A week in Generative AI: Claude, Oracle & Clicks
News for the week ending 14th September 2025
This week we have new Claude capabilities that allow it to create and edit files, but is really about helping with data analysis. We have a 45% swing in Oracleās stock price on the back of the announcement of a deal with OpenAI (which shouldnāt have been a surprise!), and some interesting research was released about consumers attitudes to advertising on AI platforms.
There are some nice little Web 4.0 articles that cover Google admitting that the open web is in ārapid declineā and news that one-third of all Internet traffic is now bots.
Lots of Ethics news too with a judge putting Anthropic $1.5bn book piracy settlement that we covered last week on hold and the FTC launching an inquiry into chatbot companions.
Thereās also a great Long Read from Ethan Mollick on Working with Wizards.
Claude can create and edit files
This is quite a confusing release from Anthropic as Claude already had an analysis tool that they released back in October 2024. The difference here is that it can now do much more than just analysis and is able to create and edit files in lots of different formats - spreadsheets, documents, slides, PDFs etc.
We havenāt previously seen an AI model thatās really capable of using Excel and I have to say that Claude is now really capable! Itās slightly surprising that we havenāt seen this from OpenAI/Microsoft first, but given Anthropicās enterprise focus it makes sense that theyāve been focused on this.
Simon Willison has run this new feature through its paces, and his review is well worth a read. These new features from Anthropic are very capable and I think will be genuinely useful for anyone who works with data, does analysis, and pulls together reports on a day-to-day basis.
As an aside, Claude AI can now also automatically ārememberā past chats.
Oracle & OpenAI Sign $300 Billion Cloud Deal
So, putting to one side that a $300bn contract is obviously a big deal on any day, this is the second large financial market shock weāve seen this year because of AI.
The first was back at the end of January when NVIDIA lost $589bn in value in a single day, the biggest market loss in history. This was off the back of DeepSeek releasing their open-weights model that rivalled OpenAIās o1 at reasoning. Itās worth noting that it took NVIDIA until the end of June, so 5 months, to recover its stock price but it has since gone on to grow by an additional 20%, becoming the first company to hit a $4tr valuation on its way.
Wednesdayās announcement of the OpenAI deal caused Oracleās shares to surge by 45%, adding over $400bn to their valuation. It also meant Larry Ellison, their Chairman, (briefly) overtook Elon Musk as the worldās richest man.
What occurs to me about both of these stock market shocks, isnāt the size of the sums involved but that they happened at all in the first place. The NVIDIA shock happened 7 days after the release of DeepSeekās R1 model, but the release was widely anticipated in the AI community as it was based on their V3 model that was released the month before.
The same thing has happened here with Oracle. The Trump administration announced Project Stargate back in March with both OpenAI and Oracle as key partners, and then when more details about the deal are announced we see a 45% stock price swing.
I think the issue here is two fold - firstly, weāre dealing with such large sums of money around AI at the moment that the stock markets donāt know what to do with it. The second, however, I think is more important. Investors are reacting not to the technology itself but to when the story about that technology goes mainstream. That delay creates wild one-day swings on news that, in reality, isnāt news at all.
AI technology is moving so fast that the markets are struggling to keep up.
New report shows fewer clicks, and a changing advertising landscape
This is an interesting piece of research that plays to the Web 4.0 narrative that Iāve been talking about this year. Itās from a company called Equativ who surveyed 4,000 consumers across North America and Europe about their AI usage, habits, expectations.
The research confirms many of the things we already know - search habits are changing, itās happening faster and to a great extent amongst younger audiences, and this shift is redefining the digital experience. What I found interesting though is that the research also dug into advertising on AI platforms, which isnāt something Iād seen before.
80% of respondents said that they were receptive to advertising on AI platforms to support free access but only if it was sponsored content thatās clearly labeled and contextually relevant. So, not display ads, not lots of clutter, and not things that are irrelevant. I still think thereās a lot of work to be done to figure out the advertising model on AI platforms, but from this research it sounds like audiences would at least be receptive to it if done right.
But, this should all be taken with a pinch of salt as Equativ are an ad tech platform, so obviously have a vested interest in pushing this narrative!
Web 4.0
Mirror publisher puts 600 jobs at risk amid AI and reader changes
Sam Altman says that bots are making social media feel āfakeā
Dead Internet Theory Lives: One Out of Three of You Is a Bot
AI Ethics News
Judge puts Anthropicās $1.5 billion book piracy settlement on hold
FTC launches inquiry into AI chatbot companions from Meta, OpenAI, and others
OpenAI secures Microsoftās blessing to transition its for-profit arm
A California bill that would regulate AI companion chatbots is close to becoming law
ChatGPT may start alerting authorities about youngsters considering suicide, says CEO
How thousands of āoverworked, underpaidā humans train Googleās AI to seem smart
Encyclopedia Britannica and Merriam-Webster sue Perplexity for copying their definitions
Thinking Machines Lab wants to make AI models more consistent
Long Reads
One Useful Thing - On Working with Wizards
āThe future is already here, itās just not evenly distributed.ā
William Gibson