Attack of the jargon cave?

I paraphrase: Help Scout’s competitors pump out AI content at scale, yet Help Scout still gets more mentions and citations in AI Overviews. The secret: writing their own content. Ethan Smith at Graphite agreed: less than 20% of articles cited by ChatGPT are generated using AI while 80% are written by humans.

Caveat: These are claims by the respective companies. However, this does align with the idea that LLMs privilege quality content.

Lots of other good AEO insights at that post too.

I dug this review of Claude Design; the highs, the lows, the comparisons with other products. One thing that hadn’t occurred to me about using AI in design: waiting for AI to process your prompts can yank you out of your design flow and elevate the risk of wandering off for snacks or social media rabbit holes.

The AI data center boom is having one not so great knock-on effect: driving up the price of hard drives and storage. A 2TB Samsung SSD that cost $159 last fall now costs $575. Ouch!

And that’s if you can find them. This is causing headaches for the Internet Archive, Wikipedia, and the Wayback Machine.

Google is trying to get people to dive into its Flow Studio. To that end they will have 4 C-suite creatives use the tool to design creative campaigns for brands of their choice. I can’t wait to see what they come up with. The initiative will also feature background on the creatives used the tool.

Jack is the co-founder and head of policy at Anthropic, and yes, he thinks there’s a 60%+ chance that LLMs will be able to improve themselves, with no human involvement, by the end of 2028. Take heart though, Jack is not sure society is ready for it. Make what you will of the prediction, but the advances he points to in his post are impressive.

“When SWE-Bench launched in late 2023 the best score at the time was Claude 2 which had an overall success rate of ~2%. Claude Mythos Preview gets 93.9%.”

And:

“In 2022, GPT 3.5 could do tasks that might take a person about ~30 seconds. In 2023, this rose to 4 minutes with GPT-4. In 2024, this rose to 40 minutes (o1). In 2025, it reached ~6 hours (GPT 5.2 (High)). In 2026, it has already risen to ~12 hours (Opus 4.6). Ajeya Cotra, a longtime AI forecaster who works at METR, thinks it isn’t unreasonable to expect AI systems to do tasks that take ~100 hours by the end of 2026.”

Clipping is taking longer videos and chopping them into shorter snippets to pump out onto social media. The practice recently was in the news because the agency Chaotic Good used the practice to boost the ~~psyop~~ band Geese on social media algorithms. Also Open AI paid $200 million, not a typo, for TBPN — a podcast with 7,000 viewers, but with all the clipping, they reach hundreds of thousands of viewers.

While some agencies report dialing back hiring for junior positions, there are still opportunities for the youngs, and different agencies are handling the situation differently. Some look to junior employees as founts of AI expertise.

May 14. Looks like a solid line-up of speakers.

A new blog where I’ll have more room to post and maybe even break some news occasionally. Profiles of interesting people and companies. Et cetera.

For folks who want to see prior newsletters.

I’m fascinated by what AI and coding are doing to the English language. The Jargon Cave is an online glossary.