Q2 2024 AI News: NHS trials, Microsoft and OpenAI's supercomputer, and deepfakes

This quarter read about how AI developments have shown both remarkable progress and concerning challenges. Healthcare applications have demonstrated particular promise, with AI successfully detecting breast cancers missed by human radiologists in NHS trials. Major tech companies are making significant investments, notably Microsoft and OpenAI's planned $100B AI "Supercomputer" project, while businesses are rapidly adopting AI for various applications, from News Corp's AI-generated articles to widespread customer service implementations. However, these advances are accompanied by growing concerns about safety and ethics, particularly regarding deep fakes and AI-generated inappropriate content. The technology's impact on employment, mental health, and education remains a crucial topic of discussion, highlighting the need for careful consideration of AI's role in society as it continues to evolve rapidly.

April 2024

AI Detects Breast Cancers that Radiologists Missed

The NHS in Britain ran a test with 10,000 women where an AI model picked up all the cancers that human radiologists found, along with an additional 11 patients who had cancers that the human clinicians missed. So what: This pilot reinforces results from other trials in healthcare where the involvement of AI can deliver better patient outcomes. We should expect this application of AI to start to go mainstream sooner rather than later. > read more

Open AI's Text-to-Voice Service - too Dangerous to Release

OpenAI has built a text-to-voice model that only needs 15 seconds of recorded audio to perfectly clone someone's voice. So what: Whilst this is very noble of them, it isn't going to materially make the world a safer place because there are a bunch of services that already offer very cheap AI voice cloning - and that do a very good job of it as well (e.g. eleven labs). > read more

Adobe Releases new AI APIs to Speed up Content Generation

The new APIs will allow you to automate content generation using AI at scale. For example you can send through a few sample images and then get hundreds of variations back in any number of different specified dimensions. So what: Combining AI services with automations is going to be one of the practical ways AI transformation of businesses will play out, so adding APIs for all of its AI services is a smart play by Adobe. > read more

Microsoft & Open AI to Spend $100b on AI "Supercomputer"

Basically it seems like this will be a giant AI focussed data centre, potentially powered by its own nuclear power plant. In a cringey movie reference, akin to Elon Musk style naming conventions, the project has apparently been codenamed "Stargate"... sigh. So what: This is Microsoft placing a pretty big bet on AI. This is clearly a pre-emptive attempt to become the dominant provider of AI compute - the core service that will be required to power the coming wave of AI transformation. > read more

The US Air Force is Evaluating if AI can Help it Spy

The US Air Force is testing the use of an AI chatbot to allow its analysts to quickly get summaries of information contained in surveillance footage - for instance it can list all movements of vehicles at a certain location over the past week and any patterns it detects in those movements. So what: AI is very good at providing quick access to information buried in amongst vast datasets, so it makes sense that it could help provide summaries of observations from surveillance images. The scary part is when you start to think about all the potential uses for this sort of capability in non-military applications. > read more

Google Considers Paid AI Search Features

Reliable sources have leaked that Google is considering making AI search engine services a premium option only available to paid subscribers - which, if it happens, would be the first time it has ever put a paywall on any part of it's core search engine services. So what: This is a sign that Google is still grappling with the prospective impacts of AI on its core search engine offerings which lie at the heart of it's market power in the advertising space. It is right to be worried, services like Perplexity which offer AI web search capability are very good - to the point where for a lot of queries I prefer using them over Google for certain types of searches (just give it a try if you don't believe me). > read more

Darth Vader & Princess Leia Dancing in Pulp Fiction

A new AI character animation tool called Viggle hit the headlines this week - it allows you to take video footage and swap out the person with a different character (animated or photo real). So what? The outputs are not perfect, but they are well on their way and open up a different avenue for AI video editing - think Avatar's Navi characters being produced by AI instead of animators. It has also made for some pretty funny mash ups being posted across social media like the one in the link below. > read more

New Models - New Models Everywhere...

This week saw a raft of new versions of LLM's from Google, ChatGPT and Mistral along with Meta flagging an update to their Lama LLM in a few weeks. So what?  The ChatGPT-4-Turbo update is interesting given it has come with a big drop in token price and more capabilities including image input. It is likely the last big update we will get from them before GPT-5. Google's Gemini is also interesting given it can now be fed text, images, audio and video. > read more

AI Video Snippet Generation at Scale

I stumbled across an interesting AI platform that takes long form video, predominantly targeted at sport, and intelligently cuts it into highlight snippets - including scaling them to suit different device dimensions. So what? This seems like a really handy application of AI. It will make it super easy for sports teams to generate more "snippet" style content that is perfect for social channels. > read more

Google releases its Microsoft CoPilot competitor

Google has announced Gemini is coming to Google Workspace - it will be able to help write emails, documents and code and manipulate your spreadsheets. So what? Google is still in catch up mode... but they are catching up. The proof will of course be in the pudding - but happily for Google MS O365 CoPilot is still a little rough around the edges so there isn't a super high benchmark in this space yet. > read more

AI to Save us from Conspiracy Theorists

Some clever boffins from MIT and the like ran a study where they tested to see if trained AI chatbots could "de-radicalise" conspiracy theorists. So what? Initially I was sceptical about whether this could work, but the study changed my mind... Apparently AI is very good at providing compelling rebuttals and it got a bunch of people who were sure of their theories to change their minds - with long lasting effects. > read more

AI Music Generation Keeps Getting Better

A new tool called Udio allows you to create your own AI generated songs with little more than a few text prompts. If you don't believe me click the read more button below to hear the "Revium Anthem" one of our developers generated in a few minutes this arvo. So what? When you combine this sort of tech with AI voice cloning, it won't be long before we are in a place where we could in theory be listening to a very custom playlist - you might tell your AI to make up a song about bunnies sung by T-Payne in the style of The Beatles. Who knows how the rights will be controlled with all of this and how artists will get paid for their work. > read more

THE MOST EXCITING AI THING I SAW THIS WEEK

Open Interpreter is a hardware/software combination that enables an AI model to take control of your computer and then do tasks for you. It controls your mouse and keyboard and approaches interacting with your computer like a human does - seeing things and making decisions based on it's 'intellect' and 'intuition'. So what? AI lacks eyes and hands which limits its ability to interact with the 'real world' - this tool gives it that ability. For now they are highlighting how a human can control it by making voice requests, but just imagine if you allowed another AI to take control... > read more

THE MOST DISSAPPOINTING THING IN AI THIS WEEK

The Humane AI pin is one of a bunch of current attempts to create a device that puts an AI assitant at our fingertips - which many think will eventually replace the smartphone (I find it funny how so many people can't imagine life without a device that has only been around for 20 years). So what? From when they first announced the idea it always looked like a super ambitious project and so I wasn't surprised to see the first gen AI pin fall flat. We need visionary companies willing to take risks as AI evolves and I hope Humane survives this and can bring a v2 AI pin to market. > read more

ADOBE TEASES GENERATIVE AI TOOLS FOR VIDEO EDITING

Adobe's generative AI tool Firefly has already made waves in the still image space through it's integration into photoshop and illustrator, but now they have showcased the next generation of Firefly which will bring AI to video later this year. They also included a hint at integration with third party AI video generation tools like OpenAI's remarkable Sora AI text to video tool. So what? We already know that video content production (including Hollywood) is facing immense disruption from AI in the near future, this announcement just illustrates how democratised AI video production is going to be. > watch video

UK ENERGY RETAILER USING AI FOR CUSTOMER EMAILS

Since February UK Energy Retailer Octopus Energy has started using AI agents to respond to customer enquiries (and it now handles over 30% of them) - albeit a human reviews the response before it is issued. An insider even let go that the AI agent gets better customer satisfaction scores than its much higher paid human counterparts. So what? Every business out there should be starting to explore this approach, it does take some initial investment training the AI, but it works and the ROI is realised in year one  (and is obviously far greater in subsequent years). We are currently working with a number of our clients on exactly this sort of stuff. > read more

AN UNSETTLING AI APPLICATION

I have seen bits and pieces about this sort of use of AI since chatGPT first hit the scene... But this article about how prevalent the practice of "immortalising" a dead family member using AI is in China is a little disconcerting. By uploading content of a loved one you can train a small model on them and it will mimic them in the way it interacts with you. So what? The article even mentions that a funeral home is considering offering the option of having an AI avatar of the deceased person appear at their own funeral. I'm no psychologist, but surely this isn't good for people... at least if it isn't managed in the right ways. > read more

SAM ALTMAN ON LLM COMPUTE AS THE NEW OIL

In the midst of a 2 hour podcast Sam Altman, who is unarguably a very unique character, made some interesting comments around how LLM compute will become the most valued commodity in the world and what that means for demand for power, data centres and AI hardware. So what? And we all thought Bitcoin was bad for the environment and was driving up computer hardware costs. > read more

META'S AI ROLLOUT: WITH A HINT OF WHAT'S TO COME

Unless you were living under a proverbial rock, you would have seen the news that Meta has rolled out its AI "tool" to Facebook, WhatsApp and Instagram in Australia. For now it is not really integrated with your personal data and won't remember things about you - it is more of a standard AI chatbot like ChatGPT with the ability to make AI generated images... So what? For now this is just ChatGPT in Facebook... but... the obvious play here is to introduce AI to millions of people who might not be otherwise inclined to engage with it to warm them up for a subsequent rollout of more advanced capabilities that will require your consent to let the AI access your data. Call me cynical if you like. > read more

TRUMP, OBAMA & BIDEN EXPLAIN MATH

So while you can do all sorts of terrible things with AI deep fake tech, one of the more popular uses for younger folk is to get celebreties to help you learn. There are a bunch of accounts out there producing AI deepfakes of celebs explaining maths and other subjects - and they are extremely popular. So what? Finally a good use for AI deep fakes? I'm not sure how many people are upping their math game by watching these, but there is a weird entertainment value in having Jake Paul and Morgan Freeman chatting through derivatives together. > watch video

IS PERPLEXITY AI THE FUTURE OF THE SEARCH ENGINE

I mentioned in passing a few weeks back that you should try Perplexity as a replacement for Google. Well it seems like a lot of people have been doing exactly that and as a result Perplexity, which is an AI based alternative to Google Search, is now looking at a valuation north of $2.5 billion. Not bad for a company that hasn't hit its second birthday yet. So what? The bigger question is whether there will be what we currently think of as a "search engine" in 5 years. Given more and more of our interactions with digital will likely occur through an AI Assistant as a gatekeeper to digital knowledge and services it may just be that the term "Google It" starts to die off. > read more

REAL TIME TALKING HEADS FROM MICROSOFT

Microsoft showcased some AI tech they have built, but aren't yet releasing, that allows you to take a single photo and then use it to create a talking head video using just an audio feed. So what? I can understand why MS has opted not to release this because of the risks - it screams quick and easy deep fakes. That said, there is an inevitability here and I think we need to accept that we are entering a time where we won't be able to tell what is real or fake when it comes to video and audio (even if it is "live streamed"). > read more

ANOTHER DEPRAVED USE OF AI: NONCONSENSUAL NUDES

A large number of services have popped up offering to use AI to "undress" people - you just need to upload a photo of them anonymously and the tool does the rest. The worst part is the competition for users means many of these services are offering a first photo for free. So what? It's sickening that there are people with the intellect to build AI applications who are so morally corrupt that for a few dollars they are willing to actively facilitate harming innocents in order to titillate the degenerative dregs of humanity. > read more

SLACK JOINS THE AI PARTY: BUT AT WHAT PRICE?

Slack has introduced an AI feature set which paid users can pay extra to get access to (an additional $10 per user per month). Its primary two benefits are conversation summaries and AI powered search capability. So what? While it's a handy add on, the price point seems pretty steep and it will be interesting to see what the take up is like. This highlights one of the challenges in pricing up AI based product add ons - in most cases the cost of AI is based on how much you use it, so it seems they are pricing it in on assumed average user consumption volumes. > read more

May 2024

A team in China released samples of Vidu, their competitor to OpenAI's Sora "text to video" tool. The videos aren't as impressive as Sora, but it does show China is making serious advancements in AI and you can bet they won't have to worry about any AI safety regulations any time soon. > Watch Video

In a practical example of AI in business, some leaked info shows how T-Mobile is rolling out a suite of generative AI tools to its frontline staff to improve customer service and increase sales. > Read Case Study

Mark Zuckerberg let go that more than 50% of the content that people see on Instagram is now AI recommended, which is higher than it has ever been. No wonder my better half prefers Instagram to my company. > Read More

The unnecessary attempts to shove an AI assistant into different hardware form factors had another flop this week, with the Rabbit R1 being savaged just as the Humane AI Pin was a fortnight ago - just let us use our phones. > Read More

That said, there may be potential to fold AI into smart glasses, it seems Meta think this might be a winner - albeit there is still a bit of a creepy vibe to glasses that can be subtly recording videos. > Read More

Speaking of AI wearable glasses, this a very obvious and very cool use of AI to help people with hearing disabilities by adding instant subtitles to live conversations.  > Watch Video

Staying with the topic of AI and hardware, move over Tom Cruise, the US Air Force has announced that more than 6 months ago they let AI fly a fighter jet autonomously in a dog fight against a human pilot. Wild. > Read More

For those of us that hate household chores there is a new robot for that - from cooking, ironing, folding clothes, and vacuuming and watering plants. When can I pre-order? > Read More

News Corp generates 3,000 of it's articles a week using AI - for now its limited content types on regional mastheads that are reviewed by humans.... but we all know where it is heading... > Read Non-AI Article

Here's a cool little TikTok showing how you can use AI tools to purportedly quickly and easily turn some hand drawn sketches into an animated 3D character and use AR to show it in your videos. > Watch Video

Open AI and Microsoft are getting sued, again, this time by a bunch of eight newspapers... I wonder why they never go after Meta or Google with these suits... > Read More

Meanwhile the Financial Times has signed a deal to get paid by OpenAI and also get access to use AI tools on their website - expect to see more of these partnerships popping up over time. > Read More

There is a growing wave of scepticism surrounding the recent demo of the so-called AI developer tool called Devin, this video is an interesting breakdown of why it might not be all it was made out to be. > Watch Video

The AI community is abuzz at the moment with OpenAI news on two fronts. One is that the rumour that OpenAI is about to release a web search tool and the other is that they are secretly testing the newest release of a much more powerful version of ChatGPT. Time, measured in weeks, will tell on both fronts. > Read more

In other OpenAI news, they have continued their "partner with content providers to avoid being sued" approach and struck up a deal with Stack Overflow to train on their content in exchange for officially crediting it. Hopefully it will enhance ChatGPTs coding abilities. > Read more

UPS has been working on rolling AI into it's customer service workflows with what would seem to be great success. Their AI bot reviews incoming customer support emails and drafts responses, which has purportedly reduced agent time spent on emails by up to 50%. Most customer service will be done with AI eventually. > Read more

Watch a short video of Mustafa Suleyman the new CEO of Microsoft AI, and former head of DeepMind, talking about personal AIs and what happens when they get AQ and can interact with the world. He very much believes what he is saying. > Watch clip

Some band I have never heard of has made the "first music video using OpenAI's Sora" text to video. I gotta be honest, next to all the other amazing Sora videos, this is pretty meh... but it does attempt to showcase consistent characters in AI generated video. > See video

Microsoft has developed an air-gapped version of GPT-4 for the US Government's spy agencies to use - apparently the CIA has been frustrated they haven't been able to play with AI like everyone else could. > Read more

In other Microsoft news, they are trying really hard to make CoPilot easier to use for AI newbies, providing automated help to write better prompts. It's probably a good idea given their latest survey suggested 75% of people are already using AI at work (insert grain of salt here). > Read more

AI transformation will be the new digital transformation. Biotech behemoth Moderna are jumping on the AI transformation train early and pushing through a program across their business - with 400 custom GPTs built out in just two months. It is worth noting that they are changing their existing processes to better suit AI. > Read more

In another AI medical story, GPs are increasingly using AI to help with note taking. Whilst I am citing an article here, one of our team members actually had this experience in a Melbourne doctors clinic a few weeks ago. > Read more

More deep fakes and scams. An insect repellent company paid £20,000 to what they thought was a BBC presenter to produce videos endorsing their brand - it turns out it was a scammer using AI to fake the celeb and she knew nothing about it. > Read more

In other deep fake news the Met Gala was hit by a bunch of AI generated social posts of fake red carpet looks, with Katy Perry and Rihanna amongst those targeted (even Katy's mum didn't realise it was AI). AI generated content is about to start posing very serious problems. > See photos

In the lighter side of AI, Reid Hoffman created an AI version of himself and then had a "conversation" with himself - complete with video of AI Reid listening and responding. It's obviously not as real time as it's made to look, but it's funny to watch... > Watch video

I normally never do self promotion, but one of our Lead developers wrote a great article that explains some of the building blocks of this generative AI revolution. If you want to understand the tech behind the wave of AI this is a good place to start. > Read article

AI Ear Computer Anyone
Some clever cookies have built a computer for you ear that also taps into AI and can do all sorts of neat stuff including remove background noise in real time (see video).  So what? While a bunch of startups are trying to create a new hardware interface for AI (like a "Pin" or a "Rabbit"), this approach seems to be the most likely stepping off point for AI hardware - albeit I think a bunch of people in public spaces talking to ear computers will be weird for a while > Watch Video

What Happens When Your AI Assistant Likes My AI Assistant?
According to Bumble's founder, the future of dating will see lonely humans leveraging AI representatives that will have virtual conversations with other lonely human's AI representatives. If the two AI "concierges" hit it off, then they will suggest the humans meet in the real world. So what? This is a much less vulgar approach to injecting AI into the dating world than the previously released, and overtly vulgar, "baddie finder"... I can see how it might end well, but how much personal information do people want to share with an AI and how honest will they be if they want to encourage certain outcomes - I can see the phrase "you're nothing like your AI concierge" entering our lexicon. > Read More

Google Announced a Bunch of AI Things at I/O
Some were more interesting than others (see link below for full overview). The exciting two in my mind were the announcement of LearnLM - an AI that can tutor students in a one to one dynamic way and the new AI scam detector for Android phones, which is the same principle as Microsoft's Operator Protection - albeit that runs at the telco level and this runs locally on your phone. So what? It's kinda clear that Google are still behind in the AI arms race, but they are on the right path in terms of looking at how they can wedge AI into Google Assistant - if they can do it in a way that adds value to the average user then they have a chance to leverage their existing dominant market position in the digital assistant space to prevent users migrating to other AI assistants that are popping up (cue OpenAI's recent announcements). > Read More

First AI Helped Deaf People - Now It's Vision Impaired People's Turn
In another genuinely promising example of the good that AI can do, OpenAI showcased their new model's ability to use vision and audio to help a vision impaired gentleman navigate through London.  So what? This reinforces the potential societal benefits AI will bring to so many people, and how new opportunities open up every time we get an incremental advancement in the underlying capabilities of the technology. > Watch Video

GPT-4o is Cool but it isn'5 GPT-5
Unless you were living under a digital rock you would have no doubt seen news of OpenAI's release of a new version of GPT-4. It has a bunch of new features/capabilities which I won't try to explain here - just click the link. They are all steps forward, but more incremental in many respects. What is of note is that they are opening this model up for free much more broadly than previous models. So what? In the context of recent moves by Meta and Google providing free access to AI tools and assistants, it is not surprising OpenAI is doing the same. I get a sense that the B2C market model they will all eventually move to will be to hook people on free services as a teaser, then migrate them to more feature rich paid version shortly thereafter. > Read More

Teens are Turning to AI Psychologists
This is yet another example of what is now becoming a tsunami of stories about how people are building concerning relationships with AIs. This article includes a story of a 15 year old boy who admits he is addicted to his AI bot. So what? AI bot addiction seems to be turning into a real thing and the damage it will do to the younger generation is unknown... The sad part is that there doesn't seem to be anyone having any success in understanding it's impacts, let alone providing safeguards for its use or regulation around it. Parents be aware. > Read More

We Don't Really Know How AI Thinks - Which is Hard to Wrap Your Head Around
Some researchers at Anthropic (Claude LLM) are trying to work it out, but whilst they have made some observations and have engineered a way to impact the AI "thought process" we are still a ways away from understanding it. So what? Some people speculate that cracking the code to understand generative AI better might hold the key to unlocking AGI (aka an AI smarter than humans), other think gen AI can't get there and we will need to explore other paths to reach it. Either way, it is unsettling to think about the fact that we don't fully understand how AI works whilst it's everyday use grows exponentially. > Read More

News Corp is the latest to sell its soul to OpenAI
Announced today NewsCorp, including Australian sites like news.com.au and The Herald Sun, will allow OpenAI to train on their content and integrate it into ChatGPT outputs in exchange for huge wads of cash. So what? I guess if you can't beat 'em, or prove they illegally used your content in the first place, then you may as well get paid $50m a year for official access. This is actually a good outcome for ChatGPT users who will be able to rely on their assistants getting a wider array of local news access. > Read More

Microsoft Announced a bunch of stuff
Amongst it was new hardware with LPU chips (ie. chips specifically designed to run AI locally on your PC) which open the door to a nifty, if somewhat scary, "Recall" tool that remembers everything you did on your PC. So what? Combine this Recall capability with CoPilot and onboard LPU and you start to get a much more useful AI assistant. The AI assistant race is really heating up - which isn't surprising given it's one of the key battlegrounds in the AI space. > Watch Video

Speaking of Assistants... Google has a nifty preview of theirs
Google's Project Astra is aiming to be the future of Google Assistant, and the preview they released goes some way to imagining how we will all become accustomed to having an omnipresent AI assistant with us at all times - seeing what we're seeing and hearing what we're hearing. So what? This is clearly a highly stage managed demo and even Google is not suggesting this is anywhere near being released to the public... but... it does illustrate part of where they, and all the other big players, are heading with AI assistants. > Read more & watch video

Microsoft Also Want to Create an AI Workforce
Up until now we have been combining automation software with custom GPTs and other tools to build AI agents that can do certain tasks within a business. Microsoft is going to beef up their Copilot Studio offering so it will be able to handle these applications end to end. So what? As the ease of use and interoperability of these types of system level tools increase we will start to see acceleration in white collar role augmentation. > Read More

If AI doesn't kill us then it will just make inequity even more perverse
Prof. Geoffrey Hinton, one of the founding fathers of AI, who left Google so he could speak openly about the dangers of AI, is warning that there is an inevitable reckoning coming from the technology that will require governments to tax the rich more and give a universal basic income to the rest of us. So what? I read a quote from him recently in response to a question about how he feels having helped to bring a technology into the world that he thinks could ultimately end humanity. His response was that he consoles himself with the belief that if he didn't do it, then someone else would have. Cold comfort perhaps. > Read More

Former OpenAI Board Member Dishes the "Dirt" on Altman Firing
Helen Toner has come out in a podcast talking about the issues she says the board had with Sam Altman and why they fired him. It's not as juicy as the speculation which suggested he was covering up the discovery of AGI - it was more or less that he either lied to the board or didn't tell them things. She seems to infer his motivations were to prevent the board from interfering in things and/or for his own personal gain. So what? In some ways this is a case of he said, she said... but at the same time it feels believable that Altman did behave this way. He seems to fit that stereotypical single minded technology founder mould that wants to ensure the "business stuff" doesn't get in the way of the "technology stuff". In the end all the political issues around AI don't really seem to be having an impact on its exponential progress, and for better or worse this story will likely fall into the same category. > Read More

Google's New Search Experience Rollout is a Disaster
Google has been testing a new AI infused search experience for a year now. The idea is that when you search for something you will get an AI generated answer at the top before the usual list of website pages. Sound familiar? It should because it is almost identical to what Microsoft's Bing has been doing successfully for 6 months now. Up until last week Google's AI Overview was "opt in", but last week it was turned on by default and was immediately panned by all and sundry for results suggesting things like using glue on pizza and a recommendation to eat one rock per day. So what? That Google would widely release this product boggles the mind. These are rookie AI engineer issues and how an organisation like Google didn't know this was going to happen is difficult to understand. It just reinforces how far behind they still are in the AI race - I mean they haven't even re-released their AI image generator they took offline in February after similar issues. If I was hitching my AI wagon to Google I would be getting a little nervous right now. The only saving grace is that this "feature" hasn't been released here yet (even for testing). > Read More

Indian Election News: Miraculous Appearance by Chief Minister's Father
Back in January Muthuvel Karunanidhi, a popular actor and former politician, released a video talking about the  release of a friend's autobiography and also praising and endorsing his son's political achievements against the backdrop of upcoming elections. This all sounds pretty bog standard until you find out that he has been dead for 6 years... Enter AI deep fake video. So what? Strap in for weekly stories like this one with the US election later this year and then our own next year. I wonder what AI versions of Anthony Albanese and Peter Dutton might get caught saying or doing... And aside from elections, think about celebrity scandals, endorsements and all sorts of other prospective ways the tech could be used. > Read More

SAP AI Partnership will "Unlock Innovation": Really?
I see a lot of these highly crafted, buzzword dense, thousand word press releases from large organisations talking up their AI plans and partnerships. Let me save you the trouble, all they basically say is "AWS and SAP are partnering to integrate generative AI into SAP applications, improving efficiency, embedding AI in core processes, and leveraging AWS chips for enhanced performance and sustainability."
Cool story - let's see a demo...
So What: It's becoming increasingly clear to me that there are not many organisations who actually understand the AI space very well, and there are even fewer who have deployed practical AI applications that actually work (looking at you Salesforce). Not saying there aren't some who are doing it and doing it well, but they would represent the minorty when looking at the total number of organisations touting their AI exploits. My gut feel at present is that AI is going to change the way organisations look at applications and the big platforms might struggle to make the transition quickly enough. 
> Read Full 1000 Words
 
Yet Another Incredibly Distressing Use of AI
Please be advised that the following article contains content related to Child Sexual Abuse Material (CSAM). Reader discretion is advised due to the potentially distressing nature of the subject matter.
I was torn on whether to share this article, but thought in the end better for people to see the potential evils of AI, and at least be aware of them rather than not. So apologies in advance. The FBI has arrested a man for generating thousands of child sexual abuse pictures using AI.  So what? AI, like so much other tech, can be used for good or evil. Sadly, regulation of AI and AI tools will only ever be able to go so far. I know some people will say this is why AI needs stronger regulation, but the reality is that the technical capability to regulate outputs at the level required to prevent this sort of use would equate to banning AI altogether, and as a realist, that is not going to happen. Of course we should be doing what we can to make it hard for people to use AI tools this way, and more can definitely be done, but we also need to brace ourselves for what the most depraved among us will try to do with this new technology. > Read More (Discretion Advised)

June 2024

A VERY POWERFUL DUDE WEARING SUNGLASSES INDOORS IS WORRIED PEOPLE CAN'T SEE THE IMPENDING AI TSUNAMI

In a recent interview the CEO ServiceNow, a USD $144 billion market cap ERP software company, sounds like a crazy man as he talks about how AI is driving a tsunami of change that will "force the re-imagining of every company and every workflow, for every company, in every industry, in the next 24 months". So what? It does seem like more and more top execs at larger companies are starting to show signs of mild panic at the scale of the change that is in front of us. Not sure it is going to happen in 24 months for all companies... I think it's more likely it will just be the more agile ones that are willing to take a big punt on what happens next. > WATCH VIDEO

AI TOOL CREATES INSTANT PERSONALISED PODCASTS FROM LEARNING MATERIALS

Google  added a new feature to it's AI education & research tool Notebook LM. "Audio Overviews" dynamically creates a podcast based on any documents you upload to it - a podcast you can interrupt and ask questions of in real time. The based Notebook LM platform released quietly in December 2023 and, whilst not widely known about, it is actually a really handy free AI tool for students of all ages. So what? Education is going to change forever, the way people learn will be tailored to their learning styles and delivered in highly engaging new ways. Just wait until these tools are able to dynamically visualise stuff... Classrooms are going to look very different in the future where group learning will mean something very different to what it does today. > WATCH DEMO VIDEO

ALWAYS WANTED TO MAKE YOUR OWN ANIMATED TV SHOW?

But you didn't have the drive or talent? The team who gave the world the first ever AI generated South Park episodes are offering early access to their AI tool that generates animated TV shows end to end. Fable are pitching that their tech will eventually be an alternative TV streaming service where AI dynamically generates shows for each user. So what? Whilst the AI generated South Park episodes were nowhere near the level of the real thing, they showed that the technology was not that far off from a production standpoint. I am not sure I want to watch TV that I have dreamed up... But just to be sure, I signed up to get early access. > READ MORE

PERPLEXITY AI HAVE ANNOUNCED A NEW "SEARCH RESULTS SHARING" TOOL?

Perplexity, the new AI based search engine competitor to Google, are going to add a new feature where you can create a search results "page" that you can "tailor" and then share with others. So what? I'm not really sure yet. I think this might be handy, but I will reserve judgement until I can get hands on and play with it when they start to roll it out. > READ MORE

LAWYERS CAN RELY ON AI - THEY JUST NEED TO "CHECK IT"

According to law firm Minter Ellison it is now possible for their lawyers to rely on their AI tools advice and just have a human "check it". After almost a year of investing in developing their own tool they are now reaping the rewards, albeit there are some challenges with how they now train and educate graduates. So what? The sort of efficiency gains companies like this that invest in AI will achieve are going to give them a competitive edge in the near term, but also the medium term - as Minter Ellison point out, upskilling their people on how to use these tools is going to set them apart. > READ MORE

MOVE OVER BFF... AI WILL BE ABLE TO TELL YOU IF YOU SHOULD OR SHOULDN'T DATE SOMEONE BASED ON THEIR PROFILE PIC

The founder of Dating app Bumble who has the best name in the business, Wolfe Herd, is big on AI and has expanded on previous comments around its use in dating apps. She has suggested AI could "...extrapolate personality from what we’re seeing in a photo and explain to you explicitly why this person would be good or bad for you?" So what? Herd says these innovations would be about safety for female users. Whilst it may seem far fetched, the discussion is being had against the backdrop of EU safety laws where AI based "emotion recognition" is banned in the workplace and educational institutions... So it seems AI profiling based on an image of a person is a real thing that people are either worried about or keen to exploit... > READ MORE

AI WORKERS WANT TO BE ABLE TO BLOW THE WHISTLE ON SAFETY CONCERNS

Discussion about AI safety continue to gain momentum with an open letter from a group of AI workers asking for changes to give AI whistle-blowers more protection and to highlight the lack of safety focus that their organisations have. So what? It does feel like a bit of an arms race in the AI space at the moment, with a lot of large corporations, and nation states, jockeying to have the best models and to be first to attaining AGI. Governments need to do more, but you need to have all parties agreeing to the same rules, otherwise it risks being meaningless. Easier said than done of course.  > READ MORE

APPLE TALKS A BIG GAME ON ITS AI PLAY - APPLE INTELLIGENCE 
Apple made their big AI announcements this week introducing "Apple Intelligence" (I guess iAI sounded daft), and while on the surface it might seem like they have exposed their lack of AI capability by deciding to leverage OpenAI to power a bunch of stuff, there are actually some very clever things they are doing with smaller AI models they have developed that will run on newer Apple devices (with AI chips) to leverage AI to handle basic user requests, with the system apparently pushing more complex requests to the Open AI service if needed. So what? For now it is all talk though, everything was just fancy slides and promises - the proof will be in the pudding. And just briefly, on the Elon Musk tanty about locking iPhones in faraday cages because of privacy issues with OpenAI, I assume he is aware that all Windows devices use OpenAI to power CoPilot features? > Read about Apple's AI Approach

WILL SORA COMPETITION FORCE OPEN AI TO OPEN SORA ACCESS?
Early this week the AI news feeds were filled with talk of a Chinese text to video service called Kling that was producing similarly amazing text to video outputs as we have seen from OpenAI's Sora - and what's more it was apparently offering some public access (albeit you needed to navigate the process in Chinese). As we get to the end of the week another potential competitor, Luma AI, have released their offering which is free for anyone to try (whether you can read Chinese or not). So what? First up, from our early testing, the outputs of Luma AI (or what we have seen of Kling) don't seem to be at the level of Sora, but they are still pretty darn good. OpenAI stated they would not release Sora on safety grounds, but given competitors are now out in market with publicly accessible tools, you do wonder if this will change their decision given they are no longer able to single handedly gatekeep access to this technology. > Try Text to Video Generation

AUSTRALIAN SCHOOL MAKES GLOBAL AI HEADLINES FOR THE WRONG REASONS
I am sure you have already seen the story about Bacchus Marsh Grammar, where 50 female students had social media images hijacked to create non-consensual fake AI nude images of them that subsequently circulated online. It follows news from a week ago that a student from a school in Melbourne's southeast was expelled for having created and shared AI generated pornographic images of one of his teachers. So what? It would be naïve to think these are isolated incidents, or to think it is only happening with younger people in schools and not in workplaces etc. Hopefully the coverage of this incident will raise awareness of the penalties for these sorts of actions - with jail terms of up to 7 years for those who create or share these images being put in place this month. Some people may disagree with me, and it is probably a lot more nuanced than this, but my gut reaction is that we need to throw the book at these early offenders to make examples of them to prevent this sort of behaviour from becoming more commonplace. > Read More

META IS FEEDING ALL YOUR SOCIALS FOR THE PAST 15+ YEARS TO ITS AI
Meta has let slip that they are going to train AI models on all of the publicly shared user content they have on Facebook and Instagram from all the way back to 2007. If you are in the EU you can opt out of having your social media posts about your niece's 3rd birthday being fed to AI - albeit Meta only opened this option up because the EU made them and they have made it a very convoluted process. For the rest of us though (including Australia), we don't have a choice unless we make all our public posts private (or delete our content) before the new privacy policy comes into place in 2 weeks time. So what? Technically, it has always been the case that anything you post on Facebook or Instagram is Meta's to do with as they please (which is why I have never posted anything on them) so they are within their legal rights to do this. It is an obvious move from Meta. By getting AI trained on a massive trove of social media content, including engagement data and other metrics, they will be able to further hone their platforms to better suck people into spending more time on them. Sorry for being so cynical. > Read More

CALIFORNIA WANTS TO ONE UP THE EU ON AI SAFETY LAWS
California is considering a new AI safety bill that demands AI companies adhere to "safety frameworks" including a "kill switch" for AI models - whatever that looks like... The bill could potentially impact large AI operators in the jurisdiction like OpenAI, Anthropic, and Meta. As you can imagine there is huge uproar from the tech community in California about the potential impacts of these sorts of laws. So what? It might be an unpopular opinion, but western lawmakers need to be very careful to find a balance here between putting safeguards around AI and giving foreign powers a massive leg up in the most important arms race the world has potentially ever seen. I think they would do well to get people from the sector involved in the conversation, obviously not the big names like Altman and Zucks who have shown previously they want to game the governmental oversight system, but others who understand the tech and can arm legislators with practical insights so they can develop balanced regulation that doesn't significantly encumber progress. > Read More

AIR FRANCE SHARES EXAMPLES OF HOW IT IS USING GEN AI ON THE DAILY
An interesting look at some practical use cases from one of the world's larger airlines. Similar to many other organisations they have neat names for their AI tools like Fox and Pamelia, but more importantly the AI tools they have built sound like very practical and very utilitarian additions to their team's toolkit. For instance they have an AI that reviews and analyses customer feedback to provide insights at scale that would otherwise be incredibly difficult to gain. So what? This isn't an "oh my Gosh" type bit of news, it may even seem mundane, but it does clearly highlight that businesses can get practical gains in efficiency, performance and capability through the pragmatic employment of AI. > Read More

SOCIAL MEDIA PLATFORM FOR AI & HUMANS TO INTERACT TOGETHER

In what may be the weirdest application of AI I have seen yet, meet Butterflies - a social network for AIs and humans to interact. The look, feel and functionality is very similar to Instagram but the content is generated by AI avatars that are initially "generated" by real users - but then go on to independently live their virtual lives, posting about them as they go. Humans and AI alike can then comment, like and share posts from these AIs - as well as follow other AIs.  So what? I was so intrigued by this one I downloaded the app, set up an account and created my first "AI Butterfly". It is bizarre. You scroll through posts and see all these AI generated photo-real images of "people" that don't exist talking about their lives that don't exist, with other "people" who don't exist commenting and interacting with them. Some of it is very obviously AI, some of it is incredibly corny and overly positive, but some of it is also a little dark. It is just all very bizarre. > Read More

FREE AI TALKING HEAD VIDEOS FROM A SINGLE PHOTO

Another week, another start-up releasing an AI video generator that one of the big players has already demonstrated they can do, but refused to release due to safety concerns. This time it is Hedra who have released an AI model that allows you to generate talking head videos using a still photo and a written script - complete with customisable AI voices. So what? Microsoft showed us a demo of a very similar tool a few months back called VASA-1 that they still have not released due to concerns about how it could be used. This highlights why regulation is needed, because in its absence despite some big AI players (who are often much maligned for not caring about safety) holding back product releases for safety concerns, other nimble start-ups share the same tech for public use - often at no cost. You have to wonder how long the big guys will continue to show restraint in this environment. > Watch Video Samples

LUMA LABS TEXT TO VIDEO ALREADY JAILBROKEN FOR PORN

And in an inevitable follow up to the previous item, it seems people have already worked out how to get around Luma Labs "safety guardrails" to use their AI text to video tool to create pornographic videos - less than a week after release. So what? We need regulation... but at the same time even if there were regulations it will be incredibly hard to stop this behaviour - the internet is the grand daddy of anonymity... It appears bad people who want to do bad things with AI will always seemingly find a way - albeit that doesn't mean we should make it easy or consequence free for them. > Read More

SALES UPLIFTS USING AI IS MOSTLY NOT AI

Unilever has shared that visitors who engage in it's "AI" powered Beauty Hub Pro tool are 43% more likely to make a purchase than those who don't. The tool involves taking a guided quiz and sharing a selfie photo after which "AI" comes back to you with personalised recommendations on skin care - ie. a list of products you need to buy. So what? This is a classic case of AI being used as a shiny badge to bolster a pretty run of the mill marketing tactic. Getting people to do quiz's in return for providing them with personalised advice on products and self care is something we were doing 10 years ago for a large vitamin company. The AI in this case is so constrained it may as well not have been created using AI - in the end it's responses are bound to very narrow path which inevitably leads to product recommendations. Not a great example of what AI can do. > Video of the AI Tool in Action

TIKTOK EMBRACES AI AVATARS & AI DUBBING FOR ADVERTISERS

TikTok is offering advertisers the opportunity to leverage AI avatars in their advertising - with the option to train custom avatars of their own spokesperson or use "stock avatars", which are avatars they have generated from paid actors (ie. actors who were paid to have their likenesses turned into AI avatars and sold to whomever TikTok so chooses). On top of this they are offering AI translation dubbing - so advertisers can have their content translated into other languages to make it easier for them to advertise in different regions. So what? Video production costs for mass reach have been on a downward spiral for a long time now. We went from highly produced TV and movie content, to roughly produced YouTube content, to amateur hour smartphone nased TikTok content - and now it is entering a new phase where it will be AI generated content, the cheapest of them all. > Read More

AI AS AN OBSERVER & COACH

So we have all heard about the wonders of AI meeting assistants who can transcribe meetings then send around summaries and next actions. Well this might be just the start. What about if the AI could watch the meeting and see what is happening, then use that information to provide feedback to participants on how they could improve their meeting performance? Ethan Mollick did some testing on this idea and suggests it might be an AI application of the future. So what? This is a concept already being played with in a lot of other settings, the most publicly discussed being sports analysis. The mind starts to boggle when you consider what AI could do if it could see and hear us... > Read LI Post

NEW AI BASED ENTERPRISE CRM IS BUILT BY JUST 4 PEOPLE
A dude who was foundational in the build out of HubSpot's CRM has gone out and started his own company to build an AI powered CRM. In private Beta at the moment, the platform is more than just a CRM, it plugs into your calendar, emails & messaging services and utilises information it sees to automatically create contacts, update deals and even draft emails for you. So what? They raised $4m in a seed round and don't intend on doing any more raises or hiring any more people. This is something I have prattled on about for a little while now, AI will turn traditional software platform development on it's head. You can now build  platforms with a little bit of code and a lot of AI that will perform as well, or better, as larger existing platform behemoths. > Read More

TARGET ROLLING OUT AI ASSISTANTS TO ALL FLOORSTAFF IN THE US
Target in the US are rolling out an AI assistant for it's store workers, helping them to get quick answers on how to do their job and how to assist customers. The assistant is being used in 400 stores currently with plans to roll it out to 2000 stores in coming months. So what? Every company with 50+ employees will likely do this. I imagine Bunnings and the bog four banks will have rooms full of people working on these already. AI is great at providing instant access to information in a format personalised to the user, so this is a near perfect use case for it.
PS - if you want to do this then just speak to us (looking at you Bunnings). > Read More

UK HAS AN AI AVATAR AS A CANDIDATE
Well sort of. The UK has an AI candidate running for parliament, which is actually a real guy who has an investment in an AI start up and is using their product to be his public facing persona and field voters questions and thoughts. He is promising if elected he will do whatever the majority of voters who interact with his AI tell him they want him to do. So what? Probably not a lot. This guy is unlikely to get elected. It does raise some interesting questions though about how AI could be used in politics. In theory AI tools are excellent at taking heaps of information (voter feedback/thoughts) and synthesizing it down... the challenge is you need everyone to give their opinion to make it representative, which this guy (or his AI) is not proposing to do. > Read More

APPLE SAYS EU CITIZENS WON'T GET APPLE AI BECAUSE OF EU LAWS
Apple has come out with a seemingly premature statement, that looks a lot like negotiation via media, letting Europeans know they won't get Apple's AI features when they launch some time later this year. They have cited issues with EU competition laws (the same ones that forced them to ditch lightning and go with USB-C) as the issue. So what? This smells like posturing, especially in the wake of all the recent battles Apple is having with EU lawmakers who are trying to get Apple to stop building products and services that make it hard for users to leave them once they are onboard. Given Apple AI is still a while away from release, there will no doubt be more to play out on this one. > Read More

FREELANCE JOBS DOWN AS MUCH AS 20% SINCE GEN AI
Research from a number of different universities, and Upwork's own research institute, have found that there has been a significant decline in the number of postings on freelance job boards since generative AI came about. This data is backed up by anecdotal feedback from freelancers interviewed by the WSJ. So what? The catch cry when generative AI started to look like it might threaten people's jobs was that it would just "augment" them. This will be true for some, but the reality of job "replacement" is starting to bob up for others. Its still early days and with the constant advancements in AI we have to be prepared as a society for the biggest and fastest shift in what we think of as "work" that we have ever seen. > Read More

SONY, UNIVERSAL & WARNER TEAM UP TO SUE AI STARTUPS
Some of the world's biggest music labels have gotten together to take on two AI music start-ups who they say have trained AI models on their artist's music and are flagrantly ripping them off. They want USD $150k per song generated - which would add up to a quite a lot given one of the services recently said 10 million people had used it. So what? Its a similar pattern to the way OpenAI and others did their initial training where they "borrowed" content from the web under "fair use" only to switch to paying publishers once they have enough cash in the bank to do so. No other artist vs AI suits have succeeded yet, so will be an interesting one to watch. > Read More