Summary this week
- Learn: A a piece of code called Wolverine, named after the Marvel comic book hero with super healing abilities, was released here on GitHub. The demo takes code written in Python and then passes the result to GPT-4 (ChatGPT's successor) to be debugged, the program then takes GPT-4s bug fixes and re-runs the code, debugging or 'healing' the code until it is bug free.
- News: An open letter from an organisation called The Future of Life Institute has released an open letter calling on companies training models of greater than GPT-4 intelligence, to be paused for six months while the world discusses how to add safeguards.
- News: Italy has banned ChatGPT for GDPR violations. The regulator has said that the ban and investigation of OpenAI will start "with immediate effect".
Wolverine is the project created by Twitter user @BioBootloader. It is a working proof-of-concept that takes a script written in Python (but could theoretically be modified for any language), and tries to run the program. If there are errors in the code it passes them to GPT-4 and asks it how to fix them. The suggested code is then returned, and automatically written back into the script, then the process starts again. This creates a loop, 'healing' the bugs in the code with each successive pass, until the code can run without errors.
Watch video below (2 mins) to get a sense of what's possible. Even if you don't code yourself, the concept is simple enough to follow along.
While it's not much more than a demo at the moment, it does show that writing code quickly (and even badly) can be a enough of a prompt to get GPT-4 to create a working version of your script. Since GPT-4 can now take thousands of words of text into it, most code files will now fit inside that range. It's not hard to see that programmers will shortly have an assistant that can automatically fix code and fix errors, once you have enough in place.
This is a clear difference in capability, right now most AIs ask permission to make any changes. This program starts an AI acting autonomously. This is clearly one of the next steps watch for.
The open letter from The Future of Life Institute has caused quite a stir in the AI community. The letter titled “Pause Giant AI Experiments” calls for a six-month pause on AI experiments that “should be developed only once we are confident that their effects will be positive and their risks will be manageable.” Signed by over 1,100 people, the letter has the support of notable figures such as Elon Musk, Steve Wozniak, Yuval Noah Harari and Andrew Yang.
This was followed up by a much more aggressive piece in Time by Eliezer Yudkowsky titled "Pausing AI Developments Isn't Enough. We Need to Shut it All Down". You can hear him articulate (very well I think as he's clearly very knowledgable) his position in the podcast below.
It’s clear that we’re seeing the beginning of an AI backlash. The technology is being hyped far beyond what it’s currently capable of, and people are leaping ahead to the inevitable doomsday scenarios. While these scenarios are absolutely possible in the future, we’re a long way from having a basic Artificial General Intelligence (AGI), let alone one that has the capability and intent to cause harm unchecked.
It’s obviously important to have discussions about AI and regulation, but these presented solutions are extreme and completely impractical. For example, do we think that bad actors will pause or stop development, because some countries pass laws? This is an arms and economics race now whether we like it or not and we should present solutions that work towards ensuring AI's are aligned, rather than give adversarial groups a chance to move ahead.
As an example it's clear that social networks have had both of good and bad outcomes since their inception, but pausing them at their earliest days would not have helped us avoid the problem as we didn't yet understand how they would affect society. I believe the intense interest in this space will provide enough oversight for the moment without ceding the lead to bad actors.
Italy bans ChatGPT
As of the morning of April 1st 2023, the ban has come into effect. Federico Vittici of Macstories.net reported on the email all ChatGPT users recieved following notification of the ban. The regulators have objected to the collection of data from individuals which is then used to train the system, in addition that data is also collected from children under 13 and there is no mechanism in place to prevent them signing up.
Text of Email from Open AI
Dear ChatGPT customer,
We regret to inform you that we have disabled ChatGPT for users in Italy at the request of the Italian Garante.
We are issuing refunds to all users in Italy who purchased a ChatGPT Plus subscription in March. We are also temporarily pausing subscription renewals in Italy so that users won't be charged while ChatGPT is suspended.
We are committed to protecting people's privacy and we believe we offer ChatGPT in compliance with GDPR and other privacy laws. We will engage with the Garante with the goal of restoring your access as soon as possible.
Many of you have told us that you find ChatGPT helpful for everyday tasks, and we look forward to making it available again soon.
If you have any questions or concerns regarding ChatGPT or the refund process, we have prepared a list of Frequently Asked Questions to address them.
-The OpenAl Support Team
This action itself should be fairly simple to remedy, but it is worth noting that the Italian regulators have already targetted and banned the AI avatar company Replika in February earlier this year for the same reasons. As a result we should expect other regulators across europe to be considering similar action. As a final point it should be noted that using a VPN still gives access to ChatGPT in Italy, but it is unclear whether use of an Italian card for purchases will be blocked as well.
Please share and subscribe
If you liked this free weekly newsletter please use the subscribe button and share or forward if you think others would benefit from this content. If you have any questions, or there is a topic or AI product you would like me to cover, please let me know at [Not active yet: firstname.lastname@example.org] or use the social icons at the bottom [Not active yet] to send a message.