AI news - Fri 21st July
Updated: Aug 10
The highlights
Meta has launched their own Large Language Model (LLM) to compete with ChatGPT and Bard - and it is open source.
Rumours are floating around that Google is testing AI to write the news.
Workers rights against AI are being talked about by trade unions and actors.
Meta’s Llama 2
Large Language Models
The majority of us have heard of (or more likely used) OpenAI’s ChatGPT. It is currently the most widely used Large Language Model. ChatGPT 3.5 (the older version) is available for free, while the most recent version 4 can be used via a paid version (currently 20 dollars a month). Google has their own version called Bard which they call “an AI experiment” and is available for free.
This week Meta (formerly Facebook) have released their contender in this LLM AI race. Their model is called Llama 2, and it is not only free to use but also open source.
Open Source
Being open source means a bit more than access to the software being free (like ChatGPT or Bard), it means the code itself is available for people to use. This is a big deal when it comes to this type of software, meaning in terms of companies being able to use it securely. At the minute, companies with sensitive data (eg fintechs, law firms, etc) cannot use LLMs as they do not control where the information typed in is stored. But with open source, companies can take their own copy of this powerful algorithm and control how it is used on their data.
Open source also usually means a community of interested people (mainly developers) help to enhance, fix and build the software - for free!
In May, a letter was leaked from a Google engineer claiming the open source AI community was a serious threat to the front runners in this AI race. The engineer stated, “Open-source models are faster, more customisable, more private, and pound-for-pound more capable.” (Source: Guardian) Therefore, it is interesting (but not unexpected) for Meta to have chosen this path.
Two reasons spring to mind for this choice, first their models are slightly lagging compared to ChatGPT, and open sourcing opens new opportunities (and potentially forces the others to follow suit). Or the second reason could be Meta returning to its original route - if you’ve ever seen the Social Network, Zuckerberg was extremely against commercialising facebook (i.e. he did not want to add advertising on the platform in the beginning - but it's worth noting this was in an attempt to attract users and keep them engaged. And ultimately allowed Facebook to advertise to their millions of users eventually). Maybe this is the first step in Meta having a redemption (similar to Microsoft in recent years). Unlikely, but let’s keep our fingers crossed.
Further Reading
Easy to understand Guardian article
Meta’s research paper
Access to the Lllama 2 model on Hugging Face
Google’s news writer
AI Journalist
Originally reported in the New York Times (behind a paywall), Google has pitched some new AI which can write news stories. This project called Genesis is being called “responsible technology” by Google, and despite the dystopian vibes given (both by the biblical name and the thought that AI should be controlling news stories), it is not as scary as it seems.
The project is actually a helper for journalists - it can automate some boring, repetitive tasks and allow journalists to focus on the creative parts.
While these sound like great intentions, it will be interesting to see how this tool plays out - will it be used as an assistant or will some journalists start to use it to create entire news stories?
Articles Written by AI
In January 2023, the news site CNET was using AI to completely write articles which were riddled with inaccuracies (eg 3% of 10k is 10,300) and has received backlash from the public and other writers (Source: Washington Post)
This week has seen a wave of articles along the lines of “The best holiday destinations according to AI” in the Telegraph, BBC and travel websites. With one article comparing responses from ChatGPT and Bard (source: Travel Daily News). While using AI for the content, these articles are still written by humans (as we think it should be).
Bias
We wouldn’t be the responsible AI technologists we claim to be without calling out the potential issues with such technology. It feels extremely unlikely that any AI algorithm will be unbiased. And yes, humans are not unbiased, and a lot of news reporting comes with biases built in. But, it is deeply known that a lot of modern technology is amplifying the biases which exist in society. Therefore, we remain very sceptical about any such algorithms, and worried about the potential implications.
Further Reading
TechCrunch article
Workers vs AI
TUC is Worried
Trade unions have raised concerns that the UK (having left the EU) is falling behind in protection for workers as AI evolves (Source: BBC). They particularly mentioned the UK lacking similar protections to the EU's AI Act - which protects workers from AI decisions around hiring, firing and working conditions.
The article also has quotes from Octopus energy boss claiming their customers prefer messages from AI than humans. Further, AI is doing the work of 250 humans. Based on this, the TUC might be right to be worried.
Writers and Actors Strike
Leaving the UK, and focusing on Hollywood for a second, one of the (secondary after fair pay) demands of the current writers and actors strike is protection against AI. Writers want to ensure scripts won’t become AI generated, and actors want to ensure their images will not be used without them. Examples of this can be seen in the Abba holograms, but what is stopping living humans images being used instead of paying the actors themselves?
If you have watched the newest series of Black Mirror, you will know the first episode does a great job dissecting some of the potential issues with both AI generated scripts and actors.
Further Reading
The writers strike, including the AI part, is well covered in this episode of the Reasons to be Cheerful Podcast (so not strictly reading...)
We will be back next week with our top AI news stories of the week. If you happen to find anything interested - feel free to share with us!