weekend ai reads for 2026-05-29

“Perhaps there is an ulterior motive for putting those messages out; raising money or whatever,” Hassabis says. “From my point of view, from DeepMind and Google's point of view, if engineers are becoming three or four times more productive, then we just [want to] do three or four times more stuff.”

We find clear racial disparities in applicant outcomes. Of all applications submitted by Asian and Black applicants, 14.74% and 25.87% are submitted to positions that adversely impact Asian and Black applicants, respectively, according to U.S. employment discrimination standards.

People complain about the bad things that happen to em that they dont deserve but they seldom mention the good. About what they done to deserve them things. I dont recall that I ever give the good Lord all that much cause to smile on me. But he did.

[Eric] Carle’s work is like: what if you use only what kids use to make a book, but you put that through one of the best designers to ever do it.

The tension between a hastily torn piece of paper against some super expert typesetting and planning is the excitement in so much of this, right from the cover, and all the way through. The spontaneous joy of when you first made anything put against the precision machinery of book production. Carle is what happens when you’re insanely good at both.

Sinisterra and Wang have so far stayed ahead by emphasising live, novel use cases that the vendor documentation does not yet cover. How long that gap stays open is a different question.

For now, the two-month waitlist is what banks are paying for. The actual training, as several recent client testimonials note, can be replicated by a moderately curious analyst with a corporate ChatGPT licence and a weekend.

The tool was part of CEO Brian Niccol's efforts to fix the coffee chain’s persistent product shortages that he has blamed for hurting sales. The app - designed to improve Starbucks’ visibility into shortages at stores - frequently miscounted and mislabeled items, such as confusing similar milk types or missing them altogether, Reuters reported in February.

New blackboard lecture with Reiner Pope: how do chips actually work - starting with basic logic gates, and working up to why GPUs, TPUs, FPGAs, and the human brain each look the way they do.

Reiner is CEO of MatX, a new chip startup (full disclosure - I’m an angel investor). He was previously at Google, where he worked on software efficiency, compilers, and TPU architecture.

Prince builds his argument on Peter Drucker's 1954 book “The Practice of Management,” dividing roles at every company into three rough buckets: builders, who develop the product; sellers, who close deals; and measurers, who handle oversight and control functions like internal audit, revenue recognition, middle management, and marketing.

…

Drucker argued that measuring matters, but you win customers by building and selling, according to Prince. The best companies pour their money into those two functions.

Closed architectures work when the vendor is winning on capability. Apple closed iOS from a position of product leadership. SAP is closing from a position of AI weakness.

But our analysis shows that only 7%—data reinventors—have reached the level of data-readiness required to scale advanced AI, which includes generative, agentic and physical AI.

As higher ed scrambles to figure out the benefits and harms of AI, the CSU offers an early look at what happens when an administration commits to a technology that its own community isn't convinced will improve education.

Starting summer 2026, UC Berkeley Law is banning AI in nearly all graded work. Students can't use it to brainstorm, draft, outline, write, revise, translate, or proofread anything they submit. Exams are completely off-limits too. The only exception is using AI for research, like finding statutes or case law. But students are on the hook for every fact they cite.

If you treat AI as a tool, you ask: how do I get the right answer? If you treat it as a medium, you ask: what happens if I push this?

The difference is immediate. A prompt stops being an instruction that returns a correct (or not) answer and starts being a sketch—something provisional, something you can revise, distort, overwork. The point is not to nail it on the first try. You are trying to see what these models do under pressure.

In my class, the first principle is simple: Do not take what comes back from prompts at face value. Interrogate it. Iterate on it. Stay with it longer than feels efficient.

…

If you are using AI for efficiency, that is a feature. If you are using it as a design medium, that is a problem—and an opening.

A second principle in the course: You can now build entire projects through prompts. That does not mean you understand them. My students are using AI to write programs that scrape data from the internet, generate information, propose interfaces, draft code, structure arguments. You can get very far, very fast. That speed is real. But it also creates a new kind of risk and opportunity: fluency without understanding.

In this article, we demonstrate that through an indirect prompt injection in a poisoned skill, attackers can exfiltrate files from M365. This is done by exploiting the fact that, unlike other sensitive actions, sending emails and Teams messages to the active user does not require human approval, and opening the compromised messages in Teams or Outlook can trigger attacker-controlled network requests.

Their capabilities and usage tend to focus on speed by increasing the amount of code that can be generated in a specified time frame. Speed is a natural byproduct of high aptitude. When it’s forced, it always leads to lower accuracy. The integration of these tools doesn't tend to focus much on deeper understanding or conciseness.