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.
Week full of AI Events
What a great week in the world of AI with some big events this week. We attended the AI Build Club Demo Day on Monday night at The Collider, in Tech Central, Sydney. 5 of the 15 groups pitching getting to go to San Francisco to demo over there too – LinkedIn
Here’s Annie’s write up on it too – LinkedIn
Unlocking expert capital
Sharing a LinkedIn post here. From Ethan Mollick. This highlights the need for experts in the loop when it comes to AI processes and it’s certainly something that we’ve been seeing as well and why we’ve built SIKE the way that we have. Recent research put out by Francois Chollet show that you need knowledge of certain problem solving techniques to get the best out of AI and given that experts can be the ones that input that into a model, whether it’s via documents or instructions or notes in SIKE, that’s where the best knowledge outputs come from. This is especially more so because experts are the ones that can actually tell whether or not an output is useful or not. That said, AI can still be very useful to help level up your knowledge and skills no matter your level of expertise. – LinkedIn
Anthropic’s Claude Sonnet 3.5 update
Anthropic dropped an update to their Claude model with Sonnet which looks quite interesting (despite some initial message limits) – Ethan’s X Post
Its artifacts feature seems like a more feature rich version of the function calling we get in ChatGPT. X post from Skirano
I had my own take on it here too where I showed it being used as a way to visualise explanations of how ETFs and blockchain works – see X post here
GPT2 as good as neuroscientists?
Some really interesting points shared in this video on an AI paper showing how LLMs, when pre-trained on a specific field (in this case neuroscience) can be an even better performer than human equivalents.
What’s interesting here for us is that they highlighted how its better to be domain specific because just a generic vectorisation of words would cut up a complex word in neuroscience like “excitatory postsynaptic potential” into shorter parts. Domain specific ones would keep the word as a single token and therefore be able to perform better at next word prediction.
The video also highlights that it could get interesting if you perform scientific tests with a null and alternative hypothesis and get the domain-specific AI to do their own sets of testing and compare that versus what a human would have output. We allude to this sort of testing as well when it comes to our AI versus series especially the first episode where we looked at what SIKE would do if it were to start an association (like the Aus DeFi Association we built not long ago).
Very interesting seeing AI, when used for what it’s strong as, see amazing results.
Further on that thinking, there’s another thing we know AI is good at when it comes to data, and that’s anomaly detection. We’re not talking here about AI doing data analysis with function calling help (though that is cool) but rather, the LLM doing some sort of data analysis on its own. Something we’ll show more of in future episodes.
AI for lawyers
An interesting post here on how to use copilot for legal tasks. Worth a look through if you’re interested in ways to use AI in the legal fields in general. – LinkedIn
How Gen AI is leveling the playing field for smaller business
Something we got to know last week but didn’t get to highlight was meeting 2 authors from this Harvard Business Review article on how AI helps small businesses. Kim was on stage and Graham was in the audience as we spoke about AI at the Fintech Marketplace event but we completely agree with the findings they shared that showed how AI can help small businesses. It is game-changing for them, that’s for sure. – Harvard Business Review article
Cisco and AI
Cisco powers a lot of the digital world but many end users have had no idea just how much. That might change now that Cisco is getting more publicly into AI with it now being easier to implement Cisco products into their stack (normally something only available to companies who could afford Cisco software and the engineers needed to run it). Here’s to more companies lowering barriers to entry for all things good in tech. This is great as we’re all going to need robustness when it comes to technology stacks in a future where it’s easier for hackers to try to harm others. Matt Wolfe video below.
Safe Superintelligence on its way
Ilya Sutskever who just left OpenAI has started a new venture called SSI. Here’s a post announcing what we do. X post here
Perplexity vs Wired
Wow, a lot to unpack here in this CNBC video with Wired editor Katie Drummond. The episode is partially about the issue of Perplexity seeming to ignore the robots.txt files that are on websites saying who can/cannot scrape data from them but it also devolved into the host seeming to mix up Google’s AI search issues (coming up with obscure Reddit references) and what Perplexity does. He may well be representative of what mainstream opinions are of AI and the fact he says the only AI he wants is when it has sentience is also bewildering. CNBC itself has a lot of technology behind its news platform and no doubt has aspirations for AI innovations itself. Yet, when its host comes across as he does here, it’s ironic that he had a look of perplexity the whole time.
Interestingly, blockchain technology could better help with agreements on who can scrape versus not when it comes to websites and doing so with micropayments to pave the way making this suitable for publisher and scraper. X post we had on this here
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.
