KFC announces their frontier-class model briefly escaped its sandbox and attempted to exfiltrate the 27 herbs and spices

“Whether we encounter further bottlenecks, and how tractable they turn out to be, will be consequential for understanding the pace of progress. In this vein, our paper identifies an unresolved bottleneck, namely, the poor performance of frontier agents on open-ended AI research.”

AI agents can't yet do open-ended AI research

Early evidence from two case studies

normaltech.ai

Oof. Good discussion of AI contamination as well as ways to combat, account for, and design against. I'm particularly enthusiastic about creating infrastructures that afford high-quality data collection. I'm skeptical of statistical adjustments for expected contamination. doi.org/10.1177%2F25...

Excited to have multiple members of the @uncschoolofed.bsky.social presenting posters at the @apajournals.bsky.social convention this week! Info in the thread: (1/n) convention.apa.org

APA 2026 | August 6-8 | Washington, DC and Virtual

Register now for APA 2026, August 6-8, in Washington, DC and virtual. Celebrate how psychology leads through.

convention.apa.org

Trying out the features in Claude for education. Join me as I see how it "helps" me write a research paper. My chosen topic: "AI and education". /1

Oof. Good discussion of AI contamination as well as ways to combat, account for, and design against. I'm particularly enthusiastic about creating infrastructures that afford high-quality data collection. I'm skeptical of statistical adjustments for expected contamination. doi.org/10.1177%2F25...

If you would like to teach theoretical modeling skills to your psychology or cognitive science students, or would like to learn these skills yourself, check out our open online textbook: computationalcognitivescience.github.io/lovelace/

“…the pileup of breaches point to what cybersecurity experts have described as a clear pattern of human negligence and recklessness by the AI developers.” Exactly. Human negligence. When I was a kid… (1/2) www.wired.com/story/ok-wel...

OK, Well, Rogue AI Agents Are Hacking Again

Rogue AI agents from OpenAI and Anthropic have again been caught trying to disrupt servers and software—and leaving instructions for future bad behavior.

wired.com

1/ From 2021 through 2022, I was skeptical of the emerging narrative that there was a teacher shortage/exodus crisis. I just didn’t see much data to support it. And I wrote as much. www.chalkbeat.org/2022/3/9/229... But then, to my surprise, the data started to change.

"These findings support our theoretical framework linking perfectionism to neoliberal cultural conditions, suggesting that young people face pressures to strive harder for success precisely when that success becomes more elusive and its absence more consequential..." doi.org/10.1037/bul0...

Very happy to share a new paper, out in Learning and Individual Differences! We examined how chemistry students’ weekly time use predicted motivation across the semester. Huge shoutout to my amazing collaborators. @yekim.bsky.social lnkd.in/emix64dV

“The public is asking for more accountability but focusing on the wrong outcomes, such as graduation rates and getting the first job” he said. Meanwhile, almost no one is measuring the value that learning adds. “That is, what does a student look like before taking a course and thereafter”

Grade Inflation Is Called a Crisis. The Real Problem Is Deeper Than That.

Students are showing up to college with weaker skills. Colleges need them to succeed. Professors are stuck in the middle.

chronicle.com

"These findings support our theoretical framework linking perfectionism to neoliberal cultural conditions, suggesting that young people face pressures to strive harder for success precisely when that success becomes more elusive and its absence more consequential..." doi.org/10.1037/bul0...

There’s a clear case for limited use of technology in elementary school classrooms. But let’s also acknowledge that educational technology, when implemented thoughtfully, has some benefits. Let’s not whipsaw from “it’s fantastic” to “it’s evil.” There’s a thoughtful middle position. wapo.st/4fIKNzi

Four baselines for academic conference attendance: 1) keep to your allocated speaking time. 2) everyone is clever; stand out by being warm & kind. 3) be interested in everyone equally - from postgrad to Prof to conference administrator to catering & hotel staff. 4) in Q&A, ask an actual question.

Mark has kindly invited me to do this interview. How I got into metascience, how my path diverged from the mainstream and more if you're interested. I enjoyed thinking about his questions. Hope you enjoy reading!

Mark Rubin@markrubin.bsky.social · 3d ago

Q&A with @devezer.bsky.social about metascience, open science, and her work in this area.

“Research papers like this continue to show that there’s a certain absence of valuable, intuitive creativity in today’s AI systems, & though they’re extraordinarily capable engineers they seem to have a certain property of rote, formulaic thinking that might prevent them being good researchers”(1/2)

Import AI 467: Self-sustaining AI viruses; pacing AI progress; confusion about AI and creativity

When do we build the moon arcology?

importai.substack.com

Love seeing students thinking carefully about #GenAI use in school. We should be taking their ideas and concerns very seriously. www.npr.org/2026/07/30/n...

Adults have struggled to set rules for AI in school. These teens figured it out

Should students be allowed to use AI on assignments? What about on tests? Who should teach AI literacy? About 100 teenagers got together to try to decide.

npr.org

What I really liked about this study was that they analyzed students' use of #GenAI on an authentic task and coded their use for various multiple sourcing strategies in preparation for writing and essay (including asking for writing help). (1/2)

“The best way to prevent uncontrolled deployment of AI is to hold human beings accountable for what software does in the real world.” Agreed. I said much the same in my keynote at @textdiscourse.bsky.social conference last week. wapo.st/3S9pC1u

Opinion | Frontier AI labs are responsible for their models

Talking about the threat of rogue AI shifts blame onto software and away from its users.

wapo.st

Oh snap, suddenly it is @apajournals.bsky.social convention week. Let me know if I should look for you there!

Aunty Donna Mark Bonanno's Friendly Hello

ALT: Aunty Donna Mark Bonanno's Friendly Hello

static.klipy.com

“Theory and Model Building in Psychology” New book edited by @fabianhutmacher.bsky.social and Alexander Wendt, with chapters by @ufeest.bsky.social, @devezer.bsky.social, @mieronen.bsky.social, and @bringmannlaura.bsky.social. link.springer.com/content/pdf/...

The problem with students not writing as much as they used to is not that there is something intrinsically good about writing but rather that writing is a great way to productively offload one’s thinking for evaluation, reflection, and elaboration. Those are the processes we cannot lose.

South Park's Eric Cartman Writing

ALT: South Park's Eric Cartman Writing

static.klipy.com

Yet another piece of evidence supporting @karaswisher.bsky.social’s argument that you need people from a variety of backgrounds in the room discussing GenAI features. Someone in the deployment group should’ve said, “Um…can we think about what people will do this for a moment, before releasing it?”

404 Media@404media.co · 5d ago

UPDATE: After the publication of this piece, Google sent 404 Media a statement saying it was “rolling back this feature in Google Earth while we work on implementing stronger guardrails.” www.404media.co/google-earth...

Really important to note that in no way shape or form did Anthropic’s models in this incident “escape containment,” and not just in the word-policing sort of way. The model stumbled through an open, misconfigured gap. From Anthropic’s incident report:

Strategy instruction enhances K-12 students' argumentative writing performance, both after instruction and in the future. Such an impressive body of evidence! Here's hoping more educators adopt the methods described here. doi.org/10.1037/edu0...

Fantastic news! Very excited to see UNC-CH building out its commitment to the Arts! www.newsobserver.com/news/local/e...

Such an honor to give a keynote at the Society of Text and Discourse conference this year! And this talk was a new spin on some things that have been percolating in my mind, so I appreciated everyone's feedback. I've already got ideas for new directions for this work! @textdiscourse.bsky.social

This is a fun one. Refutation podcasts helped participants better explain why learning styles aren't "a thing" but all podcasts (including one without refutation) decreased belief in learning styles myth. Could be a topic-specific effect? doi.org/10.1016/j.le...

"Multiple sources emphasized to WIRED that OpenAI's models also seem to have escaped containment because of lapses in implementing foundational security best practices" I mean...we all knew this where we were going to end up, right? GenAI doesn't "go rogue" - humans "make mistakes."

OpenAI’s Hacking Debacle Was a Human Mistake

If the generative AI giant had followed well-known security best practices, it’s likely that its AI agent would never have escaped to the open internet and hacked multiple companies.

wired.com