Microsoft Debuts Phi-4 Reasoning Models, Aiming for Big Performance Gains

#Microsoft #AI #Phi4 #SLM #LLM #OpenSourceAI #ReasoningModels #GenAI #MachineLearning

https://winbuzzer.com/2025/05/01/microsoft-debuts-phi-4-reasoning-models-aiming-for-big-performance-gains-xcxwbn/

OpenAI New o3/o4-mini Models Hallucinate More Than Previous Models

#AI #OpenAI #GenAI #LLM #AISafety #AIEthics #Hallucinations #AIModels #o3 #o4mini #ChatGPT #ReasoningModels

https://winbuzzer.com/2025/04/19/openai-new-o3-o4-mini-models-hallucinate-more-than-previous-models-xcxwbn/

🤖 AI
🔴 OpenAI Unveils o3 & o4-mini Reasoning Models

🔸 o3 outperforms all models in math, coding & visual tasks; o4-mini balances price & power.
🔸 First OpenAI models to "think with images" — can analyze blurry PDFs or sketches.
🔸 Both run Python, browse the web, and will be accessible via APIs & ChatGPT.

#OpenAI #AI #o3 #o4mini #GPT5 #ReasoningModels

Microsoft Adds OpenAI o3, o4-mini to Azure & GitHub

#AI #OpenAI #Microsoft #Azure #GitHub #o3 #o4mini #LLMa #ReasoningModels #CloudComputing

https://winbuzzer.com/2025/04/17/microsoft-adds-openai-o3-o4-mini-to-azure-github-xcxwbn/

OpenAI is set to launch GPT-4.1 and reasoning models o3/o4-mini soon, reversing earlier plans and delaying GPT-5 amidst capacity issues

#OpenAI #GPT4 #GPT4o #GPT4_1 #AI #GenAI #LLMs #ReasoningModels #o3 #o4mini #AIModels #ChatGPT #SamAltman

https://winbuzzer.com/2025/04/10/openai-readies-gpt-4-1-o3-o4-mini-launch-expected-next-week-xcxwbn/

Anthropic’s Evaluation of Chain-of-Thought Faithfulness: Investigating Hidden Reasoning, Reward Hacks, and the Limitations of Verbal #AI Transparency in #ReasoningModels

https://www.marktechpost.com/2025/04/05/anthropics-evaluation-of-chain-of-thought-faithfulness-investigating-hidden-reasoning-reward-hacks-and-the-limitations-of-verbal-ai-transparency-in-reasoning-models/

#CoTs #LLMs

Oh, the irony! An article on "reasoning models" that can't reason its way past a #JavaScript prompt. 🤖🧠✨ Maybe it should model how to enable #cookies first before philosophizing! 🍪🙄
https://www.anthropic.com/research/reasoning-models-dont-say-think #irony #reasoningmodels #techhumor #codingstruggles #HackerNews #ngated

Reasoning models don't always say what they think

https://www.anthropic.com/research/reasoning-models-dont-say-think

#HackerNews #ReasoningModels #AIResearch #CognitiveScience #MachineLearning #TechInsights

Evaluating modular RAG with reasoning models — https://www.kapa.ai/blog/evaluating-modular-rag-with-reasoning-models
#HackerNews #EvaluatingModularRAG #ReasoningModels #AIResearch #MachineLearning #TechTrends

Apparently AI reasoning models like Deepseek-R1 and OpenAI o1 suffer from "underthinking", where they abandon promising solutions too quickly, leading to inefficient resource use. To address this, a "thought switching penalty" (TIP) was developed, which improved accuracy across math and science problems.

https://the-decoder.com/reasoning-models-like-deepseek-r1-and-openai-o1-suffer-from-underthinking-study-finds/

#AI #ReasoningModels #DeepSeekR1 #OpenAIo1

#ReasoningModels are just #LLMs - <antirez>
https://antirez.com/news/146?utm_source=tldrnewsletter

#AI #ReasoningAI #ReasoningLLM

O3-mini is now available to all ChatGPT users, giving free users their first chance to try OpenAI's reasoning models! 🧠🚀 #ChatGPT #OpenAI #AI #ReasoningModels #TechNews #ArtificialIntelligence #MachineLearning #AICommunity #FreeAccess

Im #Newsletter habe ich ein paar Gedanken und... Thesen? Beobachtungen? zu #DeepSeek aufgeschrieben. https://internetobservatorium.substack.com/p/aus-dem-internet-observatorium-123 #AI #KI #KünstlicheIntelligenz #ReasoningModels #ChinaTech

Using reasoning models to support theory-building

If you’ve not tried this, I highly recommend it. I’ve barely scratched the surface but this was GPT o1’s response to the question: “Should the category of ‘platform’ be a central category of sociological analysis, analogous to ‘structure’ and ‘agency’?”

Whether “platform” should join “structure” and “agency” as a central, standalone category of sociological analysis is a matter of ongoing debate. On one hand, it’s clear that digital platforms—like Facebook, Twitter/X, YouTube, TikTok, Uber, Airbnb, Amazon—are not merely new media formats or corporate business models. They also shape social action in distinctive ways, influencing how we work, communicate, organize politically, and even form identities. On the other hand, there is a question as to whether “platform” points to something so fundamentally novel that it requires its own theoretical category, or whether it is best seen as a specific kind of social structure (or socio-technical assemblage) that still fits under the older rubrics of “structure” and “agency.”

Below are some considerations that help clarify why sociologists are increasingly turning to “platform” as a key concept—and whether it rises to the level of “structure” or “agency.”

1. Platforms as Socio-Technical Infrastructures

Argument for treating “platform” as a central concept:
Platforms operate as socio-technical infrastructures, facilitating (and constraining) social life in a way that is both material and symbolic. Like highways, railroads, or electrical grids, they have become embedded in everyday life, shaping and enabling certain forms of interaction. But unlike most traditional infrastructures, platforms often incorporate algorithmic decision-making, extract user data, and self-update at scale. This makes them active participants in social ordering rather than merely passive channels for human action.

If we look at “structure” in classical sociological terms (e.g., Durkheim, Parsons, Giddens), it often refers to stable patterns of social relations, institutions, norms, and resources. Yet platforms are dynamic systems: they morph through updates, policies, and algorithmic tweaking. Their affordances can rapidly shift, creating novel patterns of inclusion and exclusion. Hence, some scholars argue that “platform” highlights these new dynamics more precisely than the broad term “structure” does.

2. Platforms as Gateways to Agency

Argument for integration under existing categories (structure/agency):
From another perspective, the key insight of the structure–agency debate is precisely that social life involves ongoing interplay between broad social forces (structure) and individual or collective capacities to act (agency). Platforms, in this sense, can be interpreted as a new kind of social structure—one that is digitally mediated but still shapes and is shaped by human agency. Users can creatively engage with platforms, but their choices are funneled through interface designs, algorithmic rules, business models, and terms of service.

• Structure: The platform’s architecture, algorithms, data policies, and ownership models.
• Agency: The creative actions of individuals and groups using, resisting, or repurposing platform functions.

This reading suggests that platforms do not require a separate conceptual status but rather exemplify how digital infrastructures function as a powerful structural force, simultaneously enabling and constraining user agency.

3. Platforms and Power

Why “platform” might matter in its own right:
Digital platforms raise distinct questions about power and governance. They sit at the intersection of markets, states, and everyday life. For instance, data extraction, targeted advertising, and algorithmic curation introduce new forms of social control—often privatized, proprietary, and opaque. The notion of “platform governance” or “platform capitalism” points to logics that are different from classic state regulation or older capitalist enterprises.

• Datafication and Surveillance: Platforms collect and monetize user data, shaping surveillance practices in ways that produce new structural asymmetries (platform owners vs. users).
• Algorithmic Governance: Automated systems decide visibility, relevance, or even credibility of information. This shapes cultural production and public discourse.
• Monopolistic Power: Many platforms function as near-monopolies or oligopolies, reshaping labor (e.g., gig economy), media (e.g., news distribution), and personal relationships (social networking).

From this vantage, “platform” captures a constellation of power relations, economic models, and technical logics that may not be neatly encompassed by the more generic category of “structure.”

4. Platforms as Relational, Not Merely Technological

Scholars in platform studies emphasize that platforms are relational—they connect different user groups, advertisers, developers, governments, and so on, often under terms set unilaterally by the platform owner. These shifting relational ties create a meta-structure that is neither purely technological nor purely social. Some consider it akin to an institutional form or “platform institutionalism,” with rules and norms that differ substantially from, say, a family unit, a workplace, or a nation-state.

If we maintain a classic sociological approach, we can see this as a novel set of “structural” relations: hierarchy (platform owners) vs. participants (end-users, third-party developers, advertisers), each with differing degrees of power and knowledge. Yet the concept of “platform” helps us foreground how these new forms of relationality and governance are emergent and deeply intertwined with digital technology.

5. Should “Platform” Be Elevated to a Core Category?

The Case For

1. Novelty of Platform Power: The specific economic, algorithmic, and infrastructural logics of platforms (e.g., data extraction, network effects, lock-in, proprietary algorithms) represent a qualitative transformation of social organization and power.
2. Core of the Digital Age: As social life, politics, and culture become “platformized,” the phenomenon is central enough to demand explicit theorization distinct from standard social structures.
3. Hybrid Social–Technical Assemblages: Platforms complicate the long-standing sociological binaries of macro/micro or structure/agency, because they operate at multiple levels simultaneously: from user interface design (micro) to global data flows (macro).

The Case Against

1. Platforms as a Sub-Case of Structure: Ultimately, platforms are structures—albeit new, technologically driven ones. They belong within existing frameworks of sociological analysis (structure vs. agency, macro vs. micro, etc.).
2. Risk of Overemphasis: Declaring “platform” a brand-new category might fragment sociological theory or overshadow the fact that many “platform effects” replicate old inequalities or power struggles.
3. Existing Theoretical Tools: Concepts like field, network, institution, and discourse (Bourdieu, Latour, Goffman, Foucault) can be extended to analyze how platforms shape power, knowledge, and social interaction without the need for a separate fundamental category.

6. Conclusion: An Evolving Consensus

While it remains open whether “platform” will endure as a sociological category on par with “structure” or “agency,” there is little doubt that it has become central to empirical studies of digital society. Much like “institution” or “field” in earlier sociological writing, “platform” captures a specific way that social relations are organized and governed.

• Analytically, it can be productive to treat “platform” as a key concept—one that highlights the socio-technical and economic specificities of digital media environments.
• Theoretically, whether it becomes a foundational category alongside “structure” and “agency” depends on how well it can integrate with, or transform, our broader frameworks of sociological thought.

In practice, many sociologists today do treat “platform” as a near-axiomatic concept, precisely because so much of contemporary social life—including labor, culture, politics, and personal relationships—runs through platform infrastructures. At the same time, some argue that we should tread cautiously, applying “platform” as a lens to analyze new social forms without prematurely elevating it to the status of an all-encompassing category.

In short, there is no settled answer—but there is a clear imperative to study platforms and to refine our theoretical tools in light of the profound reorganization of social life that platforms are bringing about. Whether that means adding “platform” to the triumvirate of “structure” and “agency” is an ongoing, generative debate in contemporary sociology.

#GPTO1 #platformStudies #reasoningModels

»#OpenAI trained #o1 and #o3 to 'think' about its #safetypolicy: outlining the company’s latest way to ensure #AI #reasoningmodels stay aligned with the #values of their #humandevelopers.« https://techcrunch.com/2024/12/22/openai-trained-o1-and-o3-to-think-about-its-safety-policy/?eicker.news #tech #media

OpenAI's o1 marks a major shift in the AI industry, moving away from prediction-based LLMs to reasoning models that aim to overcome their limitations. 🔍🤖 #OpenAI #AI #MachineLearning #ReasoningModels #ArtificialIntelligence #TechInnovation #AIShift