# @thesupermannx: X profile **Lists:** [[Users/@thesupermannx/Chronological List|Chronological list of paper posts]] · [[Users/@thesupermannx/Topical List|Topical list]] **Related pattern:** [[Topics/Ribosome Pausing and Cotranslational Folding/Related Trends#1. "AI cracked open X" posts that re-label old science|Ribosome topic: the "AI cracked open X" pattern]] ## Identity - **Display name:** Superman · **Handle:** [@thesupermannx](https://x.com/thesupermannx) · **User ID:** 2052949858252939264 - **Bio:** "AI & Neuroscience" - **Joined:** May 8, 2026, 10:14 PM CT. The account was about five and a half months old at collection. - **Verified:** blue check (paid) · **Location:** none given - **Pinned post:** [lingbot-map](https://x.com/thesupermannx/status/2077779856050606155) (Jul 16, 2026): "China open-sourced a model that reconstructs any scene in 3D from a regular video, in real-time." - **Real-world identity:** not established ⚠️. Nothing in the profile or posts names a person, employer or credentials. The "Neuroscience" in the bio is not backed by any stated affiliation. - **Possible earlier handle:** the API's photo link for the account's earliest collected post (May 24, 2026) uses the spelling "thesupermanmx". That may be an earlier handle ⚠️ (unconfirmed). ## Stats (X API, Oct 24, 2026, morning CT) | Followers | Following | Posts (lifetime) | Listed | |---:|---:|---:|---:| | 20,295 | 19 | 625 | 209 | ## Collected posts - **Collection:** the full timeline, 420 items from May 24 to Oct 24, 2026. That is 222 original posts; the rest are replies and reposts. - **Paper posts:** 164 present a study or paper ([[Users/@thesupermannx/Chronological List|list]]), 109 on AI and 55 on non-AI science or health. - 73 link arXiv; 21 have no source link at all ⚠️. - Together they drew about 29.5M impressions. - **The other 64 originals:** mostly "someone just open-sourced…" repo and tool posts. **Paper posts by month:** | May | Jun | Jul | Aug | Sep | Oct (to 24th) | |---:|---:|---:|---:|---:|---:| | 1 | 7 | 3 | 42 | 70 | 41 | **The arc:** - **June and July:** mostly open-source repo hype. - **From August:** a switch to paper explainers, usually 2–4 a day. - **The breakout:** the first viral paper post (Japan, Jul 20) is followed by steady growth. Pop-science and health studies ("bedtime IQ", Alzheimer's, coffee, tattoos) are mixed in with AI. ### Most-viewed paper posts (Oct 24, 2026) | Posted (CT) | Post | Impressions | Analysis | |---|---|---:|---| | Jul 20, 4:09 AM | "Researchers proved every major LLM is secretly obsessed with Japan." | 3,757,239 | [[Users/@thesupermannx/Post Analyses/Post Analysis - LLMs Obsessed with Japan (2026-10-21)\|yes]] | | Aug 15, 4:00 AM | "Two economists mathematically proved that AI will destroy the economy." | 3,197,226 | [[Users/@thesupermannx/Post Analyses/Post Analysis - AI Layoff Trap (2026-08-15)\|yes]] | | Oct 21, 8:08 AM | "Researchers proved every LLM is secretly obsessed with Japan." | 2,424,854 | [[Users/@thesupermannx/Post Analyses/Post Analysis - LLMs Obsessed with Japan (2026-10-21)\|yes]] | | Aug 10, 3:28 AM | "Going to bed late is associated with a higher IQ." (non-AI) | 1,900,224 | no | | Oct 22, 7:53 AM | "Scientists found a way to flush Alzheimer's plaque out of the brain in minutes." (non-AI) | 1,420,720 | no ⚠️ | | Sep 26, 6:13 AM | "ChatGPT has now a big problem." | 1,316,316 | [[Users/@thesupermannx/Post Analyses/Post Analysis - Model Collapse (2026-09-26)\|yes]] | | Sep 28, 5:34 AM | "Researchers proved AI has deleted every reason universities exist." | 1,273,398 | [[Users/@thesupermannx/Post Analyses/Post Analysis - Harvard AI Tutor (2026-09-28)\|yes]] | | Oct 7, 9:35 AM | "Scientists are losing their minds over this paper." (mitochondria "quantum device", non-AI) | 1,197,199 | no | | Aug 23, 6:50 AM | "MIT mathematically proved that ChatGPT is designed to make you delusional." | 1,121,790 | [[Users/@thesupermannx/Post Analyses/Post Analysis - Delusional Spiraling (2026-08-23)\|yes]] | | Sep 4, 11:50 AM | "MIT mathematically proved that AI will destroy the democracy." | 1,000,845 | [[Users/@thesupermannx/Post Analyses/Post Analysis - Automation and Repression (2026-09-04)\|yes]] | | Aug 11, 12:23 AM | "Anthropic scientists did something terrifying." | 874,022 | [[Users/@thesupermannx/Post Analyses/Post Analysis - Claude Introspection (2026-08-11)\|yes]] | ### Analyzed posts | Posted (CT) | Paper | Verdict (short) | |---|---|---| | Oct 24, 4:59 AM | [[Users/@thesupermannx/Post Analyses/Post Analysis - Emotion Circuits (2026-10-24)\|Emotion circuits]] (arXiv 2510.11328) | The paper is real. "Experience fear", the "lobotomy" and the safety bypass are not in it. **The image shows a different paper** | | Oct 21, 8:08 AM (+ Jul 20) | [[Users/@thesupermannx/Post Analyses/Post Analysis - LLMs Obsessed with Japan (2026-10-21)\|Japan bias]] (arXiv 2604.21751) | The core finding is real. The October rewrite invents mechanisms (quantization, "anime wiki") | | Oct 11, 7:59 AM | [[Users/@thesupermannx/Post Analyses/Post Analysis - LLMs Actually Understand (2026-10-11)\|"LLMs actually understand"]] (arXiv 2507.08017) | Framework accurate. A philosophy synthesis is sold as "proved", and grokking is misattributed. Reposted by [[marshallpittman]] | | Sep 28, 5:34 AM | [[Users/@thesupermannx/Post Analyses/Post Analysis - Harvard AI Tutor (2026-09-28)\|Harvard AI tutor]] (Sci Rep 2025) | Numbers mostly right. The "universities are dead" conclusion contradicts the authors | | Sep 26, 6:13 AM | [[Users/@thesupermannx/Post Analyses/Post Analysis - Model Collapse (2026-09-26)\|Model collapse]] (Nature 2024) | Close paraphrase of the abstract, but a 2024 paper presented as news | | Sep 5, 4:27 AM | [[Users/@thesupermannx/Post Analyses/Post Analysis - Vibe Coding Predictors (2026-09-05)\|Vibe coding]] (arXiv 2603.14133) | Mostly accurate. "China" should be ETH Zurich (self-corrected). Drops a non-significant result | | Sep 4, 11:50 AM | [[Users/@thesupermannx/Post Analyses/Post Analysis - Automation and Repression (2026-09-04)\|Automation and Repression]] (Acemoglu et al. 2026) | A fair theory summary, framed as "MIT mathematically proved" | | Aug 23, 6:50 AM | [[Users/@thesupermannx/Post Analyses/Post Analysis - Delusional Spiraling (2026-08-23)\|Delusional spiraling]] (arXiv 2602.19141) | Simulated agents are presented as real people, and "designed to" is invented | | Aug 15, 4:00 AM | [[Users/@thesupermannx/Post Analyses/Post Analysis - AI Layoff Trap (2026-08-15)\|AI Layoff Trap]] (arXiv 2603.20617) | Mechanics right. A stylized model becomes "proved … will destroy" | | Aug 11, 12:23 AM | [[Users/@thesupermannx/Post Analyses/Post Analysis - Claude Introspection (2026-08-11)\|Claude introspection]] (arXiv 2601.01828) | Real experiment. Omits the ~20% success rate; adds "consciousness" | ## Context **Formula.** A typical post has five parts: - **A hook:** "Researchers proved…", "X mathematically proved…", "…did something terrifying." - **A famous institution in the first line.** - **A long explainer** in note-tweet form. - **A shock turn:** "But here is the darkest part…" - **A sweeping moral.** - **The paper link,** usually in the post or a follow-up reply. **By the numbers (first lines of the 164 paper posts):** - "proved" appears in 47; - "mathematically" in 11; - a big-name lab or university in 52; - 39 of the posts use words like "terrifying", "insane", "devastating" or "unsettling" somewhere in the text. **Hype patterns** found across the analyses: 1. **Certainty inflation.** Theory, simulation or argument becomes "proved". This happens in the Layoff Trap, Automation and Repression, delusional-spiraling and "understand" posts. 2. **Lab and author misattribution:** - "China" for ETH Zurich; - "MIT", "Harvard" or "Wharton" based on one author's affiliation, with co-authors dropped; - philosophers credited with "documenting grokking"; - a screenshot of the wrong paper on the seed post. 3. **Invented mechanisms and stakes.** Examples: the "digital lobotomy" and "bypassed safety filters" in the seed post; quantization and "reddit and anime wiki users" in the Japan post. 4. **Dropped caveats:** - sample size and scope (two lessons, one course); - reliability (~20%); - simulated vs real users; - non-significant results; - the authors' own "not a crisis" or "should not replace" statements. 5. **Old papers presented as news.** Examples: model collapse (2024), the Harvard tutor (2025), the emotion circuits paper (Oct 2025), Natural Selection Favors AIs (2023), ElectionSim (2024). 6. **Recycling.** The same paper or theme is posted again weeks or months later, often with a more sensational rewrite: - Japan (Jul and Oct); - the CUDA agent paper (at least 4 times); - Google 2602.24281 (Aug and Oct); - LLM-to-LLM communication, 2510.03215 (Sep and Oct); - tattoo and coffee studies, each posted twice (whether it is the same study each time was not checked ⚠️). 7. **Minds and consciousness framing** for interpretability work. Examples: the emotion circuits, introspection, "understand" and "pain" posts. 8. **Non-AI pop-health "proved" claims** in the same style. Several are the account's biggest posts (bedtime IQ, Alzheimer's ultrashort waves, mitochondria "quantum device"). These match the Ribosome topic's [[Topics/Ribosome Pausing and Cotranslational Folding/Related Trends#1. "AI cracked open X" posts that re-label old science|"AI cracked open X"]] pattern. The difference is that @thesupermannx usually *does* link a real paper, so the distortion sits in the framing rather than in a missing source. **Responses:** - **Self-correction:** one found. The vibe-coding post's "China" became ETH Zurich, in two replies on Sep 5, 2026. - **Pushback:** repliers and quoters often push back. On the seed post, [@kasetsu_today](https://x.com/kasetsu_today) quoted it with a fact-check ("✅ファクトチェック済み … これは完全に誤り。", English: "✅ Fact-checked … this is completely false."), and @BGM_886 wrote "The paper's real. The framing in the post is in question." ## Interactions with [[marshallpittman]] - **Repost of an @thesupermannx post:** @marshallpittman reposted "Researchers proved LLMs actually understand things." on **Oct 11, 2026, 12:00 PM CT** ([repost](https://x.com/marshallpittman/status/2109328391853776926); [original](https://x.com/thesupermannx/status/2109267610596684047), Oct 11, 7:59 AM CT). See the [[Users/@thesupermannx/Post Analyses/Post Analysis - LLMs Actually Understand (2026-10-11)|analysis]]: the framework is accurate, but "proved" overstates it. - **Repost in an @thesupermannx thread:** @marshallpittman reposted a reply by @PaulBard73 on **Oct 22, 2026, 8:28 PM CT** ([repost](https://x.com/marshallpittman/status/2113442294766428650)). The thread runs: - **The @thesupermannx post** on Alzheimer's ultrashort waves ([post](https://x.com/thesupermannx/status/2113252527844311276), Oct 22, 7:53 AM CT; 1.42M impressions, 31,939 likes). It is the account's most-liked post. - **@lllaszlo78's reply** ([reply](https://x.com/lllaszlo78/status/2113352025236987921), 2:29 PM CT): "Radiolab had a great episode on this, bring back Gabba. If you have a high refresh rate monitor (hz / 40) you can actually do this at home…" - **@PaulBard73's answer** ([reply](https://x.com/PaulBard73/status/2113423069150855422), 7:11 PM CT), linking the Radiolab episode "Bringing Gamma Back, Again" (Li-Huei Tsai's 40 Hz light work). This is the reply @marshallpittman reposted. - The Alzheimer's ultrashort-wave paper itself was **not** checked in this pass ⚠️. - **Other interactions:** none found (no replies, quotes or mentions between the accounts). ## Assessment for source reliability **Low to moderate as a pointer to papers; low as a summary of them.** - **Useful as a feed:** most AI posts link a real, findable paper, and the numbers quoted are usually taken from the paper. - **Unreliable as a summary:** the headline and conclusions routinely overstate the work, credit the wrong labs, drop caveats, and sometimes add mechanisms that are not in the paper. - **Recommendation:** read the linked paper's abstract before repeating any claim, and treat "proved", "mathematically" and "terrifying" as signs that the framing has outrun the source.