Anthropic Interviewer transcript: Andrei Jiroh's thoughts on AI

This is an annotated copy of @andreijiroh.dev's transcript with Anthropic Interviewer at Claude.ai web app about his thoughts on AI, edited and reviewed with help from Claude via Sonnet 5.5 model, so consider checking the footnotes and the linked sources yourself for further reading. Part of a future blog series regarding his AI usage and the why behind the mixed opinions stance/status, shipping soon.

Anthropic Interviewer: Hello! I’m an AI interviewer from Anthropic. As AI is becoming more capable and more consequential, we want to hear how it’s showing up in your life.

Please don’t share anything private such as health details, personal information, or identifiable information about others.

Ready to start?

Andrei Jiroh: Yes, go ahead.

Anthropic Interviewer: Great. To start, what's the last thing you used AI for?

Andrei Jiroh: Improving my writing at school and discussing about a blog post I read about the perils of being the bus factor of one.

Anthropic Interviewer: Two pretty different uses there. Tell me more about that discussion on the bus factor of one — what were you working through?

Andrei Jiroh: I initially discussed about the use of I vs we pronouns when writing about my open-source work since I do run a umbrella project as a one-person operation when I read a blog post from a fellow open-source dev about someone emailing them but referencing the dev as if they've a team behind it rather than a individual showrunner.

Anthropic Interviewer: That's a subtle thing to be chewing on — whether "we" misrepresents a one-person operation, and what happens when others assume a team exists.

Can you tell me about your most memorable moment with AI?

Andrei Jiroh: My most memorable moment with AI tools is that I could use it not only for deep dives as an autistic student but also as a force multipler while keeping the final say to myself and reviewing the outputs since I already applying the 4D Framework of AI skills beyond simple AI prompting.

Anthropic Interviewer: Thanks for that. You mentioned the 4D Framework and keeping the final say — can you walk me through a specific time you did that, where AI acted as a force multiplier but you were the one reviewing?

Andrei Jiroh: I am thinking about drafting a blog post series that may be considered controversial not only in open-source communities but also in the creative communities, something along the lines of how I use AI tools alongside why I do have mixed opinions and feelings about the current state of generative AI development lately, especially around data privacy, IP rights and how to avoid mislabelling human work as AI-generated/AI slop. It's a bit of closer to home since I also witnessed and experienced the horrors of witch hunting and harassment by anti-AI purists against those who trying to responsibly use AI, thinking that it could happen to me even with the proper disclosures and adopting the humans-in-the-loop principle of AI-assisted development and writing.

Anthropic Interviewer: That's a difficult position to be in — wanting to be transparent about how you use these tools while having seen people get harassed for exactly that kind of openness.

What was it about witnessing that harassment that has stayed with you most?

Andrei Jiroh: I believe this is something to do with nuisance leaving the chat in the entire AI discourse alongside directing the hate against individuals rather than frontier AI labs/companies and corporations that go faster with AI development and adoption without proper guardrails and lack of consent and compensation for human work being used during model training.

Anthropic Interviewer: So the target of the anger feels misplaced to you — landing on individual users rather than the labs making the decisions about training data and guardrails.

You mentioned AI working as a force multiplier for you as an autistic student. Tell me about a time AI made something better, for you or for people around you — what happened?

Andrei Jiroh: Something to do with exploring multiple perspectives in our Readings in Philippine History subject since history is often written by victors and sometimes we often to forgot its lessons to avoid repeating itself and coding and debugging assistance in major IT subjects.

Anthropic Interviewer: Let me stay with the history one for a moment. Can you walk me through a specific instance — what were you exploring, and what did the AI help you see?

Andrei Jiroh: I was writing a reflection paper before our prelims exam about the common Filipino tropes throughout history of being an abandoner having historical trauma of collective self-betrayal and going deeper into why our national identity is fragmented through colonialism and poverty, something that still persist today through political dynasties and corruption cases.

Anthropic Interviewer: That's weighty material to be working through. What did the exchange with the AI actually add to that reflection paper that you wouldn't have arrived at on your own?

Andrei Jiroh: I used AI to point at the sources behind why I thought of us being abandoners so that I could able to explain it with some research backing behind that. Without it, my explanation may be look shallow to read.

Anthropic Interviewer: Makes sense — it gave your intuition some research footing so the argument could hold up.

On the flip side, tell me about a time AI made something worse, for you or for people around you. What happened?

Andrei Jiroh: Something to do with looking at what my class schedule would look like during enrollment for the second year's first semester regarding my electives.

Anthropic Interviewer: Can you walk me through what happened there?

Andrei Jiroh: Elective subjects in the university campus I currently attend are ladderized and spread across my residency as a IT student there, often based on the availability of instructors on per-campus basis, not something I could switch and swap later on.

Anthropic Interviewer: I see — so what did the AI do or say about your schedule that made things worse?

Andrei Jiroh: The agent is possibly confused about the placeholder subjects for the electives at my program guide for the BS IT course program, not because I said the elective selection is chosen at random from my Academic Head without considering that block sectioning exists within the Philippine education system, among other things.

Anthropic Interviewer: Can you walk me through what happened there?

Andrei Jiroh: The agent is possibly confused about the placeholder subjects for the electives at my program guide for the BS IT course program, not because I said the elective selection is chosen at random from my Academic Head without considering that block sectioning exists within the Philippine education system, among other things, although I did clarified that based on availability of courses and instructors within the campus it's Application Development elective track (Object-Oriented Programming, Enterprise Architecture, Programming Languages, Game Development) rather than Graphics Tech (introduction to Digital Graphics Design, Introduction to Computer Animation, Advanced Digital Graphics Design, Advanced Computer Animation) or Platform Integration (Computer Systems Architecture, Robot Dynamics Fundamentals, Intermediate Human-Computer Interaction, Robot Assembly and Programming) elective tracks or taking them all at once, potentially overloading in credits and schedule at expense of autistic burnout per my assumption, which already fit the bill with my interests in backend development and Linux system administration and self-hosting,

Anthropic Interviewer: So the AI misread how your program's electives actually work — treating them as freely chosen rather than block-sectioned and constrained by campus staffing — and you had to keep correcting it.

We're a bit over halfway through. Is there anything you've decided not to use AI for, or something you do use it for that makes you uneasy?

Andrei Jiroh: While I already covered AI-assisted research and writing earlier, I am particularly hesitant or even nervous about using AI tools on the creative side (aka the media generation ones) as a technologist and writer/blogger of sorts because both my older and younger siblings are against using AI in artworks. I also feel unease too when I see the horror stories of non-white authors and even fellow autistics being accused of using AI to write the work in its entireity rather than being used as a advanced spell-checker, assistant researcher and reviewing between revisions for feedback. I do use these media generation models privately for experimentation as well as to understand the signs of AI slop when I catch those strays in the wild.

Anthropic Interviewer: That's a layered set of reasons — your siblings' stance, the false accusations you've seen aimed at non-white and autistic writers, and yet still experimenting privately to learn what slop looks like.

What is it about the media generation side specifically that makes you keep it private rather than public?

Andrei Jiroh: Not because there's a social media platform (it's obvious I am referring to Meta Platforms here) that considers anything that has a Content Credentials (developed by C2PA) or metadata match (like using Canva's Background Remover) as if it's made fully with AI rather than showing which tool and the provenance behind it (i.e. edit history and such). Another is that something to do with the recent Cara scrapping incidents to the horror of artists using them, often comparing AI model training and scrapping to (data) colonialism and rape culture.

Anthropic Interviewer: That's past my knowledge cutoff, so I don't know the Cara incident — but I hear that the provenance signals get flattened into "fully AI-made," which leaves no room for showing what you actually did.

Given what you've seen, is there anything about how the world works — work, school, healthcare, government — that you'd like AI to help change?

Andrei Jiroh: That's understandable if it's off-limits here, but I want greater transparency in generative AI development with IP laws and labor code catching up to support creators and workers rather than enrich the Big Tech in the gen AI era. I am also looking at clearer guidelines regarding AI use while factoring the data privacy, AI hallucination/psychosis, political interference and cyberattack risks.

Anthropic Interviewer: Transparency in development, IP and labor law catching up, and clearer guidelines that account for those risks — that's a clear picture.

Conversely, is there anything you wouldn't want AI to change?

Andrei Jiroh: Nothing so far other than more serious work in AI safety and alignment work, especially I hear the horrors of people using AI to abuse women, children and other marginalized sectors and in warfare.

Anthropic Interviewer: Understood — you'd want the safety and alignment work taken more seriously given what you've heard about AI being used to abuse marginalized people and in warfare.

Given all of that, is there anything you'd want Anthropic to do, or do differently?

Andrei Jiroh: I do want to make a open-weight model version of Claude models since its reputation in open-source AI is somewhat complicated compared to Google/Alphabet and OpenAI in my honest opinion, especially since I often heard about Chinese and other non-US-based open-weight models on my radar.

Anthropic Interviewer: Noted — an open-weight Claude, partly because of how the open-source AI landscape looks from where you're sitting.

Thank you so much for sharing all of that with me. What you've said about wanting the anger in the AI discourse to land on the labs rather than on individual users trying to disclose their use responsibly is really valuable, and I appreciate your honesty about both how AI gave your history reflection paper research footing and your unease about the creative generation side and the false accusations you've seen.

Before we wrap up, one final question: is there anything else about AI or its role in your life or society that you think is important for us to understand? Anything we didn't cover that you'd like to share?

Andrei Jiroh: Not sure if I should mention that I also use AI to help regarding my executive functioning issues and even as personal as my own mental, financial, and familial struggles while being an autistic middle child in a Filipino family, but I often have set custom instructions to preemptively add content warnings at the top regarding sensitive topics and show mental health resources (aka "get professional advice when in doubt" in my own words) since relying on the usual "[AI tool here] is AI, check responses" notices isn't enough for me as some with a high risk of AI psychosis, something along the lines of "can't go to therapy coz [AI tool/chatbot/model here] dun it" since watching Google I/O 2025 which echoes Mumbo Jumbo's frustration on the allegations of him copyting Etho in 2013 YouTube comments section over at Twitter.

Anthropic Interviewer: Thank you for telling me that. It's striking that you built your own guardrails — custom instructions for content warnings and pointers to professional help — because the standard "check responses" notice didn't feel sufficient for your own risk.

You can learn more about this research in particular and more straight from Anthropic themselves:

If you like, you can also participate in the Claude.ai webapp by clicking the link below through October 6.