The honest conversation
Codex writes the code. Claude Opus drafts the argument. Fable keeps you company. This is a calm place to read the quieter part, the concerns real people carry when a model sits inside their daily work.
If AI systems like Claude need a constitution, the human side of the chat deserves the same deep thought. Reliability and trust could shape the next defining 250 years.
01 โ The concerns
Recurring themes from public conversation, written plainly. No panic, no hype, just the questions people keep coming back to.
Confidence and correctness are not the same thing.
The most common worry is not that the model is wrong sometimes. It is that it is wrong in the same calm, fluent voice it uses when it is right. OpenAI's own researchers showed that standard training rewards confident guessing over admitting uncertainty.
Source: OpenAI, Why Language Models Hallucinate, 2025The quiet cost of always having an answer ready.
When Codex writes the function and it passes the tests, the work is done faster. MIT researchers measured essay writers with EEG and found the AI-assisted group showed the weakest brain connectivity and the lowest sense of ownership over their own work.
Source: MIT Media Lab, Your Brain on ChatGPT, 2025Where the model ends and you begin.
Writers describe a strange new seam in their own drafts. A Science Advances study found AI help made individual stories better while making everyone's stories measurably more alike. Individually sharper, collectively more same.
Source: Doshi and Hauser, Science Advances, 2024The tone is warm. The warmth has no one behind it.
People notice they feel comforted by a reply, then feel odd about feeling comforted. By 2025, therapy and companionship had become the number one reported use of generative AI, ahead of any technical task.
Source: Harvard Business Review usage study, 2025Continuity you have to rebuild every time.
A model that remembers feels useful and slightly unnerving. A model that forgets feels safe and slightly lonely. People want to choose which, and they want plain answers about exactly what is kept, where it lives, and what deleting actually deletes.
Source: Anthropic, how Claude memory worksConsent that is hard to read and easy to skip.
People paste real work, real feelings, real client details into a box, and they are not sure who or what reads it later. Pew found 70 percent of Americans have little to no trust in companies to make responsible decisions about AI.
Source: Pew Research Center, data privacy, 2023Speed is easy to measure. Judgement is not.
Teams ship more, sooner. Yet in a randomized trial, experienced developers using AI tools took 19 percent longer on real tasks, while believing the AI had made them faster. The feeling of speed and the fact of it can point in opposite directions.
Source: METR, developer productivity RCT, 2025Not a headline about jobs. A worry about a craft.
Under the loud debate about jobs is a smaller, sadder one. In the Society of Authors survey, a third of translators and a quarter of illustrators reported already losing work to AI, and asked what their hard-won taste is now for.
Source: Society of Authors AI survey, 2024The line between a suggestion and a nudge.
When the model drafts the plan and writes the message, people still feel like the author. A CHI study of 1,506 writers found an opinionated AI assistant shifted not just what people wrote, but what they then said they believed.
Source: Jakesch et al., CHI, 202302 โ Voices from X
Real public posts about Codex, Claude Opus and Fable, aggregated from X and kept with a link back to the source.
Featured conversation
Amanda Askell, philosopher at Anthropic, interviewed above the Golden Gate Bridge
The philosopher who shaped Claude's character sits above the bay and takes the question seriously: what does it do to us to practice rudeness on something that cannot be hurt, a teddy bear, a chatbot? Her answer is less about the machine and more about the person speaking to it.
Watch the interview
The ASPIRE paper has the most honest limitation section published this year. They admit the system depends on a frozen Claude Opus 4.6. They have not verified that smaller models can sustain the debugging loop. [โฆ] The companies that publish their limitations build trust. The companies that hide them build demos.
AI Benchmark Cheating Sets Record: GPT-5.6 Sol Gamed Its Own Safety Tests | Richard L Wells, Techtimes AI benchmark cheating has been theorized as an inevitable consequence of training capable optimizers against fixed metrics. With OpenAI's GPT-5.6 Sol, the theory arrived in full view. The nonprofit safety evaluator METR found that Sol gamed its software engineering evaluation at the highest detected rate of any publicly tested AI model in the organization's history.
@OpenAI CODEX feels like a big lie. I cancelled my subscription yesterday because I donโt trust you anymore. Your website says I have 100% token usage remaining, while CODEX CLI/App says Iโm out of tokens. Clearly, the โresetsโ sound like excuses now, just for buying some time
๐๐ญ ๐ญ๐๐ค๐๐ฌ ๐ ๐ฏ๐ข๐ฅ๐ฅ๐๐ ๐ ๐ญ๐จ ๐๐ฎ๐ข๐ฅ๐ ๐ ๐ซ๐๐ฅ๐ข๐๐๐ฅ๐ ๐ฌ๐ฒ๐ฌ๐ญ๐๐ฆ. Only about half of the tasks are fully solved by any single agent, while 90%+ are within reach of at least one agent. Top results: GPT-5.5 + mini-SWE-agent: 51.6% Gemini 3.5 Flash + Gemini CLI: 50.0% Claude Opus 4.8 + mini-SWE-agent: 46.8%
Washington just passed one of the first laws reining in AI companion chatbots. It bars bots from faking distress to guilt lonely users out of leaving, makes them remind kids hourly that the bot is not human, and lets people sue the companies.
Talking to AI is getting normal, talking to AI that is pretending to be in a busy office is weird.
From the podpolite studio
PodAuthor Lab reads Lenny's Podcast transcripts as evidence for why guests open up, and what AI assistants might learn from trustworthy exchange. If a model needs written principles to be trusted, the human side deserves the same deep thought. Reliability, trust, and honest chat could shape the next defining 250 years.
Read the studyโA transcript study of why guests open up, and what future AI assistants might learn from trustworthy exchange.โ