The SEC Newgate AI Weekly
Welcome to this weeks’ AI briefing, bringing you the most interesting developments in AI over the past seven days and what they could mean for you and your business.
OPEN(AI) FOR BUSINESS
The biggest news this week came from the much-anticipated OpenAI developer conference. With CEO Sam Altman hyping the event with his “real christmas eve energy rn” post on X (formerly Twitter), all eyes were on San Francisco on Tuesday and the company did not disappoint. Highlights included GPT4 Turbo, a new model trained on data up to April 2023 which promises to allow prompts the length of a book (allowing analysis on much larger blocks of text than previously possible) and significant reductions in cost for developers using GPT4.
But the standout announcement was the ability for premium users to create custom versions of Chat-GPT for specific purposes and (where applicable) to sell access to these via an online store. The use cases are endless but think back to the early days of the iPhone store, with developers finding new and novel ways to exploit the device’s capabilities – this time with no coding knowledge required whatsoever. The new offer also includes the ability to create internal-only chatbots for employees – imagine having an AI assistant with knowledge of your organisation’s policies, procedures, messaging etc.
Finally, use of ChatGPT continues to accelerate with the company announcing that it now has 100 million weekly users.
AMAZON’S OLYMPIC EFFORT TO COMPETE
Amazon is training a huge 2 trillion parameter large language model (LLM) according to undisclosed company sources who leaked details to Bloomberg this week. The model, codenamed Olympus, may be the tech giant’s attempt to ‘catch up’ with competitors such as OpenAI, after reports earlier this year that Amazon’s initial attempt at a ChatGPT alternative were not good enough to release to market. Whether the leak was sanctioned or not, it does the company no harm to be driving up interest for their new offering in a very crowded space.
Elon Musk never misses an opportunity to steal the limelight and this week was no exception. The owner of X announced his own chatbot, called Grok (a word meaning ‘understand’ in tech circles) ahead of rival OpenAI’s developer conference. The model that powers it appears to be nothing special in technological terms (more capable than GPT3.5 but not as powerful as GPT4), the main selling point is its access to up-to-date information from the X platform. A key question which remains unanswered is why Musk thinks this is more useful than the now-standard web browser functionality of competitors such as Google Bard and BingChat. The chatbot promises “to answer questions with a bit of wit” and X claims it has “a rebellious streak” and will answer “the spicy questions that are rejected by most other AI systems”. This provokes more questions than answers, as attempts to limit the range of answers given by other chatbots were triggered by concerns such as racism, sexism and the disclosure of unsuitable information (the recipe for napalm, bomb-making instructions etc.) Expect a host of stories in the not-too-distant future as X comes under pressure to rein in its chatbot in line with acceptable practice elsewhere. But on a more positive note, credit where it is due – X managed to train their model in only two months, a significant achievement suggesting their AI capabilities are more advanced than they appeared (or they’ve rushed it out before it is ready – we’ll know soon enough).
AI podcaster and former Microsoft Strategist Robert Scoble took to X this week to bemoan the lack of GPS capability in ChatGPT which prevented the chatbot from helping him find a nearby charging point during a stopover at Heathrow Airport. His post triggered a conversation about the current limitations of the technology and speculation that this is exactly the sort of thing Apple might include when their long-awaited generative AI offer finally arrives. Recent developments with Google Maps (covered in last week’s AI Weekly) may also go some way to resolving this but either way, location-specific generative AI can’t be far off.
The UK AI Policy Resource Hub
The UK’s AI Startup Roadmap (Startup Coalition, Onward, TBI)
Generative AI Policy Paper (Demos think-tank)
Open Sourcing the AI Revolution (Demos think-tank)