A new research paper from Stanford University purports that the models running ChatGPT have definitely changed over the last few months. It seems like GPT-4, the newest, more advanced model, has gotten less accurate and reliable, while GPT-3.5, the intermediary model released right before it, has improved regarding specific tasks. It seems the product is morphing, and I’ll note an analysis I dug into from AI Snake Oil cites less a degradation and instead a shift, and that this is, in fact, the nature of changing AI models. Their statement – this is the challenge of building products on changing models.
Microsoft Inspire was this week, and there’s lots of news about their AI efforts.
CEO Satya Nadella called next-generation AI a partner ecosystem opportunity that could span $4 trillion to $6.5 trillion.
Microsoft 365 Copilot, the company’s generative AI-fueled productivity tool, will be priced at $30 per user per month. No release date yet. Microsoft also announced its Bing Chat Enterprise service, an AI-powered organizational chat with security features, which rolled out under preview mode Tuesday. The new solution will be included for free in certain business- and enterprise-grade Microsoft 365 bundles. The differentiator – higher levels of data protection and privacy. Chat data isn’t saved and is not used to train new models. CoPilot is coming to Teams phone and chat, adding real-time call summarization and pulling out key points from chat threads.
Microsoft is also adding Meta’s family of Llama 2 open-source foundation models to Azure, available now. This announcement, concurrent with Meta announcing the model itself, is commercially available. The latest version of Meta’s A.I. was created with 40 percent more data than the company released just a few months ago and is believed to be considerably more powerful. And Meta is providing a detailed road map that shows how developers can work with the vast amount of data it has collected. Last week, Meta announced CM3Leon (“Chameleon”), a new text-to-image generation model.
Researchers at SlashNext revealed the existence of WormGPT, a tool promoted on hacker forums, which aims to be a blackhat alternative to ChatGPT. “one that lets you do all sorts of illegal stuff and easily sell it online in the future.”
And Apple has created its own Apple GPT internally… but is still figuring out what to do with it. That is from reporting in Bloomberg.
Amidst all of this… A majority of executives in a survey by Hitachi Vantara, 76%, feel their current infrastructures “will be unable to scale to meet upcoming demands” — aka, AI and associated analytic workloads. 60% of those execs are “overwhelmed” by the amount of data they manage.
Why do we care?
Microsoft is leaning into AI. Obvious and clearly on display with this set of announcements. Does it warrant a doubling of the cost of Microsoft 365? I’m not sure, and I would love to hear from listeners on this one.
I stand by my earlier assessments that the value of providers here is in product and model analysis. I’ll add that the dimension of models changing this frequently was not in my initial thinking. There’s more reason to consider this an opportunity due to that increased complexity. The models get paired with applications to drive results – even the criminals have noticed that. Apple too.