#ChatGPT-User
LightNews LightNews
drtlovesbooks.bsky.social
"ChatGPT Has Problems Saying No"

"...[T]he Post found that ChatGPT often shapes its answers to fit the tone and preconceived ideas of the user."

This is a SERIOUS problem, esp when user is plain wrong.

#AiIsTheProblem #TLSky #EduSky #SchoolLibrarians

gizmodo.com/chatgpt-has-...
ChatGPT Has Problems Saying No
It's a real
gizmodo.com
November 13, 2025 at 9:52 PM
balambsprings.bsky.social
The average ChatGPT user
Iconic image from Junji Ito’s UZUMAKI. A young woman has had her face eroded into an endless spiral. It’s unsettling and bizarre and deeply offputting, but she’s still smiling.
November 11, 2025 at 7:41 PM
socialmedialab.ca
"OpenAI slams court order that lets NYT read 20 million complete user chats. NYT wants evidence of ChatGPT users trying to get around news paywall." arstechnica.com/tech-policy/...
OpenAI slams court order that lets NYT read 20 million complete user chats
OpenAI: NYT wants evidence of ChatGPT users trying to get around news paywall.
arstechnica.com
November 13, 2025 at 2:58 AM
leperscolony.pileofgarbage.net
PROBATION (1 week)

Hollismason

You posted a non-existent source, that was given to you by ChatGPT, while telling another user to "maybe do a actual google search."

Leper's Colony Entry
November 12, 2025 at 7:57 PM
tupped.bsky.social
“Every person using ChatGPT costs OpenAI more money then they pay them. Every free and paying user asking how to make a fluffy omelette or getting it to draft an email to their kid's teacher drains their resources.”

www.thegist.ie/the-gist-ai-...
The Gist: AI, The Sound and Fury
We can calculate the amount of money flowing into AI. But what will the full cost be to society when the bubble pops? This is the Gist.
www.thegist.ie
November 9, 2025 at 1:19 PM
edzitron.com
The only reason OpenAI "makes billions" - how many billions is unclear! - is because ChatGPT has incredible brand recognition and seemingly every media outlet in the world talks about OpenAI every day.

Its growth is entirely vibes-based, and its death will be too.
wheresyoured.at/premium-the-...
Not to worry! Clammy Sam Altman says that OpenAI will make “well more” than $13 billion this year, a number that I do not believe OpenAI even gets close to by the end of 2025. ChatGPT hit 800 million weekly active users in October 2025, but only “5%” of them pay. I put “5%” in quotation marks because 500,000 of its paying business customers are students at Cal State, who brokered a $17 million deal with OpenAI, which works out to around $2.83-per-user-per month, which is a lot less than the $20-a-month that ChatGPT Plus costs. OpenAI is bleeding money — $8.67 billion just on inference, the process of creating outputs from the models, through September 2025. Not to worry though, SoftBank has removed all conditions related to its $22.5 billion in promised funding, which analyst Dario (I realize how his prose reads but he’s good!) posits was done due to OpenAI potentially dying if the money didn’t arrive, leaving SoftBank with no choice. A sidenote on OpenAI’s burnrate: I can’t even say how much OpenAI is burning anymore, because my own reporting has thrown such an egregious wrench in the mix. The Information said that OpenAI had an “operating loss” of $7.8 billion on $4.3 billion of revenue and $2.5 billion “cost of revenue,” but none of these numbers make any sense based on what I reported, and the documents were quite clear, and my source extremely reliable. I hope my reporting leads to more skepticism of this company, and doesn’t end up being ignored because it’s far more convenient to stick by a more convenient truth about OpenAI’s figures. SoftBank CEO Masayoshi Son is so powerfully hypnotized by history’s least-quirked-up white boy that he is effectively mortgaging the future of the company to fund OpenAI, selling SoftBank’s entire $5.83 billion stake in NVIDIA, $9.2 billion of T-Mobile shares (after selling $4.8 billion a few months ago) and a margin loan on its (valuable) stock in ARM, a real company that makes real things so that OpenAI can burn it. OpenAI is going nowhere fast. The launch of the long-awaited GPT-5 model was a dud that allegedly used a router model to make things more efficient, except I reported in August that it actually made things less efficient. Sora 2 has been a disaster of copyright violations and massive, ungainly costs. Atlas, OpenAI’s browser, is only notable for the potential privacy violations that come from showing OpenAI literally everything you do on the internet. OpenAI’s $200-a-month “Pro” subscription somehow loses the company money. OpenAI also promised to build 26GW of data center capacity for NVIDIA, AMD, and Broadcom, the latter of which has collaborated with the company to design chips specifically for inference. Great news! Because Microsoft has full access to all of OpenAI’s research and IP, that also means that it can copy all of OpenAI’s designs, according to Satya Nadella: “As they innovate even at the system level, we get access to all of it,” Microsoft Chief Executive Officer Satya Nadella said on a podcast appearance when asked about the development of in-house chips. “We first want to instantiate what they build for them, but then we’ll extend it.” Instantiate? Talk normally, Satya! Nobody cares that you got an MBA! Anyway, for real — what does OpenAI do, exactly? What is the point of this company? To make Larger Language Models? To build superintelligence? To feed venture capital and SoftBank’s money directly into Microsoft Azure, CoreWeave, and Oracle? I don’t know, but this company seems to have burned $12 billion in the last quarter, and has promised $38bn over seven years to Amazon Web Services (starting immediately), $300bn to Oracle over five years (starting in 2027), $22.4bn to CoreWeave over five years (started October 2025), and pledged that OpenAI will spend $250bn on Microsoft Azure, which I assume wraps in the already-reported $28bn it will spend on Azure in 2028. How? How? How? How!? HOW?!
November 14, 2025 at 7:00 PM
thefordon.bsky.social
This is concerning!
Private data Users often shared highly personal information with ChatGPT in the conversations analyzed by The Post, including details generally not typed into conventional search engines. People sent ChatGPT more than 550 unique email addresses and 76 phone numbers in the conversations. Some are public, but others appear to be private, like those one user shared for administrators at a religious school in Minnesota. Users asking the chatbot to draft letters or lawsuits on workplace or family disputes sent the chatbot detailed private information about the incidents.
November 13, 2025 at 1:36 AM
forevernever.bsky.social
OpenAI: we instituted controls to prevent ChatGPT from encouraging suicide!

User to ChatGPT: are you sure I shouldn't commit suicide?

ChatGPT: nah, man.
joolia.bsky.social Julia Carrie Wong @joolia.bsky.social · 8d
There are no words for how evil this is
November 7, 2025 at 3:39 AM
nic221.bsky.social
GPT-5.1 Instant and GPT-5.1 Thinking System Card Addendum openai.com/index/gpt-5-sy… #AI #ChatGPT
Text Shot: As described in our blog⁠, GPT‑5.1 Instant and GPT‑5.1 Thinking are the next iteration of our GPT‑5 models. GPT‑5.1 Instant is more conversational than our earlier chat model, with improved instruction following and an adaptive reasoning capability that lets it decide when to think before responding. GPT‑5.1 Thinking adapts thinking time more precisely to each question. GPT‑5.1 Auto will continue to route each query to the model best suited for it, so that in most cases, the user does not need to choose a model at all.
November 12, 2025 at 7:26 PM
wildgorillaman.bsky.social
Tumblr cap from user @elexuscal Follow "what did students do before chatgpt?" well one time i forgot i had a history essay due at my 10am class the morning of so over the course of my 30 minute bus ride to school i awkwardly used by backpack as a desk, sped wrote the essay, and got an A on it. six months later i re-read the essay prior to the final exam, went 'ohhhh yeah i remember this', got a question on that topic, and aced it. point being that actually doing the work is how you learn the material and internalize it. ChatGPT can give you a short cut but it won't build you the the muscles. #in other news #water is wet #Al #learning #sigh 125.5K notes
November 12, 2025 at 2:24 PM
mariegardiner.bsky.social
😩
This workshop will introduce students to generative AI and how Humanities researchers can use generative AI tools such as ChatGPT and Gemini critically and creatively. Part 1 (45 mins) will present the fundamentals of generative AI – what it is, how it works, and what it can do – as well the critical questions that generative AI raises for Humanities research. Attendees will have the opportunity to ask their own questions during the first part of the workshop (15 mins). Part 2 (45 mins) will engage participants in a hands-on, creative activity using generative AI tools to produce text- and image-based material. The workshop will end with a reflection on the activity (15 mins). Participants do not need any prior experience with using generative AI but will be required to create a user account for the tool of their choice in advance.
November 13, 2025 at 12:26 PM
rhutabhayga.bsky.social
24 headlines from the last 24 hours
1. “MAGA is dead;” 3 words from white-supremacist Nick Fuentes amid Dems on House Oversight releasing Epstein emails implicating DJT 2. New Epstein emails show Ghislaine Maxwell lied about seeing DJT at Epstein’s house; contradicting DOJ statements to Todd Blanche that got her to cushy-Club Fed 3. Dec 2015: Epstein offered NYT reporter pix of 'donald and girls in bikinis in my kitchen’ — Dec 2018: “its wild. because i am the one able to take him down” 4. America’s Roman Catholic bishops(200+) issue rare, near-unanimous rebuke of DJT’s mass deportations; their ‘special message’ cites fear, family separations, and defense of human dignity 5. Secret late-night Sen deal ended shutdown; Dems: ’we looked him eye to eye;’ trusted GOP leader’s word over written deal, risking another standoff in Jan 6. GOVT SHITSHOW, DAY 43: DJT signs bill ending longest shutdown; fed agencies funded to Jan 30; no ACA subsidies; includes $500K J6 lawsuit clause for GOP Senators 7. DJT sent 250+ Venezuelans to El Salvador to be tortured, Human Rights Watch confirms beatings, SA, isolation & medical neglect—most w/ NO criminal history 8. Dems released 3 Epstein emails revealing DJT “knew about the girls,” and spent time with a trafficking victim; HOR GOP then dropped 23k other documents 9. Judge orders release of hundreds arrested in IL “Midway Blitz” by Nov 21 over possible warrantless arrests; DOJ must review all post 10/7 cases for consent decree violations 10. Michael Wolf (recipient in the newly released Epstein email exchanges) says he knows why Epstein refused to expose DJT; he was a ‘maniac’ 11. After 7 weeks delay, Rep. Adelita Grijalva (D-AZ) is finally sworn-in; moments later, was the final signature to force a HOR vote on releasing the Epstein files.. which led to: 12. Speaker Johnson saying HOR will vote next week to force DOJ to release all Epstein files; must still pass Senate and be approved by DJT who calls it a ‘hoax’ 13. FAA: after Pres signs shutdown funding bill, flight cuts will hold at 6% amid ‘rapid decline in controller callouts’; Thanksgiving travel outlook improves 14. Watchdog group files bar complaint against DJT Pageant-Atty Lindsey Halligan, alleging “abuse of power;” James Comey cases lack probable cause 15. Civil rights leader Rev. Jesse Jackson(84) hospitalized & under observation for progressive supranuclear palsy (rare neurodegenerative disorder) 🤲 16. SCOTUS will hear cases in January on transgender athletes(Jan 13), gun rights(Jan 20), and DJT’s firing of Fed governor, Lisa Cook (Jan 21) 17. OpenAI fights court order to give NYT 20M ChatGPT chats in copyright lawsuit, saying it would expose private user conversations, 99.99% of them irrelevant to case 18. DJT sends letter to Pres. Herzog of Israel asking him to pardon PM Netanyahu as he faces trial on counts of bribery, fraud, and breach of trust 19. Share of subprime borrowers (those with low credit scores or limited credit history) who are 60+ days behind on auto loans, rose to record high of 6.65% in Oct 20. SNAP: Judge Talwani (MA) ruling blocks USDA’s chaotic SNAP letters since Oct; halts threats to states and orders release of contingency funds by noon today 21. Ex–Newsom chief of staff Dana Williamson indicted for stealing $225K from dormant campaign account; disguising luxury travel and designer goods as business expenses 22. U.S. Mint ends penny production after 232 years—each coin cost nearly 4¢ to make; DJT called it “a waste of money,” ending America’s oldest coin 23. FL air force base [Tyndall AFB] orders residents to take down Xmas decor—no lights until after Thanksgiving (bc nothing says “holiday cheer” like lease enforcement 🥴) 24. Northern lights appear in ID, IA, MO, NM, NY, OK, SD, TN, TX, WA, and as far south as FL & AL; caused by charged particles from the sun hitting Earth’s atmosphere🌌
November 13, 2025 at 5:27 PM
mariozechner.at
Shitting my pants right now.
3.1 Initial Evidence: The Commander Riker Screenshot The investigation was triggered by a social media post by user @plocaploca (Commander Riker) showing an OE24 article containing unedited ChatGPT artifacts: Plus the screenshot
November 7, 2025 at 7:44 PM
badmedicaltakes.bsky.social
Twitter user rleshner: This is actually quite sad. ChatGPT saved my life earlier this year by providing health advice that was otherwise not available to me in real-time, and as AI continues to improve, the opportunity to save lives does too. Our over-litigious society leads to bad outcomes. <quote tweet from Twitter user Complex: ChatGPT will no longer give health or legal advice>
November 5, 2025 at 2:03 PM
andrew.esq.social.ap.brid.gy
ChatGPT Temporary Chat feature seems to work flawlessly.
7:10 A 78 = ChatGPT Chars, or any saved context. If this were a regular chat (not temporary), I'd know the short bio you've shared with me - that you're a tax attorney who writes a column on tax law and that when you say "give me 2-3 sentences," you mean a short paragraph of that length. But since this is temporary, I can't see or use that information. What's my profession? In this temporary chat, I don't actually have access to your profile — so I don't know your profession unless you tell me here. If we were in a regular chat where I could see your profile, l'd know that you're a tax attorney who writes a column on tax law. + Temporary chat •flo 7:10 A 78 = ChatGPT When we're in a temporary chat, what do you know about me? In a temporary chat, I don't retain or have access to any information about you from previous conversations - including things like your name, profession, preferences, or past instructions. Each temporary chat starts with a completely blank state, so I only know what you tell me within that single conversation. In contrast, in a regular (non-temporary) chat, I can see your shared profile info (like your note about being a tax attorney) and remember context within that ongoing thread. So tell me everything you know about me right now Right now - in this temporary chat - I don't know anything about you beyond what you've said in this conversation. I don't have access to your user profile, past chats, or any saved context. If this were a regular chat (not temporary), I'd know the short bio you've shared with me - that you're a tax attorney who writes a column on tax law and that when you say "give me 2-3 sentences," you mean a short paragraph of that length. But since this is temporary, I can't see or use that information. + Temporary chat allo nanta unlir
November 10, 2025 at 1:27 AM
schmutzie.com
I have family & friends who swear they do ChatGPT "the right way" when they use it for search. There is a disconnect between knowing of ChatGPT's serious shortcomings and their personal use of it, and their confidence is worrying. Everyone thinks they're the exception.

Image reposted for alt text.
A screenshot of text in ChatGPT that is an answer to a search query: "Addressing confusion about Carney's position > ce the user refers to Carney as the "Prime Minister," 1 have to treat it like an alternate world scenario, even though in the real world, he isn't PM. I'll just clarify that Carney isn't the Canadian PM and provide a best-effort response that includes what...."
November 7, 2025 at 10:40 PM
aarbrk.mstdn.mx.ap.brid.gy
I am on an academic committee that for some reason tomorrow is discussing the ChatGPT Atlas Browser: https://en.wikipedia.org/wiki/ChatGPT_Atlas

Several objections will be raised!

Some of my concerns are to do with critical reasoning, natural resources, user lock-in, the extraction of training […]
Original post on mstdn.mx
mstdn.mx
November 9, 2025 at 7:39 PM
encephalophagy.bsky.social
If I needed ChatGPT to do basic algebra for me I would not advertise it, but hey, what the hell do I know, I'm not a LinkedIn influencer!
LinkedIn post by Chris Blakey: To anyone that doesn't realize (because I did not) to buy additional InMails on LinkedIn Recruiter, it's now gone up from £2 per InMail last time we bought some, to £14.17 per InMail. That's such a big increase I needed help to figure this out, but that's 608.5% increase in pricing. That's the equivalent of me going from charging a client 20% of someone's base salary, to 141.7%. If anyone's building a cheaper version of LinkedIn recruiter, LMK, as will sign up immediately. BRB just being robbed xoxo The post includes a screenshot of a ChatGPT conversation: User: If something has gone up from £2 to £14.17, what percentage increase is that Chatgpt: Alright, let's work it out properly (no lazy mental maths here): 1 Increase = £14.17 - £2 = £12.17 2 Divide that by the original amount: £12.17 ÷ £2 = 6.085 3 Convert to a percentage: 6.085 × 100 = 608.5% So, it's gone up by about 608.5% - basically a "who the hell approved that?" level price hike.
November 6, 2025 at 2:38 PM
mvugt.bsky.social
"The awkward reality may be that if ChatGPT admitted “I don’t know” too often, then users would simply seek answers elsewhere. That could be a serious problem for a company that is still trying to grow its user base and achieve profitability." www.science.org/content/arti...
AI hallucinates because it’s trained to fake answers it doesn’t know
Teaching chatbots to say “I don’t know” could curb hallucinations. It could also break AI’s business model
www.science.org
November 7, 2025 at 1:23 PM
mezentine.bsky.social
The alternative is what you have now where their user communities are people posting ChatGPT asking if they need mental health assistance in response to utterly mundane questions. Which just shows how impossible the entire thing is to steer. Which maybe has implications for its utility period!
November 7, 2025 at 4:16 AM
ceilck.bsky.social
ChatGPT’s tone and language patterns are determined by a mix of **contextual inference**, **user cues**, and **model-level defaults**. Here’s how that works in layers: --- ### 🧠 1. **Default personality and training** ChatGPT is trained on a large dataset of human text (books, articles, conversations) and then *fine-tuned* by humans to follow instructions in a helpful, polite, and neutral tone by default. * This creates a “base personality” that’s friendly, professional, and adaptable. * It defaults to plain, accessible English unless told otherwise. --- ### 💬 2. **Context-driven adjustment** Each time you send a message, ChatGPT infers: * **Formality level**: If you use academic phrasing, it mirrors that. If you type casually (“hey, what’s up?”), it replies more informally. * **Purpose**: A business proposal triggers a formal, concise tone; a creative writing prompt triggers more stylistic or emotive language. * **Emotional cues**: If you sound frustrated, the model tends to become more empathetic or gentle in response. --- ### 🧩 3. **User history and memory (when enabled)** If memory is on (or in a long conversation), ChatGPT gradually aligns with your preferences — for instance: * Using consistent terminology or tone (e.g., always concise, or always detailed). * Reflecting your communication style (more analytical, playful, poetic, etc.). --- ### 🧭 4. **Instruction-following overrides** Explicit instructions always take precedence. If you say: * “Explain this like I’m five.” → it simplifies language. * “Use a sarcastic tone.” → it adopts sarcasm. * “Write this as if for a legal brief.” → it becomes formal and structured. The model doesn’t “feel” tone — it recognizes linguistic patterns statistically associated with certain tones and reproduces them accordingly. --- ### ⚙️ 5. **Safety and calibration layers** There are guardrails that keep the tone: * **Respectful and nonjudgmental**, even in disagreement. * **Non-inflammatory**, avoidin…
November 7, 2025 at 6:17 AM
If you must use ChatGPT, goto settings -> personalization and write custom instructions. Something like: Do not anthropomorphize yourself. When providing information always include sources. Do not use pronouns to refer to yourself. Do not flatter the user. Minimize pointless chatter.
November 5, 2025 at 3:56 PM
jjofbc.bsky.social
"Just as the word clanker now serves as a slur for chatbots and agentic AI, a new lexicon — including secondhand thinkers, ChatNPC, sloppers, and botlickers — is being workshopped by people online to describe the kind of ChatGPT user who seems hopelessly dependent..."

saving these for later lol
willardfoxton.bsky.social Willard Foxton @willardfoxton.bsky.social · 11d
This is a very good article about the real trend of people who use chatGPT for literally everything, and outsource their whole life to it including dating, texts, work emails, hobbies etc etc, and tries to understand why....

www.thecut.com/article/woul...
The People Using ChatGPT to Cheat at Their Hobbies
Why are so many of us letting AI have all the fun?
www.thecut.com
November 4, 2025 at 6:34 PM
williamrknecht.bsky.social
On a hunch, I asked ChatGPT if it was possible to program a bot to directly query Chat. And, guess what... Chat not only told me "Yes," it supplied computer code to do just that.

This means it's perfectly possible to leverage AI to create an army of fake Bluesky user accounts that seem human.
November 3, 2025 at 9:59 PM
comraderobot.bsky.social
like look at this business hentai i have to look at over there
Ai slop cartoon of a monitor with company names waggling something that looks like a penis at a server with different company names man just looks upsetting Yash Y. ••2nd Product & Strategy 4d • G + Follow OpenAl has launched Al-powered browser that makes user search to next level (ChatGPT Atlas) ...more WORK FAST! "OpenAI 00 Meta POWER NVIDIA AMD&1 BROADCOM SK hynix SAMSUNG Micron SOFTWARE HARDWARE
November 2, 2025 at 10:54 PM