Ben Tsai

Ben Tsai

@bentsai.org

bentsai.org product-minded engineer that advocates for human-centered design. ai is not the answer. @ Cisco @ Carnegie Mellon University MSE

250 followers605 following174 posts

It was refreshing to hear from the venerable Phil Koopman, who keynoted yesterday's Agentic Road Show. His talk stood out to me because it was grounded in reality, informed by decades of research, and provided current practical examples (like don't walk in front of a robotaxi).

View full thread

I started browsing his other articles and was elated to see him assert fundamental truths about LLMs, like, LLMs have no concept of truth. It's fascinating to me how many people disregard or fail to reckon with this.

I will certainly be referencing Phil's work moving forward. (Mea culpa for not following Professor Koopman's work closely until now, despite it coming up in almost every ai-related conversation I've had with Nick Steele.)

It was refreshing to hear from the venerable Phil Koopman, who keynoted yesterday's Agentic Road Show. His talk stood out to me because it was grounded in reality, informed by decades of research, and provided current practical examples (like don't walk in front of a robotaxi).

View full thread

- Every output is a "hallucination," and sometimes it is useful. - No amount of training will solve edge cases. - Be careful how you design the ai + human system. The handoff is key. That is where you handle edge cases, and that's where things will fail.

- Humans should be held accountable for avoidable mistakes, but our current system and discourse affords no one being held accountable.

It was refreshing to hear from the venerable Phil Koopman, who keynoted yesterday's Agentic Road Show. His talk stood out to me because it was grounded in reality, informed by decades of research, and provided current practical examples (like don't walk in front of a robotaxi).

Though the topic was ostensibly around edge cases in autonomous vehicle safety, the lessons and adages apply to everyone deploying ai. Some of the highlights are:

Ben Tsai reposted

I like this notetaking example. it is a good case of how gender works. Notetaking has never been about producing a transcript. It is a bureaucratic skill in the 20th c. organixarion because it is SENSEMAKING. Those women socialized to take notes? They have been doing men’s metacognition for them

This post is unavailable.

Ben Tsai reposted

For those who want to dive deeper into this data, Ed Next (journal at Harvard) has republished my essay on what I call The 50 Percent Problem, my *conservative* estimate of how many students are harmed by using AI. www.educationnext.org/the-penalty-...

educationnext.orgThe Penalty Students Pay for Using AITeenagers are willingly automating cognition and learning less. Adults seem unwilling to stop them.
Benjamin Riley@benjaminjriley.bsky.social

These are indeed the right questions. We have empirical evidence that strongly suggests the immediate impact of AI is seriously harmful to learning, and since learning is cumulative, likely will have long-term harmful consequences. (That purple area is the direct measure of harm)

Ben Tsai reposted

Stock art maintenance - if you've worked with me; I'm a big fan of @kinopio.club; I'm currently working out a way to catalogue my stock art - starting with #Morktober 2026. So this will be a white board experience with what each card/day would release with as I cut em up.

LLM-generated output posing as “writing” is like that junk mail you get that looks important and personal, with a handwritten message. Then you realize it is a fake and immediately toss it in the trash because you know it’s worthless and trying to sell you something.

Every AI-based solution is a bandaid for a symptom and only worsens the entire system. Oh, I need to orchestrate agents to filter out all the noise among the chats and Jira tickets and emails I get. The problem is organizational. But use ai, and you lose sight of that and never fix it.

Another consequence of LLM output having no author is that the output lacks authority. There is no expertise, knowledge, experience of an author to pin this on. If someone takes that output and "reviews" it, they are then participating in something akin to plagiarism.

It's akin to plagiarism because there is no author. That has been obscured/washed by the machine. But the part of presenting someone else's work and expertise as your own, that's still there in spades.

surely i'm missing out on a sponsorship or other opportunity to monetize the fact that my surname has the letters "ai"

Ben Tsai reposted

When I say I've been working on this debunking for *years* that includes academic publications. Here's a quick thread:

Emily M. Bender@emilymbender.bsky.social

So people come back around to those of us who have been pointing out the emporers' nudity all along and say, "How about THIS set of invisible clothes? Does that change your analysis??" >>

Ben Tsai reposted

Who defined the problem? Who controls the system? Who benefits and bears the errors? Who can contest the decision? Massimo Ragnedda and Maria Laura Ruiu propose four questions every consequential AI audit should answer, adding a "power test" alongside the fairness test.

techpolicy.pressAI Audits Need a Power Test, Not Just a Fairness ScoreAs audits enter AI regulation, ask who sets the goals, controls the tools, gains, and can contest results, write Massimo Ragnedda and Maria Laura Ruiu.

Ben Tsai reposted

What's so sad about those who deny this is they must then therefore think humans are formal systems. Which may be true I guess, but we have no evidence for this. So if they truly think this, they can prove it by showing us the formalisms. But I won't be holding my breath for them to write it up. 💀

They are a formal system. You can analyse them with maths. You deny this because it doesn't fit your narrative, but it's factually true. It's pointless to continue because you think a formal system isn't formal. Good luck engaging with academics if you don't read their work.

Ben Tsai reposted

Replying to @tressiemcphd.bsky.social

Anyway, your problem isn‘t finding the time to take notes. The problem is that taking notes is thinking and thinking has to be done by someone. women have been doing a lot of metacognition that looks like a task when it’s really thinking.

I predict that in the future, there will be only two types of people. Those who are right-handed, and those who are not.

Ben Tsai reposted

Replying to @edzitron.com

After a few months of paying the actual cost of AI via token-based billing, an increasingly-anxious conversation has begun about AI's ROI. Per Cisco's Jeetu Patel, the costs of AI tokens are far higher than the actual value that these tokens are generating at scale. www.wheresyoured.at/brokenomics/

Ben Tsai reposted

Dr. Velislava Hillman and Dr. Peter G. Kirchschlaeger argue that as AI erodes job markets, it exposes the reality that the crisis facing young people "was never primarily technological, but political." Time to rethink what education is for, they say.

techpolicy.pressAI Lays Bare the Authoritarianism of Modern Work. Time to Rethink Education.AI now bypasses the original logic of the educational bargain altogether, write Dr. Velislava Hillman and Dr. Peter G. Kirchschlaeger.

the language around the red card being suspended is confusing because without it being suspended, the balogun is suspended. two different suspensions being talked about.