The text is coming from inside the brain

A few datapoints from Creative Boom’s survey of 882 creatives in the US and UK: half the respondents feel less secure financially versus a year ago; 8% are thinking about leaving the industry; 86% use AI at work but only 10% think AI’s impact on the industry is positive. Randomly, I think the post was written by AI. The survey also asked what the best solutions were. Networking and community came out on top at 57%, with mentorship at 53%.

The financial operations platform Ramp released a study on how AI adoption affected headcounts at 21,559 companies in the United States. Companies that were heavy adopters expanded their workforces by 10%, and entry-level hires grew by 12%. Low-intensity adopters saw no significant change in the size of their staffs. This is consistent with most of the headlines I’ve seen on this topic. Some jobs will go away, but overall, the AI rollout will likely add jobs to the economy.

There is some evidence that specialized AI agents are getting better at making predictions, including stocks. They are not there yet, compared to humans, according to this post, but it’s an interesting (very long) look at the topic. I wouldn’t take the post as gospel, but interesting regardless.

Old school: you figure out how to help people buy something in an app. New school: oops, now AI is also talking to those users and might be able to buy stuff for them. UX designers have to weigh in on how human-AI conversations should play out in digital products.

Weird AI trends: Anthropic and OpenAI release products that compete with their customers (e.g. Figma), and the line between customer and competitor is getting blurry (e.g. Salesforce promoting Anthropic’s competing Slack agent, Claude Tag).

Pharmaceutical manufacturers use Claude to make new drugs happen, but now Anthropic wants to start discovering and developing their own treatments for rare diseases.

That means Anthropic is selling software to its competitors, but Big Pharma is also buying or building AI tools of their own. Anthropic’s terms of service state that it will not train its products on customers’ data when those customers are also competitors.

Maybe not surprising for consumer queries, but this is increasingly happening in b2b queries too, at least for searches around best software. Unsurprisingly, Gemini mentions YouTube 7x more than Instagram and 10x more than TikTok videos.

There are boosters and skeptics, which is basically true for any and all aspects of the AI tsunami.

Now is the time to short the stocks of companies that make keyboards. Meta used a non-invasive magnetoencephalography device on test subjects’ heads, combined with their AI data sauce. The 61% accuracy they achieved with Brain2Qwerty is an almost 800% boost in accuracy vs prior best-in-class options. Amazing stuff.

Last week Mark Zuckerberg, who flushed $80 billion down the toilet on the metaverse, confessed that firing 8,000 Meta employees might have been a mistake. The main problem with the AI agents is that they aren’t … accelerating, it seems. The benefits of the reposturing haven’t materialized. Darn!

The BBC has found that Instagram is running ads promoting child sex abuse in India, including video ads showing minors having sex. In March Meta announced it was increasing its use of AI to screen content instead of third-party human monitors. The ads apparently were given the green light by IG’s moderation technology.

Greater capability does not translate into broader applicability, and that’s true not just in AI labs, but also in biology, where organisms that adapt to “everywhere” die. This seems to not bode well for the arrival of artificial general intelligence.

The fashion magazine tried to roll out some AI-generated content, and engagement with those posts got half the engagement of carbon-based posts. More broadly, “Made with AI” resulted in 7-8% fewer likes across 1,000,000 TikTok posts.

This is from February but has good insights into what we mean by an AI bubble and whether or not we are in one. The comparisons between the AI industry and the browser and operating system wars were interesting. Google, Amazon, Meta, and SpaceX can all build LLMs, like OpenAI and Anthropic, but those companies also have services and products that billions of people use every day. And if Apple wanted to jump into the LLM game, the cost of doing that is going down, and model competition from other companies (e.g. DeepSeek) is increasing.

There are interesting similarities between the current frenzy around AI stocks, the dotcom era, and the global financial crisis. There are also some big differences. A bad day for the AI trade could mean a temporary drop in stock prices, maybe a couple of big names go belly up but the system absorbs them. Or it could mean a systemic shock. There’s less concern when the billions of dollars in infrastructure investment come from a company’s profits. The greater risk is when the build-out is financed by debt. The article refers to a Bank for International Settlements report, which predicted the 2008 financial crisis.

I added a new thing to the AIC website: The AI Bubble Watch. The page tracks 4 measures (for now) that may signal problems in the financial system. I hope to tweak and improve it over time. You can sign up to get an email alert if the dials start freaking out.