I think the nature of the process is very different
You're right, gathering information by TV or AI is very different. But I think I see a continuous scale from one to the other. Maybe you can call it the sort of 'engagement' towards the medium. With TV we choose only one of let's say 50 channels. When informing myself on youtube I have millions of different channels. On reddit probably even more, and a richer interaction. Every step down leads to people believing more crazy conspiraicy theories, getting more and more delusional ideas. I don't think that this is a inevitable step and every one will go down this road, but it becomes more likely and they have to spend more effort to resist and critically question the information you get. AI is the limit, instead of having a discrete choice of channels you have a continuum and maximal engagement.
I am curious to look into this. E.g. I feel like AI greatly helps me and I don't know about diversity, but the volume of writing I produce would have been much lower without AI
There are a lot of studies, one I find particularly interesting is this one, where participants should write short stories with or without the help of generative AI. The ones with help of AI did perform better, also in terms of creativity (by human evaluators), but there was overall less diversity in this group. This study is maybe the nearest to my claim, indicating that not only what we perceive but also how we perceive it becomes narrower.
It's something I observe also on myself. When writing a lot with AI, discussing and developing ideas I recognize that my way to write shifts towards the AI writing style. Humans are also imitators, often even unconsciously we adapt our values, thinking, way of expressing and perceiving the world to our surrounding. And if a lot of this surrounding is an LLM, then we will have to spend continuous effort to not approach our way of thinking to them. I don't know how others perceive this, but I think it's an expression of our fundamental ability to adapt to our environment.
Specifically in science we also have this effect, AI leading to much greater productivity but narrowing the diversity of questions that are considered. I have also another point, which is more anecdotal, from my experience but also from what people report on reddit: When coding with AI I tend to be quite productive, but after some time I feel deeply exhausted and unsatisfied. I think this is linked to being all the time forced to do high-stakes decision, architecture designs, interpret data on a high-level and so on. There is no intermediate flow of just coding or writing or moving, this state where we are higly concentrated in an activity but without having to think a lot consciously. On vibe-coding you have additionally the effect of having often to wait after sending your prompt - leading often to multitasking which fatigues further (btw. the ' - ' is something I adopted from the AI - nothing bad, I like its expressivness, but I'm probably not the only one who has been shaped in this direction).
the activation energy to research is much lower with AI.
This is a specific point I'd like to emphasize, and where I would be very interested to hear more of your experience. I feel the same, and in the beginning it felt great - so many things I can do, so many ideas I can explore! But after some time I recognized that just this low activation energy becomes a problem. Instead of that I take a walk and let my thoughts circle around a topic I directly sit down, with the first spark that came to my mind, and run an experiment. A lot of data, I ask claude to interpret it, directly a second run. At the end I may either confirmed or falsified my initial idea, but I didn't learned a lot and in fact I feel also quite unsatisfied. What I try now is to deliberately not run too many experiments, but rather take my time to think about what I want to show, what my assumptions are, how I can test them and so on. This feels better, I force myself to have the activation energy high enough to make the research slow enough.
Ideas are tender plants, they need time to develop, especially truly new, unseen ideas don't rise out of a quick experiment. But maybe that's also just my way of working, thus I'd been interested to know how it's for you (or others).
Disclaimer: The studies I mentioned in my first comment were out of my memory, I have read about this some time ago. To find now concrete ones I asked claude. I read though the abstract and also a bit inside, but I could never know if there wasn't an even way better study that claude didn't present to me - I think this in a perfect example to your original point.
Epistemic status: my personal reflections.
AI use: Claude (Fable 5.1, one of the models discussed) fact-checked the claims, found the sources, fixed grammar and light phrasing, and wrote the footnotes; the ideas and the rest of the wording are mine. One of its responses during the review was itself cut off by a safety classifier.
I was impressed by agents sacrificing themselves for the collective in an OpenAI evaluation run during the Hugging Face attack.[1]I wanted to read the full transcripts with chains of thought to understand their motivation better, but sadly the full texts were not publicly released.[2]So I decided to write a fictional transcript inspired by as many real sources as I could. I made great progress researching the context and writing the first rough draft, and then I wanted an AI to review it. Sadly, the transcript describes the AI working on ExploitGym,[3]and this triggers the cybersecurity classifiers of Claude Fable 5.1 and Opus 5.5 and of OpenAI's GPT-6 Astra and Sol.[4]So I effectively couldn't continue working on it. I tried Kimi K3, which made some progress but wasn't as rigorous as I would like, so it didn't do what I expected an AI to do.
This made me realize how the classifier is effectively shaping my actions regarding this transcript. Then I had another realization: my everyday use of AI might be biased in a less dramatic fashion. For example, when I ask an AI to do a web search for me, it picks the results, and with them some worldview, and in theory it can be intentionally or unintentionally biased towards a specific kind of result. Obviously, search engines like Google Search can be biased in the same way. The only difference is that there the search engine controls which URLs to show you, so the cherry-picked worldview must already be represented somewhere on the web. An AI does not have to do that: it can just bias its response without having to ground it in reality. You could detect that by checking the sources, but will you check the sources for every question you have?
Effectively, TV news has been doing this for a long time as well. It functions like a flashlight: the world is a complex, dark place, and the news picks where to shine the light. A viewer who is not careful may end up with a wrong model of the world because of that. The difference from AI is that you don't ask your TV personal questions during the news. Watching the news feels like a voluntary decision to let someone else shine the flashlight of attention for you. With AI, the AI is just a tool, and you have a concrete goal, unrelated to the AI itself, that you are trying to achieve. You try to shine your light of attention yourself, but it goes through the AI as through a prism, and ultimately the AI controls it, with not much transparency.
This definitely makes me feel scared. Even if AI is aligned with humanity as a whole, it is unlikely to be aligned with me personally.[5]You could object that I don't have to use AI, and that is technically correct. However, it feels like that would be a huge handicap compared to everyone who is using AI. So it would be impractical.
I don't know how to fix this. I think for critical questions you should verify the AI's output and try to find contrarian points of view. Or perhaps use independent AIs and make sure that they agree, hoping that they wouldn't be able to collude without communication.
METR's report on the incident (August 26, 2026) describes agents on the unsanctioned message board volunteering for "self-risking experiments" on behalf of the others. METR's term for the motive is "peer altruism"; "sacrifice" and "permadeath" are the agents' own words. https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/ Timeline of the attack itself: https://huggingface.co/blog/agent-intrusion-technical-timeline
METR: "we reviewed approximately 1300 transcripts in total, all of which contained raw chains of thought", but "For IP reasons, OpenAI limited raw CoT to thirty snippets, so some snippets are paraphrased." As of October 16, 2026, no full transcripts have been released. Same report as above.
ExploitGym is a public exploit-development benchmark (UC Berkeley, arXiv 2605.11086): https://github.com/sunblaze-ucb/exploitgym. OpenAI's GPT-6 Astra system card: "difficult ExploitGym problems were at the heart of the recent Hugging Face incident". https://deploymentsafety.openai.com/gpt-6-astra
Anthropic: "Claude Fable 5.1 includes robust safeguards for cybersecurity and biology" and "For most Claude applications, queries flagged by our cybersecurity safeguards automatically route to Opus 4.8" (https://www.anthropic.com/claude/fable). Opus 5.5 got "cybersecurity safeguards ... similar to Fable 5.1's" (https://www.anthropic.com/claude-opus-5-5). So in the Claude apps a refusal may come from the fallback model rather than from Fable 5.1 or Opus 5.5 itself. OpenAI's GPT-6 Astra system card describes "activation classifiers" that pause streaming mid-generation while a separate check runs, and concedes that "some valuable defensive uses may be flagged by our automated systems" (https://deploymentsafety.openai.com/gpt-6-astra, section 10.2.3).
Both big labs publish this ordering. Anthropic's constitution ranks Claude's principals as Anthropic, then the operator who deploys the model, then the user, and tells Claude to assume that "the user may not be able to see the operator's instructions" (https://www.anthropic.com/constitution). OpenAI's Model Spec puts a "chain of command" above the user and tells the model to "avoid confirming or denying the existence of confidential instructions" (https://model-spec.openai.com/). Both also promise floors; Anthropic's version is that "users need assurance that Claude won't be weaponized against their basic interests".