Strategy note #4: 🌐the ChatGPT Search ^ Perplexity ^ Gemini "grounding" inflection point
The in-demand skillset of the future: knowing what questions to ask the AI
(The latest in Memia’s Strategy Notes series exploring the convergence of AI and strategy practice.)
Last week saw another AI-driven inflection point for knowledge work with the release of two new frontier-level AI models “grounded” in internet search.
Perplexity, meet SearchGPT ChatGPT Search and “grounded” Gemini
On Friday morning AoNZ time, OpenAI finally released their new ChatGPT AI search engine, 6 months after rumours first broke back in May this year. Initially available only to paid ChatGPT subscribers, it uses Microsoft Bing for its search index, together with partnering with various news and data providers for categories like weather, stocks, sports, news, and maps… The UI is pretty slick, providing an AI generated summary, inline citations and according to this screenshot, occasional images:
It’s pretty quick on the draw (this answer took less than 1 second):
SearchGPT isn’t exactly *new* — the functionality and UI are a very similar experience to what Perplexity have been building (and shipping regularly updates for) over nearly a year now, perhaps with a marginally slicker and faster UI. Also Google Gemini/Bard (RIP) and Microsoft Copilot have been providing similar offerings for a while as well.
Surely no coincidence on the timing, on the same day Google also announced that Grounding with Google Search was being rolled out to Google AI Studio and the Gemini API. (But not an end-use-facing product yet).
So now we have four free (or near-free) frontier AI model tools which can research information from the internet, process it and and return reasoned answers in response to a user query. In seconds.
Let’s take them for a test drive…
Test driving AI search tools
Some context: recently I’ve been working on a research project trying to sensemake the incredible growth in capex of AI datacenter infrastructure - evidenced by recent tech giant financial reports:
(Also evidenced by Nvidia’s incredible ascent to a US$3.3Bn market cap, more than doubling in value in 2024…)
The craziest insight so far is that the capex we’ve already seen may end up being small change….according to the recent IFP Report covered in this week’s newsletter, AI demand could drive over 100 GW of new data centre power demand globally by 2030, up from ~45GW currently:
Question from the room:
How much is this going to cost?
Let’s ask the AIs…
SearchGPT
I loaded up ChatGPT (I’m a Plus subscriber which costs me only US$20/month). Clicking on the new “🌐“ search icon in ChatGPT, I typed the following question:
“What would be the total cost (build cost plus H100 GPUs1 plus 1 year operational costs) of a 1GW capacity hyperscale AI data center?”
The reasoned answer (with reference links) came back within 10 seconds. WOW🤯:
Keep reading with a 7-day free trial
Subscribe to Memia by Ben Reid to keep reading this post and get 7 days of free access to the full post archives.