CALIFORNIA SET TO PASS AI SAFETY BILL - AI PEEPS UNHAPPY
The Safe and Secure Innovation for Frontier Artificial Intelligence Models Act (SB-1047) has been passed by the Californian State Assembly and now just needs the governor's signature to be passed into law. The bill will require AI companies that operate in California, or offer AI products citizens in California, to implement safeguards, such as enabling quick shutdowns, protecting models from unsafe changes, and testing for risks of significant harm.
SO WHAT: California is going it alone again (see CCPA) so of course OpenAI, Anthropic and others are all unhappy. Two ex OpenAI employees thinnk its good that the government is moving as, in their opinion, "Developing frontier AI models without adequate safety precautions poses foreseeable risks of catastrophic harm to the public" - something which they think OpenAI is actively doing.
> Read More

AUSTRALIAN GOVERNMENT LEGISLATES TO COMBAT AI DEEPFAKES
A few months back we had the disturbing case of a male student at Bacchus Marsh college creating deepfake nudes of 50 female students using an AI tool. The federal government has now legislated against the sharing of non-consensual deepfake sexually explicit material - with penalties of up to 6 years in prison.
SO WHAT: This is a positive step and will be the first in what will have to be a raft of new laws to tackle the potential negative use cases for AI, let's hope the Australian government can be more proactive with legislating in this space so the laws are in place to deter bad actors before any harms occur.
> Read More

AI TO MARK SCHOOLWORK IN THE UK
The UK government is putting funding into exploring the use of AI to mark students work which would remove a mundane and repetitive task from teachers' plates to free them up to invest more time in more impactful activities.
SO WHAT: This is a perfect use case for AI - augmenting human work by taking away the boring stuff no one likes doing. Hats off to the UK government for being willing to invest in this project, I imagine the ROI on this will be pretty significant if they can get it right - albeit the outcomes will be measured in improved experiences for students and academic outcomes, rather than raw dollars and cents.
> Read More

LEAKED AUDIO OF AMAZON CEO REVEALS DEVELOPERS ROLES WILL CHANGE SIGNIFICANTLY OVER COMING YEARS
"If you go forward 24 months from now, or some amount of time — I can't exactly predict where it is — it's possible that most developers are not coding," said Matt Garman (Amazon Web Services CEO) in a leaked audio recording from an internal discussion. He went on to state that developers will no longer need to write code, but instead just focus on what needs to be built and how to best architect that.
SO WHAT: Every month we are seeing improvements in coding co-pilot tools. I think that in the next 12 months most developers will start to use AI coding assistants. Once that happens, you will see a gradual shift from where most code is written by the human with AI assistants reviewing it and doing a little bit of the coding... to where most of the lines of code are written by the AI assistant with the human reviewing and doing a little bit.
> Read More

ANTHROPIC SHARES BASE PROMPTS GIVING A WINDOW INTO HOW LLMS ARE TOLD TO BEHAVE
Anthropic, who built Claude (which is a very capable AI assistant that competes with ChatGPT), have shared the base prompts they give Claude to get it to behave as an assistant. There are a bunch of intersting insights from this information showing how it is prompted to engage "thoughtfully, showing curiosity and intelligence", and also to provide clear, objective information, while avoiding assertive openings like "certainly" or "absolutely."
SO WHAT: For most people though this is probably boring and uninteresting... but for AI nerds like me this is really interesting stuff and so I couldn't help but mention it.
> Read More

CBA USES AI TO DETECT SCAMS & ASKS TELSTRA IF THE CUSTOMER IS ON THE PHONE
The Commonwealth Bank and Telstra have expanded their "Scam Indicator" tool to cover landlines and mobile. The CBA uses AI to monitor customers activity and detect suspicious behaviours. If it detects something it thinks is suss, it will then contact Telstra (if the customer uses their service) and find out if the customer is on the phone (as well as perhaps some other data points) - if they are, then the CBA will take steps to block the activity and alert the customer.
SO WHAT: I am a customer of both organisations and had no idea Telstra could tell CBA if I was on the phone or not. Now I'm sure I signed something somewhere that gives permission for this, and I don't have an issue with them putting these sorts of processes in place to prevent me giving all my money to that very charismatic Nigerian prince who keeps calling me... but I do wonder how many people realise this sort of thing is happening in the background.
> Read More

AI SHORTS

More rumours abound about OpenAI's Project Strawberry (a version of ChatGPT that is supposedly going to be better at maths and reasoning) with some suggestion we are only a month or two away from it's release - more
Getty images have entered the AI image generation space, with a model you have to pay exorbitant fees to even test, that has pretty cruddy looking examples compared to Flux, MJ and Ideogram... but I guess at least it is paying the creators whose data it trained the model on - more
It seems like every week an AI image generator releases a new version that puts it at the top of the pile... this week it is Ideogram and the outputs are pretty neat - more
Salesforce is working on AI sales agents that can cover off initial online BDM outreach (read spam emails) and coach sales teams through role playing scenarios with AI avatars, they are slating release for October this year (ours will be ready in September - no jokes) - more
The new Pixel 9 phone with AI chip on board can do on device photo edits which can be pretty cool, or you could breach the terms of use, bypass safety guardrails and use them to do the wrong thing - more
Elderly Aussies are being ripped off by people apparently using AI deepfake video tech to disguise themselves - more