Welcome, humans.

There is actually SO much going on in AI right now that it’s now virtually impossible to fit everything in this newsletter.

So today, we're going LIVE to break down Artificial Analysis, the best resource (in our opinion) for tracking which models to use, where to run them, and what the latest performance/pricing charts are telling us about the frontier.

New models drop weekly. Prices shift. Benchmarks change. Like, all the time. We'll show you how to actually make sense of it all and pick the right model for your needs.

Livestream starts at 10am PST / 12pm CT - click here to attend (if you’re early, click “Notify me” on the video link once you click through to YouTube so you can join live).

Here’s what happened in AI today:

1. We break down a new paper on which jobs Americans support automating.
2. NVIDIA invested $2B in xAI’s new $20B funding round.
3. SoftBank acquired ABB's robotics business for $5.4B.
4. South Korea lost 858 TB of government data in a datacenter fire.

Here’s what the general public finds acceptable to automate vs what workers actually want to automate

Ethan Mollick shared a new study asking 2,357 Americans which jobs AI should automate. According to researchers Simon Friis and James Riley, the public is surprisingly supportive:

• Currently, 30% of occupations have majority support for full automation.
• When told to imagine AI that outperforms humans at lower cost, support nearly doubles to 58%.
• Translation: most resistance isn't about principle, it's about performance.

But there's a hard floor. About 12% of occupations like clergy, childcare workers, therapists, judges were considered “morally repugnant” to automate. No matter how good AI gets, these jobs must remain human.

Here’s the most “morally repugnant” jobs to automate according to the public, vs the least repugnant (sorry file clerks!).

Compare that to what Erik Brynjolfsson shared about recent Stanford research asking actual workers what they wanted automated (paper):

• 46% of tasks got positive ratings from workers.
• Top reason? “Freeing up time for high-value work” (69%).
• Workers want to automate boring stuff: scheduling, data entry, filing.
• They DON'T want to automate writing, creative work, decision-making.
• Most want “equal partnership” with AI, not full automation.

Comparing the data across both studies, Erik created this graphic:

Interestingly, the moral floor is consistent across both studies: caregiving, therapy, spiritual guidance remain off-limits for AI broadly speaking.

One twist: the occupations that are seen as morally repugnant to automate typically employ more women and provide higher wages. If this resistance holds, it might mitigate gender inequality, but reinforce current economic inequality.

Translation? We probably need to finally implement some societal guardrails around this stuff before companies decide what to automate based purely on technical possibility (“whether or not we could”) over moral acceptability (“whether or not we should”).

P.S: want to see the future of YOUR job? The Stanford team made their findings public. You can look up your occupation right now in their interactive data explorer and see which tasks workers want to keep versus automate. The full WORKBank database is on Hugging Face with responses from 1,500 workers across 844 tasks, and the GitHub repository has all the analysis code if you want to dig into the methodology.

FROM OUR PARTNERS

Turn Customer Feedback Into Action With Generative AI

Drowning in complaint backlogs and messy customer feedback? In this live demo, see how VantageCloud with generative AI makes resolution faster and smarter.

Learn how to detect complaints in real time, use text clustering, topic detection, and sentiment analysis to uncover root causes, and generate clear explanations with LLMs.

We’ll also show how external LLMs integrate seamlessly with in-database analytics to accelerate resolutions and capture institutional knowledge. While the demo focuses on banking, the approach applies across industries to strengthen retention and customer lifetime value.

Watch Now

Prompt Tip of the Day

You can now “yap to app” on Gemini AI Studio. Just hit the little microphone button for “speech to talk” and you can give your apps live feedback via chat, so you can describe your app idea out loud and watch it get build in real-time. If you want help getting started with building an app like this, we find this guide incredibly helpful.

Treats to Try

1. GLM-4.6 powers coding assistants like Claude Code and Cline to write code, debug errors, and build polished front-end pages, performing close to Claude Sonnet at a fraction of the cost ($3/month for 3× usage, GitHub here, or compare cloud providers here).
2. Retool AI’s new AppGen turns your descriptions into working internal apps connected to your live databases; just type “show me all unshipped orders over $5k with filters” and get a functioning dashboard wired to your Postgres data (free to try).
3. Jamba-Reason-3B runs reasoning tasks on your laptop, like reading 100-page contracts and pulling out critical terms or triaging support tickets by urgency—if you have LM Studio, get it here.
4. Micro AGI provides an “endless stream” of real-world video data of people performing physical tasks, like cooking a meal, welding metal, or fixing a sink, to train your robots and embodied AI systems.
5. NeuTTS Air clones any voice from just 3 seconds of audio and generates speech on your laptop or phone in real-time—try it out here.
6. Deepsite is a fun way to build your own website on a HuggingFace Space; try it out!
7. Tasklet is aiming to be a true “agent” for automating business workflows without flowcharts: set it to create Slack alerts when deals close in HubSpot, and it runs automatically, handling errors and edge cases on its own (founded by ex-Firebase co-founder Andrew Lee, backed by USV & Lightspeed, blog).

Around the Horn

Full video, feel free to get angry and disagree as many have; P.S. he used Tiermaker to do this.

• News:
  • OpenAI's Sora video app reached 627K iOS downloads in its first week, exceeding ChatGPT's launch despite being invite-only.
  • South Korea’s National Information Resources Service datacenter lost 858 terrabytes of data across 70 different government services due to a lithium battery fire; luckily, many companies (ex 1, ex 2, ex3) are investing in new sodium batteries that are less flammable to avoid this issue going forward.
  • OpenAI and Anthropic reportedly explored using investor funds as “self-insurance” to help cover potential multibillion-dollar copyright settlements.
  • SoftBank agreed to buy ABB’s robots business for $5.4B, a play to fuse “Physical AI” with industrial automation; ABB said proceeds will fund core electrification/automation tech and possibly larger acquisitions, with closing targeted for mid-to-late 2026
  • The Bank of England warned that equity valuations appear stretched, particularly for AI-focused technology companies, leaving markets exposed to a sudden correction if expectations around AI's impact become less optimistic.
• Watch:
  • Sam Altman’s interview with A16z on Sora, energy, and building an AI empire, where he said GPT-5 has already made early novel scientific discoveries, OpenAI’s planning more infrastructure partnerships, he’s concerned about DeepSeek's open-source dominance in universities, and reiterated ChatGPT hit 800M weekly active users.
  • NVIDIA invested $2B in xAI's $20B funding round tied to GPU purchases for Musk's Memphis data center.
  • Boston Dynamics released a video showcasing their new second-generation 3-fingered robot gripper that can grab almost anything.

FROM OUR PARTNERS

‘AI consultants are getting paid $900/hr’ –Fortune

Fortune Magazine interviewed Innovating with AI’s founder to discuss a crazy stat – AI engineers are being deployed as consultants at $900/hr.

Why did they interview Rob? Because he’s already trained 1,000+ AI consultants  – and IWAI’s exclusive consultant directory has driven Fortune 500 leads to graduates.

Want to learn how to turn your AI enthusiasm into marketable skills, clear services and a serious business? Enrollment in The AI Consultancy Project is opening soon – and you’ll only hear about it if you apply for access now.

Click here to request access to The AI Consultancy Project

Thursday Trivia

One is real, and one is AI. Which is which?

Which is AI, and which is real?

Which is AI, and which is real? The answer is below, but place your vote to see how your guess everyone else (no cheating now!)

A Cat’s Commentary

Trivia answer: TBH, we still don’t know. If you found the answer in the comments of the post, let us know! Write in via the poll below and add “additional feedback.”