectors thanks to Gen AI
Generative AI opening up new attack vectors. See the ZDNet article here.
Kyutai makes it mark
French AI company, Kyutai made a splash last week with super low latency voice to voice model which they’ve made open source. Here’s a a post where we showcase us trying out the technology. It gets a bit confused but that will no doubt improve over time. Link here.
You can try it out at Kyutai’s website where you choose whether you access the US or European servers. Website here.
Perplexity issues highlight a blockchain tech opportunity
We saw our friend, Gareth Rydon at Friyay post on his love of Perplexity, the AI + search tool that’s rivalled Google (and currently beating it whilst Google works out the kinks in its AI-enabled search that was giving funny answers last month). A comment in there highlighted that Wired was suing Perplexity for not adhering to acceptable internet practice of not scraping when the robots.txt page on the website doesn’t allow you (for more on robots, see this summary).
What’s interesting here is that we can see things like this as a use case for blockchain where you have tighter restrictions and even firewalls tied to smart contracts that allow you to scrape only if you’ve been white-labelled which may be tied to you being a current subscriber or not. Additionally, if transactions are small and pay as you go, you could use blockchain for the various micropayments that would be made here. Traditional financial systems aren’t great at micropayments because of the cost but blockchain is built for this.
A combination would be great but we shall see. Read more of that thread here.
Training the training data, but with issues
The LAION-5B dataset, a model used to train many other AI models, has been found to contain identifiable photos of Australian children and there are likely more according to an investigation by Human Rights Watch. Check out the article here.
SIKE for creating content in your style
We’ll leave you with an example we showcase in our PromptHub document which showcases how you can get SIKE to write content in a similar way to your style of writing (or that of your company). We use Dreams for this and show step by step how you can do it. Check that out here inside our SIKE Product Guide.
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.
