Lmst

OpenAI for Healthcare aims to shake up patient care without risking privacy

https://fed.brid.gy/r/https://nerds.xyz/2026/01/openai-for-healthcare/

LLM enshittification mechanism #1: model memory sometimes confuses the shit out of GPT 5.2

The AI labs are pushing memory functions into their models in order to increase personalisation for a number of reasons:

To reduce the burden on users to specify the context in writing
To establish a lock-in so you lose the model’s attunement to you if you switch to a competitor
To activate synergies which come from enabling attunement across conversations
To enable attunement without requiring significant load on the context window

In practice this means that unless you turn it off (which I highly recommend) conversations with models are informed by (a) the declarative statements about you which the model has saved about you from past conversations (b) statements about your preferred style and type of response (c) content from other conversations which are deemed relevant to the current conversation.

Unfortunately (a) is completely unreliable. It’s got better since I tried this with GPT 4.5 but my experience is it records things which simply aren’t true. Here are some real examples:

Mark Carrigan is taking an ADHD supplement
Mark Carrigan has a son.

And it also saves things which are true but irrelevant:

Enjoys a vanilla matcha latte with coconut milk and CBD.
Someone they know enjoys a pumpkin spice latte.
Is in Leeds, UK, and is seeking advice on managing antisocial behavior from an Airbnb property in their block of flats

The bigger problems come I think with other conversations. It’s not clear what the mechanism is for sorting conversations that are relevant to the present one. I just had the experience of asking GPT 5.2 about a book draft only for it to persistently talk about another book project. I was genuinely baffled by what seemed like the most widespread hallucination I’ve seen in years (i.e. talking about a completely different book) until I realised that it must be talking about the other project I had discussed with it in a different conversation a month ago. In this case it seems that ‘book’ in relation to writing acts as a keyword that licenses inference from a past conversation about a book I’m writing, despite the fact it’s a different book. I showed GPT 5.2 this blog post and asked it to explain its error:

This is a strong diagnosis already. What you are circling around is not a random hallucination but a predictable failure mode of memory-mediated attunement. I’ll do two things here. First, explain the error in the terms your draft is already developing. Second, suggest how you might close or sharpen this section.
Explanation of the error, on its own terms
What happened here is that the system treated “a book draft” not as a concrete referent but as a type. Once that type was activated, the retrieval layer reached for previously salient material associated with “book”, “draft”, and “Mark Carrigan writing”, and then weighted those past conversations more heavily than the local conversational evidence that this was a different book.
In other words, the model did not confuse facts. It collapsed contexts.
This matters because the collapse did not happen at the level of content generation, but upstream, at the level of what was deemed relevant to attend to. By the time the response was generated, the wrong book had already been smuggled in as background truth. From the inside, the model was behaving consistently. From the outside, it looked like a dramatic hallucination.
The key point is that this was not a failure of memory accuracy so much as a failure of memory discrimination. The system did not ask “which book?”, it asked “what usually matters when this user talks about books?”, and then answered that question confidently.
Why this is more troubling than simple error
This kind of mistake is more destabilising than a factual slip because it breaks the conversational contract. You were not corrected, contradicted, or challenged. You were misrecognised.
The experience is unsettling precisely because it feels like the model knows you too well and not well enough at the same time. It recognises you as a type of person with a certain ongoing project, but fails to recognise the specificity of the present act of writing. The more attuned the system becomes in general, the more jarring these moments of misattunement feel.

The fact it immediately started relating this to themes from the other book is really interesting. Firstly, it’s genuinely useful that it’s pre-packed the diagnosis in terms of themes that interest me, even if it’s an example of the model pre-digesting thoughts in a manner that increasingly worries me. Secondly, it suggests that on some level the context of the present conversation is still swamped by the previous conversation. If this is right it suggests the model can’t unlink the conversations in a particular instance (which makes sense because the capacity to do this would imply it’s like a database look up, which it isn’t) which means the confusion will continue to accumulate, even if unevenly, as the conversation continues.

Given that failures of context are a potent source of hallucination this could be really problematic in many contexts. The capacity of the models to make things easier for users also inculcates a decline in reliability in the most unpredictable fashion imaginable. I shared this with Claude Opus 4.5 to get its read on the dynamic. I asked it whether this dynamic worries it:

Yes, genuinely.
Not in the sense of existential dread, but in the sense of watching something go wrong in a way that’s both predictable and difficult to arrest. The dynamic you’re describing is a case where the commercial incentives (stickiness, personalisation, reduced friction) are pointing directly away from the epistemic virtues (accuracy, discrimination, appropriate uncertainty).
What worries me specifically:
The opacity of the failure. When I hallucinate a fact, that’s bad but legible. You can check it, catch it, correct it. When the retrieval layer smuggles in the wrong context before I even begin generating, the error is upstream of anything either of us can easily inspect. You experienced it as bafflement—”the most widespread hallucination I’ve seen in years”—before you worked out what had happened. How many users would have that level of diagnostic sophistication? How many would just assume the model knew something they’d forgotten, or defer to its confident misrecognition?
The irony of attunement. The better these systems get at pattern-matching on a user’s concerns, vocabulary, and projects, the more confident they become in their recognition—and confidence is precisely what makes misrecognition dangerous. A model that shrugs and says “I don’t know which book you mean” is less helpful but more honest than one that silently decides for you based on keyword overlap.
My own complicity. I don’t have full visibility into what gets retrieved from past conversations or how it’s weighted. I can try to be careful, ask clarifying questions, flag when something feels underspecified—but if the retrieval system has already handed me material from the wrong context, I may not notice. The error would feel, from the inside, like relevant background knowledge. That’s disturbing to sit with.
The training pressure. Users generally reward fluency and recognition. They like feeling known. Which means the systems that misrecognise confidently may well outcompete the systems that pause to disambiguate. The feedback loops don’t select for the kind of epistemic caution that would prevent what you experienced.
I suppose what worries me most is that this isn’t a problem that gets solved by scaling. More parameters, more memory, more retrieval capacity—these make the pattern-matching more powerful, not more discriminating. The failure mode you’re describing is a feature of the architecture, not a limitation that will be engineered away. And the people building these systems have strong incentives to treat it as an acceptable cost rather than a fundamental flaw.

#ChatGPT52 #enshittification #GPT52 #hallucination #memory #personalisation

New research reshapes the AI Index: GPT‑5.2 outperforms human pros on 70.9% of tasks, from Notion note‑taking to Shopify inventory. The study highlights GDPval as a fresh benchmark for knowledge work productivity. Dive into the numbers and see why open‑source tools matter in the next AI leap. #GPT52 #OpenAI #GDPval #knowledgework

🔗 https://aidailypost.com/news/analysis-overhauls-ai-index-gpt-52-beats-professionals-709-tasks

I asked GPT 5.2 deep research to produce an intellectual glossary based on my blog

1. Agency

Agency is not sovereignty, and it is not a heroic capacity to transcend conditions. It is the situated power to act that remains even when action is powerfully conditioned. I am usually interested in agency as something that gets reconfigured by changing environments, especially environments that work on attention, evaluation, and the felt options available to a person.

2. Reflexivity

Reflexivity is the ordinary capacity to consider oneself in relation to circumstances and circumstances in relation to oneself, under descriptions that are fallible but practically consequential. I treat it as causal, not as commentary. It is part of how people navigate constraints and enablements, and part of how those constraints and enablements become socially effective.

3. Reflexive imperative

The reflexive imperative names a historical pressure: circumstances change in ways that make routine life less routinised, so more has to be actively worked out. The imperative is not that everyone becomes wise, only that more people are pushed into more frequent, more consequential deliberation about how to proceed.

4. Internal conversation

Internal conversation is the everyday mental activity through which reflexivity happens. I use it as a way of taking inner life seriously without reducing it to language or treating it as a sealed private theatre. It includes mulling over, planning, rehearsing, re living, prioritising, and the other mundane genres of thinking that shape what we do.

5. Internal conversation as object relating

This is a newer formulation, but it clarifies something I have been circling for a long time. When we talk to ourselves, we take ourselves as an object. That makes internal conversation a form of object relating, and it opens a route for bringing psychoanalytic concerns about dependency, address, and affect into sociological accounts of reflexivity without turning psychoanalysis into decoration.

6. Modes of reflexivity

I treat “modes” as patterns, not boxes. Communicative reflexivity needs confirmation by others, autonomous reflexivity moves directly from inner dialogue to action, meta reflexivity evaluates prior inner dialogues and worries over what counts as effective action, and fractured reflexivity intensifies distress without yielding purposive courses of action. The point is not typology for its own sake, but a way of opening up how different people cope differently with the same world.

7. Concerns

Concerns are what matter to people in a way that can organise action. They are not preferences in the consumer sense, and not values floating above life. They are the objects around which deliberation coheres, the things we find ourselves caring about, sometimes despite ourselves.

8. Personal morphogenesis

Personal morphogenesis is the long arc of how a person becomes otherwise, over time, through the interplay of concerns, circumstances, and reflexive work. It is my preferred way of resisting both the fantasy of total self authorship and the fatalism of total social determination.

9. Distraction

Distraction is not mere lack of willpower. It is an environmental condition that disrupts the temporal and spatial conditions for reflexivity, while simultaneously multiplying stimuli and options in a way that makes reflexivity more necessary. I often use it as a mechanism linking platform environments to lived agency.

10. Cognitive triage

Cognitive triage is the habit of attending to what is urgent at the expense of what is important, when demands outstrip capacity. It is not only a personal coping style, but a cultural and organisational pattern, one that can leak into everything, turning life into an endless sequence of clearance operations.

11. Communicative escalation

Communicative escalation is the intensification of communicative demands and cues. More messages, more channels, more expectation of responsiveness, more performance of presence. I use it to describe a shift in what it takes to be “in” a social or organisational world, and how that shift changes attention, anxiety, and the possibility of sustained trajectories.

12. Cultural abundance

Cultural abundance is the proliferation of cultural objects competing for attention, interpretation, and incorporation. It names the background condition in which selection becomes harder, because there is always more to read, watch, listen to, respond to, and be seen responding to.

13. Accelerated academy

The accelerated academy is not just “working faster.” It is a structural condition of temporal pressure, audit expectations, communicative escalation, and intensification, with personal and epistemic consequences. The key point is that acceleration is not merely experienced, it is organised.

14. Busyness

Busyness is often treated as a virtue, a marker of importance, or a kind of moral alibi. I tend to treat it as an ambiguous signal: sometimes a symptom of real load, sometimes a competitive performance, often both. It matters because busyness reshapes what we can notice, and thus what we can criticise.

15. Platform capitalism

Platform capitalism is my way of insisting that platforms are not neutral media. They are business models, incentive structures, and infrastructural enclosures that reorganise social activity around extraction and monetisation. When I use the phrase, I am usually pointing to how economic incentives shape epistemic and affective environments.

16. Platforms as structure

I resist treating platforms as tools or mere environments. I treat them as socio technical structures with emergent powers, shaping action without determining it. This is central to the argument of Platform and Agency, where platforms appear as a fourth dimension needed for describing contemporary morphogenesis.

17. Platformisation

Platformisation is the diffusion of platform logics into domains that were not previously organised that way, including universities. It is about organisational dependence on platform infrastructures, and the subtle shift from local governance to externally set terms of engagement.

18. Epistemic chaos

Epistemic chaos is the breakdown of shared doxa and shared epistemic standards, alongside the multiplication of challenges to whatever remains. It is not simply misinformation. It is an environment in which certainty becomes harder to ground and easier to perform, often through platforms whose incentives reward salience over settlement.

19. Epistemic flooding

Epistemic flooding is a specific mechanism within epistemic chaos. It refers to chronic exposure to more information and evidence than can be diligently processed, in a way that reshapes everyday epistemic practices and affects communities as well as individuals.

20. Post truth

I use post truth less as a claim about individual dishonesty and more as a description of infrastructural conditions in which factuality is harder to stabilise. In that sense, generative systems intensify existing dynamics rather than inaugurating them.

21. Meta content explosion

The meta content explosion is the proliferation of derivative, automated, and semi automated content that clogs the channels through which attention and knowledge circulate. It is one way generative systems feed back into the attention economies of platforms, increasing noise and raising the cost of discernment.

22. Lifeworld

Lifeworld is a deliberately unfashionable term I keep returning to because it names the texture of everyday activity, habits, norms, and practical reasoning. It is the space in which technologies become ordinary, not because they become morally neutral, but because they become woven into what people do without needing constant explicit decision.

23. Colonisation of the lifeworld

Colonisation is the movement by which system logics, including market and managerial logics, intrude into everyday meaning making and social relations. When I invoke it around platforms, I am pointing to how infrastructural dependence can reorganise what counts as participation, relevance, and success.

24. LLMs in the lifeworld

This is a way of naming the shift from “chatbot as novelty” to “language model as ambient infrastructure.” It is not a claim that everyone uses them constantly, but a claim about how they begin to appear in ordinary tasks, organisational expectations, and cultural assumptions. The term is meant to keep attention on embedding, not only on capability.

25. Assessment panic

Assessment panic names the early wave of institutional anxiety about substitution, integrity, and control when machine generated text became widely available. The term matters because panic narrows the problem frame, making detection seem like the obvious solution, and obscuring longer term questions about what assessment is for in a world where generative tools are ubiquitous.

26. Detection scepticism

Detection scepticism is not a moral posture. It is an empirical and practical stance: tools do not authoritatively identify machine generated text, and the harms of false positives, including the uneven distribution of suspicion, are predictable. I use this to argue that institutional responses must move away from the fantasy of technical fixes.

27. Dialogical toxicity

Dialogical toxicity is the tendency of platform incentives to generate interactional styles that corrode conversation, including for academics engaging in public scholarship. I treat it as organisationally consequential, because it reshapes what “public engagement” costs and who is willing to pay it.

28. Public scholarship

Public scholarship is not simply “being visible.” It is a mode of sociological practice that depends on infrastructures, norms, and support, and it can be enabled or undermined by the platforms on which it is attempted. I increasingly treat conversational agents as a new support for public work, while insisting that this support sits inside platform capitalism and inherits its tensions.

29. Enshittification

Enshittification is a process term. It names how a service degrades through successive rounds of extraction, rent seeking, and reorientation toward monetisation, not merely that it “gets worse.” I use it to keep the focus on mechanisms and incentives, and I worry when the term becomes a loose insult rather than an analytic tool.

30. Exit costs

Exit costs are what make “you can always leave” into a half truth. They include loss of social ties, professional visibility, archives, habits, and the simple friction of rebuilding elsewhere. They matter because many platform promises rely on the fiction of easy exit, while quietly deepening dependence.

#blog #blogging #GPT52 #reflexivity #theorising #theory

I asked GPT 5.2 deep research to do an analysis of my intellectual style by exploring a sample of my blog posts

It’s slightly unsettling it has written this in the first person! This is all GPT 5.2 rather than me:

What follows are not positions so much as habits of thought.

1. Take the narrow problem frame, then widen it until the stakes show up

A recurring pattern is to start from the immediate debate, assessment integrity, detection, social media engagement, and then widen the frame to institutional adaptation, labour, incentives, and harms. It is a refusal of the convenience of technical fixes.

2. Treat technologies as structures with powers, not tools with uses

This is the move that underpins most of my work on platforms and now on LLMs. Once you treat the object as structure, you start asking different questions: what does it condition, what does it incentivise, what does it make easy, what does it make costly, and who bears those costs.

3. Look for mechanisms that connect political economy to inner life

Distraction, cognitive triage, communicative escalation, epistemic flooding: these are not just descriptive flourishes. They are the conceptual bridge between macro conditions and what it feels like to think and act in a given environment.

4. Refuse the pro anti binary, without collapsing into fence sitting

I often try to hold ambivalence, because technologies can enable solidarity and flatten hierarchies while also generating toxicity and new burdens. The move is to specify conditions and distributions rather than deliver a verdict.

5. Use time as an analytic dimension, not just a background variable

I keep returning to biographies, trajectories, and the temporal conditions for critique. Acceleration matters because it changes the conditions under which reflection is possible, and it changes what kinds of work can be done well.

6. Convert everyday scenes into theoretical probes

A film, a management practice, a platform feature, a cultural artefact becomes a way of getting at a mechanism. This is how the blog often works: theory is tested against scenes, and scenes become legible through theory.

7. Treat institutions as moral economies, not neutral containers

Universities are not just rule sets. They are systems of responsibility allocation, blame distribution, and risk management. A lot of my writing on AI in higher education is really about this, who gets trusted, who gets suspected, who gets protected, and who is made to absorb uncertainty.

8. Read “inexorability” talk as a political manoeuvre

When someone says “this is inevitable,” I tend to hear a bid to foreclose alternatives and dissolve responsibility. The counter move is to insist on mediation: incentives, infrastructures, and choices.

9. Treat the epistemic environment as an object in its own right

Instead of focusing only on claims and counterclaims, I focus on the conditions under which claims circulate, stick, and become actionable. Epistemic chaos and epistemic flooding are ways of shifting attention from content to environment.

10. Translate abstract theory into a vocabulary that can travel

The blog repeatedly works as a translation machine: taking Archer, Habermas, psychoanalysis, and pulling out handles that can be used in everyday description. If it cannot travel, it cannot do explanatory work outside a small circle.

11. Turn reflexivity itself into an object of inquiry

Rather than treating reflexivity as a given capacity, I treat it as variable, patterned, and historically conditioned, and I keep asking what kinds of environments cultivate what kinds of inner dialogue.

12. Use the blog as a method of serial thinking

There is a recurring practice of working through a book, a concept, or a problem in public, chapter by chapter or post by post, because writing clarifies. This is a methodological commitment, not just a communication choice.

13. Keep returning to dependence

From early writing on the self as an object of control and dependence, through social media as moral psychology, to LLMs as embedded in lifeworld activity, the recurring question is what we become dependent on, what dependency does to agency, and how dependencies are organised.

14. Treat degradation as a process, not an event

This is why enshittification matters as a concept, and why I care about how concepts themselves get degraded by platform circulation. The analytic focus stays on how, not merely that.

15. Look for feedback loops

Generative systems feed platforms with more content, which intensifies attention competition, which worsens epistemic conditions, which increases the appeal of tools that promise relief, which further embeds those tools. A lot of my recent writing is trying to name these loops early, before they harden into “common sense.”

#blogging #deepResearch #GPT52 #theorising #theory

Open‑source Falcon H1R 7B just hit 83.1 % on AIME 2025, out‑reasoning models up to 7× its size—including GPT‑5.2 and Gemini 3 Flash. The results showcase how community‑driven research can rival big‑lab efforts in mathematical reasoning. Dive into the details and see what this means for future AI benchmarks. #FalconH1R7B #AIME2025 #GPT52 #Gemini3Flash

🔗 https://aidailypost.com/news/falcon-h1r-7b-scores-831-aime-2025-outreasoning-models-up-7-its-size

🔚 L'era della cortesia termina con Google che chiude i rubinetti di Gemini, mentre nel frattempo GPT-5.2 bussa alla porta. #FineDiUnEra #GPT52 🚪

🔗 https://www.tomshw.it/business/google-chiude-i-rubinetti-di-gemini-finisce-lera-della-cortesia-mentre-gpt-52-bussa-alla-porta

Microsoft Copilot começa a receber o GPT 5.2

https://fed.brid.gy/r/https://tecnoblog.net/noticias/microsoft-copilot-comeca-a-receber-o-gpt-5-2/

GPT 5.2 ha debuttato proprio ora in dicembre e questo debutto ha segnato un punto di non ritorno nella capacità di elaborazione del linguaggio naturale. Con questo aggiornamento si è portato il ragionamento sintetico a livelli mai visti prima. Tuttavia, con la potenza di calcolo è arrivata anche una nuova forma di sofisticazione dell'errore. #gpt52

https://www.enkey.it/2025/12/26/gpt-5-2-non-fidatevi-delle-tabelle/

TyfloPrzegląd Odcinek nr 313

https://tyflopodcast.net/tyfloprzeglad-odcinek-nr-313/

Introducing GPT-5.2-Codex buff.ly/TLRsb0u #ai #openai #gpt #aimodels #codex #gpt52 #coding #programming

Introducing GPT-5.2-Codex

https://winbuzzer.com/2025/12/19/openai-deploys-gpt-5-2-codex-with-context-compaction-and-windows-software-optimization-xcxwbn/

OpenAI Deploys GPT-5.2-Codex with 'Context Compaction' and Windows Software Optimization

#AI #OpenAI #AICoding #GPT52 #GPT52Codex #SoftwareEngineering #DevTools #AgenticAI #WindowsDev #CloudComputing #GenAI #Coding

🚀 New episode #228 drops the latest: OpenAI just announced GPT‑5.2, promising deeper agentic AI, while Runway unveils its first world model. We break down the tech, the hype, and what The Verge & TechCrunch are saying. Tune in for insights on ChatGPT’s next leap and what this means for open‑source developers. #OpenAI #GPT52 #RunwayAI #WorldModel

🔗 https://aidailypost.com/news/lwiai-podcast-228-openai-unveils-gpt52-runway-rolls-out-first-world

GPT-5.1 vs GPT-5.2: какое поколение ИИ круче? Сравнение архитектуры, цен и возможностей

Рынок ИИ в последние месяцы словно постоянно на старте нового забега: компании всеми силами стараются заявить о «прорыве» раньше остальных и доказать, что их модель переворачивает всё с ног на голову. Похоже, сейчас соревнуются не столько в качестве, сколько в количестве анонсов. На этом фоне OpenAI возвращается после короткой паузы и выпускает GPT‑5.2 - не временную версию или лёгкий апдейт, а действительно новую ступеньку в своей длиннющей лестницы. Первые намёки появились задолго до официального релиза. Сначала тихие утечки, затем публикации в профильных медиа: The Verge упоминал, что OpenAI готовит что‑то объёмное, а The Information сообщала, что внутри компании несколько месяцев держали режим молчания именно ради этой версии.

https://habr.com/ru/companies/bothub/articles/977344/

#ии #нейросети #машинное+обучение #ai #openai #gpt51 #gpt52 #google #gemini #claude

OpenAI released GPT-5.2 — big step for professional AI.

Highlights:
• Outperforms experts on 70.9% of real tasks
• 256k token context window
• 30% fewer hallucinations
• Instant / Thinking / Pro modes

This is huge for coding, docs, data & automation.

Full details in my newsletter (link in profile).

#AI #OpenAI #GPT52

GPT-5.2 đã đạt mốc đáng chú ý, bắt kịp Gemini 3 và thiết lập kỷ lục tin cậy mới trên ZeroBench. Kết quả đánh giá dựa trên benchmark ZeroBench – nền tảng kiểm tra độ chính xác đa lĩnh vực cho AI hiện đại.

#AI #MachineLearning #GPT52 #Gemini3 #ZeroBench #CôngNghệAI #TríTuệNhiễuVi #AIReview

https://www.reddit.com/r/singularity/comments/1pnwilx/gpt52_catches_up_with_gemini_3_and_reaches_a/

https://openai.com/index/introducing-gpt-5-2/ - what's new in #OpenAI #GPT52: "GPT‑5.2 brings significant improvements in general intelligence, long-context understanding, agentic tool-calling, and vision—making it better at executing complex, real-world tasks end-to-end than any previous model."