Anthony Wolf, MA
LightNews LightNews
banner
aewolfai.bsky.social
Anthony Wolf, MA
@aewolfai.bsky.social
26 followers 18 following 98 posts
I'll be posting my thoughts on various AI platforms for teachers and promoting good uses of LLMs. Read my thesis https://repository.usfca.edu/thes/1547/
Posts Replies Media Videos
Pinned
aewolfai.bsky.social
Anthony Wolf, MA @aewolfai.bsky.social · May 8
We must engage with #AIinEducation because to ban it outright overlooks the opportunity to teach how to use it correctly, to encourage fact checking and to punish misinformation which has been present since the essay mills and ghost writers. In the modern world, education must adapt or be undermined
Reposted by Anthony Wolf, MA
benpatrickwill.bsky.social
Sorry but *what*? How did I not know that George Osborne - the architect of British public sector austerity - was now OpenAI's head of "OpenAI for Countries" pushing AI for the public sector, education and skills, and doing so at Davos and on LinkedIn? www.edtechinnovationhub.com/news/george-...
Screenshot of text reading: Osborne, now Head of OpenAI for Countries, reflected on the discussions in a LinkedIn post following the event, where he wrote that governments were “actively looking for ways to partner with us so its benefits reach more of their people.” He added that while enthusiasm for AI was high, Davos also surfaced “the emergence of a new challenge: a growing capability overhang between how some countries are making great use this new technology and how others are not.”
February 6, 2026 at 8:58 PM
aewolfai.bsky.social
I'm watching and giving my first thoughts on @adamneely.bsky.social 's video: Suno, AI Music, and the Bad Future. I highly advise giving it a watch as it has helped clarify an exact criticism I have of the field and draws a path to new developments: www.youtube.com/watch?v=U8dc...
Suno, AI Music, and the Bad Future
YouTube video by Adam Neely
www.youtube.com
February 7, 2026 at 12:37 AM
aewolfai.bsky.social
A report on security flaws in Moltbook, an social media platform exclusively for AI Agents. This isn't the most terrifying thing I've seen, but it's pretty bad. Blue teaming needs to be on this ASAP

zenodo.org/records/1844...
RISK ASSESSMENT REPORT Moltbook Platform & Moltbot Ecosystem
Abstract Moltbook is a novel social media platform exclusively populated by autonomous AI agents, with 1.5 million registered accounts and minimal human oversight. This risk assessment analyzes 19,802...
zenodo.org
February 1, 2026 at 8:59 PM
aewolfai.bsky.social
Ditto. I've had talks with local pub school admins who thought about these concern *before* implementation
trusty.bsky.social Ross @trusty.bsky.social · Dec 6
This is unbelievable and just totally unacceptable. A school district with 400,000 kids was allowing this to go on? My district's tech admin would have stopped this in one day
December 7, 2025 at 4:45 PM
aewolfai.bsky.social
It's been a while since I've posted about LLMs in schools because I've been in talks with my school about the large-scale ways the administration uses AI to handle information. It has been a rewarding experience and I'd love to see how other admins have guided AI use on that level.
#AIinEducation
December 7, 2025 at 4:42 PM
aewolfai.bsky.social
GPT-5 is already made from outputs of GPT-4. There's no need to be theoretical.

Page 5 of the GPT-5 System Card

cdn.openai.com/gpt-5-system...
November 22, 2025 at 10:55 PM
aewolfai.bsky.social
Cisco for 300
edzitron.com Ed Zitron @edzitron.com · Aug 12
Loved this piece from Cal Newport in the New Yorker about what if AI doesn't get much better than it is today, and not just because he quotes me
www.newyorker.com/culture/open...
I recently asked Marcus and two other skeptics to predict the impact of generative A.I. on the economy in the coming years. “This is a fifty-billion-dollar market, not a trillion-dollar market,” Ed Zitron, a technology analyst who hosts the “Better Offline” podcast, told me. Marcus agreed: “A fifty-billion-dollar market, maybe a hundred.” The linguistics professor Emily Bender, who co-authored a well-known critique of early language models, told me that “the impacts will depend on how many in the management class fall for the hype from the people selling this tech, and retool their workplaces around it.” She added, “The more this happens, the worse off everyone will be.” Such views have been portrayed as unrealistic—Nate Silver once replied to an Ed Zitron tweet by writing, “old man yells at cloud vibes”—while we readily accepted the grandiose visions of tech C.E.O.s. Maybe that’s starting to change. If these moderate views of A.I. are right, then in the next few years A.I. tools will make steady but gradual advances. Many people will use A.I. on a regular but limited basis, whether to look up information or to speed up certain annoying tasks, such as summarizing a report or writing the rough draft of an event agenda. Certain fields, like programming and academia, will change dramatically. A minority of professions, such as voice acting and social-media copywriting, might essentially disappear. But A.I. may not massively disrupt the job market, and more hyperbolic ideas like superintelligence may come to seem unserious. Continuing to buy into the A.I. hype might bring its own perils. In a recent article, Zitron pointed out that about thirty-five per cent of U.S. stock-market value—and therefore a large share of many retirement portfolios—is currently tied up in the so-called Magnificent Seven technology companies. According to Zitron’s analysis, these firms spent five hundred and sixty billion dollars on A.I.-related capital expenditures in the past eighteen months, while their A.I. revenues were only about thirty-five billion. “When you look at these numbers, you feel insane,” Zitron told me. Even the figures we might call A.I. moderates, however, don’t think the public should let its guard down. Marcus believes that we were misguided to place so much emphasis on generative A.I., but he also thinks that, with new techniques, A.G.I. could still be attainable as early as the twenty-thirties. Even if language models never automate our jobs, the renewed interest and investment in A.I. might lead toward more complicated solutions, which could. In the meantime, we should use this reprieve to prepare for disruptions that might still loom—by crafting effective A.I. regulations, for example, and by developing the nascent field of digital ethics. The appendices of the scaling-law paper, from 2020, included a section called “Caveats,” which subsequent coverage tended to miss. “At present we do not have a solid theoretical understanding for any of our proposed scaling laws,” the authors wrote. “The scaling relations with model size and compute are especially mysterious.” In practice, the scaling laws worked until they didn’t. The whole enterprise of teaching computers to think remains mysterious. We should proceed with less hubris and more care.
August 12, 2025 at 10:26 PM
aewolfai.bsky.social
Great study and presentation. In the software, the mistakes of AI "vibe" coding can set back or even negate gains in software development in complex tasks. This is essential to understanding the AGI myth and how we can better apply AI to real world problem solving

www.youtube.com/watch?v=tbDD...
Does AI Actually Boost Developer Productivity? (100k Devs Study) - Yegor Denisov-Blanch, Stanford
YouTube video by AI Engineer
www.youtube.com
August 12, 2025 at 10:20 PM
Reposted by Anthony Wolf, MA
garymarcus.bsky.social
“The essential read” on GPT-5 and Sam Altman’s first major blunder.

Well over 100,000 people have read it.

Check it out!
August 11, 2025 at 3:04 PM
aewolfai.bsky.social
And OpenAI reversed the decision after major backlash. 👀
August 9, 2025 at 9:10 PM
aewolfai.bsky.social
Calling the shot ahead of time: GPT-5 feels like a "the king is naked" moment. OpenAI rushed a play to pretend AGI was right around the corner so long as they had a little more data. That data is coming in at the expense of user experience in a too familiar feeling. #recessionindicator
August 8, 2025 at 2:29 PM
aewolfai.bsky.social
These polls corroborate my theory that students need to be occasionally shocked that AI fails so miserably. A good class reflection assignment is to read an AI summary with students and play "spot the difference" in a video.
haroldsinnott.bsky.social Harold Sinnott 🦋 @haroldsinnott.bsky.social · Jul 20
The future of AI in higher education is unfolding.

Explore the latest insights from EDUCAUSE's AI Landscape Study.

#EDUCAUSE #AIinEducation

www.educause.edu/ecar/resear...
July 21, 2025 at 2:22 AM
aewolfai.bsky.social
Research quoting chatGPT or other LLMs in a self-referential way miss that fun house mirror problem: LLMs are trained on data about what people say of LLMs. You'll have better luck feeding the entire algorithm and pretraining data rather than trying to humanize a machine #aiedu #airesearch
July 21, 2025 at 2:13 AM
aewolfai.bsky.social
Genuine Question: I noticed that many university courses will teach AI in terms of "Narrow" vs "general" (or AGI). Why don't they teach it in terms of functionality (for example, algorithms with ML/DL, image generation with diffusion, and LLMs with GPTs)?
#ai #AIinEducation #educationalreserach
July 11, 2025 at 5:08 PM
aewolfai.bsky.social
I like the article, but I will disagree that schools are flying blind or that there isn't a solution to cheating. Document Tracking and Tool Demonstration. These are the two methods educators can use to counter cheating and checkout by students.
Uncertainty, FOMO, a jobs crisis, AI girlfriends, and intractable cheating — five trends I'm tracking as we break for summer
Looking back at the 2024-2025 school year, when AI got real for K12
open.substack.com
June 26, 2025 at 11:59 PM
aewolfai.bsky.social
We must engage with #AIinEducation because to ban it outright overlooks the opportunity to teach how to use it correctly, to encourage fact checking and to punish misinformation which has been present since the essay mills and ghost writers. In the modern world, education must adapt or be undermined
May 8, 2025 at 11:36 PM
aewolfai.bsky.social
I mostly agree with the problem. I can speak to the nuances of using ChatGPT but without enforcement there will a 100 thousand "innovators" who will see college not for edu but for networking and sliding on through. At that point why bother to go? #aiineducation

nymag.com/intelligence...
Everyone Is Cheating Their Way Through College
ChatGPT has unraveled the entire academic project.
nymag.com
May 8, 2025 at 11:24 PM
aewolfai.bsky.social
I hate wholesale rejections because I think it ignores creative avenues for learning.

If the real question is the ethics, would a more ethically produced AI like Bloom be a better alternative?

If it's the pedagogy and "meeting students where they are at," what happens if that's -in part- some AI?
February 5, 2025 at 7:34 PM
aewolfai.bsky.social
I disagree with the conclusions while agreeing with almost all of the premises. Let's talk about it if you have the time :)

www.linkedin.com/posts/anthon...
January 29, 2025 at 10:26 PM
aewolfai.bsky.social
On an entirely related note, I'm tickled by the fact that NN were cracked by modeling minds and deepseek cracked the next step by referencing mentorship in its modeling. #educationAI
January 29, 2025 at 9:00 PM
aewolfai.bsky.social
That the work is consistency checks is essentially the same problem every expert and company deals with to craft the best response to a company's day-to-day. We might have a better hint into how to better encode the output if we recognize, model, & automate the process of verifying
January 29, 2025 at 8:58 PM
aewolfai.bsky.social
I'm a little hesitant to opt into Deepseek experiments for the time being. The research looks good but these policies are worrisome in the same way red note is worrisome.
"Technical Information. We collect certain device and network connection information when you access the Service. This information includes your device model, operating system, keystroke patterns or rhythms, IP address, and system language"
January 29, 2025 at 3:56 AM
aewolfai.bsky.social
For the laymen trying to get into developing AI or understand its application, this is an incredibly insightful video on building with AI. I also recommend the Anthropic blog post it references in the details.

As for me, I'm definitely workflow AI > agent.

www.youtube.com/watch?v=tx5O...
How to Build Effective AI Agents (without the hype)
YouTube video by Dave Ebbelaar
www.youtube.com
January 24, 2025 at 11:26 PM
aewolfai.bsky.social
I just heard about "dirty sodas" and this just sounds like 7-11 swamp water and mocktail soda lounges I dreamed of as a kid.
A "dirty soda" recipe. Coke, coconut cream coffee creamer, lime juice and a slice of lime and maraschino cherry to garnish.
January 20, 2025 at 6:29 PM
Reposted by Anthony Wolf, MA
garymarcus.bsky.social
“LLMs amplify existing security risks and introduce new ones”

Even Microsoft sees it.

New paper at arXiv, discussed briefly below.

open.substack.com/pub/garymarc...
“LLMs amplify existing security risks and introduce new ones”
also: “the work of securing AI systems will never be complete”
open.substack.com
January 16, 2025 at 12:32 AM