With it being Thanksgiving week in the US, its been a quieter week in the world of generative AI, but that didnāt stop Anthropic from releasing a new frontier model before the holidays!
Unsurprisingly, OpenAI, Google and Perplexity all released new shopping assistants in time for the yearās biggest shopping weekend. I think weāre going to see huge growth in the use of generative AI platforms for purchasing throughout next year. There was also a great article from The Verge this week on how ChatGPT canāt tell the time which is well worth checking out.
There are some good Long Reads to take a look at as well with a great piece from Helen Toner on Taking Jaggedness Seriously and The Verge explores how language is not the same as intelligence in Large Language Mistake.
Anthropic introduce Claude Opus 4.5
The headline frontier AI model release of the week was Anthropicās Claude Opus 4.5, their largest and most capable model, which surpasses Claude Sonnet 4.5 across most benchmarks.
Claude Opus 4.5 is cheaper than the previous Opus model, uses fewer reasoning tokens, and has much better safety alignment. Simon Willison has a great write up on it and why evaluating new LLMs is increasingly difficult.
āTis the seasonā¦
So, the yearās biggest shopping season of the year is upon us, so of course many of the major AI platforms wanted to release their new shopping assistants to help us all spend our hard earned cash.
Theyāre all a variation on a theme - a conversational shopping experience that takes your prompt, asks a few questions, narrows down your options, and then presents you with some final recommendations in a comparison table along with images, descriptions, prices etc.
The differences between them are mostly in emphasis. ChatGPT tries to be a research companion that builds a personalised buyerās guide. Google leans into āagentic checkoutā, where the system does more of the work for you and tries to close the loop. Perplexity sits somewhere in the middle with conversational discovery and the option to purchase directly for some retailers.
Iād say theyāre ok-ish. Iām quite a discerning shopper and usual know what I want before I buy it, so product recommendations are never something I really need. For example, I tried it out ChatGPTās Shopping Assistant for my Secret Santa this year and whilst it bought back some interesting ideas it took about 5 minutes to research and didnāt really present me with anything better that I would have got in a standard conversation with ChatGPT. They all feel a bit uninspiring because the experience doesnāt yet beat either traditional research or a quick prompt or search.
I think this is another AI solution in search of a problem. An interesting idea, but one that I donāt think has a compelling enough use case yet, at least for me.
ChatGPT shopping research builds you a buyerās guide using AI
Perplexity says its AI personal shopper āputs you firstā
Why canāt ChatGPT tell time?
This is a fantastic little piece from The Verge about why ChatGPT (and other LLMs) canāt tell the time. It really grabbed me.
The reasons are simple, but I think the issue goes far beyond being able to tell the time. The models, and by extension the AI platforms built around them, have no sense of time at all and I think thatās both a problem and a huge missed opportunity.
If our ambition is for generative AI models and the platforms they power to be truly intelligent and to help people in their day-to-day lives they are going to have to have a very strong awareness and sense of time in general. Time governs all our lives and I think itās a fundamental feature these AI systems will need to have. I can also see huge benefits of having time appended to our conversations - it would open up another dimension (pun intended) for training AI models and help them better understand āturn-takingā in a conversation and allow them to be more proactive.
I hope we see more in this space next year!
AI Ethics News
AI could replace 3m low-skilled jobs in the UK by 2035, research finds
Suno AI Rep Turns the Internet Against Suno AI With One Simple Post
Warner Music Group partners with Suno to offer AI likenesses of its artists
Epic CEO Tim Sweeney says Steam should drop its āMade with AIā tags
Long Reads
Helen Toner - Taking Jaggedness Seriously
The Verge - Large Language Mistake
Benedict Evans - AI, networks and Mechanical Turks
Emerson Collective - In Conversation: OpenAIās Sam Altman and LoveFromās Jony Ive
āThe future is already here, itās just not evenly distributed.ā
William Gibson