😺 Wikipedia is fine (for now)

PLUS: Insane video demos from Google Veo

Welcome, humans.

Quick correction from yesterday’s newsletter: The ChatGPT Voice Mode we discussed is the earlier version, not the new one showcased in OpenAI’s recent demos. The updated Voice Mode isn’t available for Plus users yet. Apologies for the hallucination!

Here’s what you need to know about AI today:

Wikipedia’s traffic hasn’t dropped since ChatGPT.
Perplexity released a feature that lets users share their learnings.
Google showcased more demos from its AI video generator, Veo.
Microsoft developed an AI that predicts extreme weather.

On today’s podcast: A history of Nvidia, WTF are GPUs and intelligence factories, and even more AI chips (Apple Podcasts, Spotify, YouTube).

ChatGPT’s impact on Wikipedia: not horrible (for now).

Post-ChatGPT, some thought it was RIP for Wikipedia.

Why bother with Wikipedia when ChatGPT offers instant answers, follow-ups, and tailored explanations? We’ve personally ditched Wikipedia since ChatGPT. #sorrynotsorry!

But new research shows that Wikipedia is doing just fine. Researchers looked at Wikipedia’s metrics—page views, visitors, edits—before and after ChatGPT’s launch.

Here’s what they found:

There’s no evidence of a decline in engagement.
Page views and visitors have actually gone up since ChatGPT.
So has engagement in langauges that weren’t offered in ChatGPT.

Google Trends, ironically

The main reason is that ChatGPT and Wikipedia serve different needs: Wikipedia is for detailed research and verified facts, while ChatGPT offers quick answers and summaries.

Why it matters: All of Wikipedia’s content is in ChatGPT’s training data, so its ongoing relevance hinges more on trust and content formatting than anything else. Indeed, there are literally Wikipedia GPTs that can browse Wikipedia and regurgitate the same details you’d find online.

We ultimately think AI will transform how we search for information. Imagine a type of Wikipedia that can take any page and summarize it, customize it for the individual reader, and answer follow-up questions.

That’s all possible today with AI. And while we’re cheering for Wikipedia (crowdsourced content matters!), if they don’t adapt soon, someone else will.

Someone like Perplexity, which recently launched a type of Wikipedia called Perplexity Pages (in beta).

Here’s how it works: a user explores/learns about a new topic → Perplexity generates a tailored web article on it. It’s similar to a Wikipedia page, but can be customized for specific audiences (e.g., piano lessons for seniors) and created in 5 days instead of 5 weeks.

Some examples shared by the team:

How to Use an AeroPress (link).
Beginner's Guide to Drumming (link).
Steve Jobs: Visionary CEO (link).

Wikipedia vs Perplexity

FROM OUR PARTNERS

Build Your AI Customer Support Stack.

AI is here to stay, so make it work for you.

Help Sc out’s ebook, Building your AI Support Stack, defines what AI can do for customer service, weighs the pros and cons of integrating AI, runs through the latest AI-powered support tools, and provides expert tips for a thoughtful implementation plan.

Click here to download the guide.

Around the Horn.

Microsoft built a weather forecasting AI named Aurora that can predict and help mitigate extreme weather events.
Anthropic’s Chief of Staff published an article saying that “these next three years might be the last few years that I work” due to development. Hacker News is more skeptical.
AMD announced new AI chips; Nvidia’s 2026 chips will be called “Rubin”.

Treats To Try.

*Get paid to learn LLM training - Evaluate AI-generated responses and help us build better models. Work from anywhere in the US at your preferred time.
Ashby automates parts of the recruiting process for hiring managers (raised $30M).
Weave Bio uses AI to help drug developers draft, review, and submit regulatory documents faster (raised $10M).
daydream uses GPT to generate SEO-optimized content (raised $3.8M).

*This is sponsored content. Advertise in The Neuron here.