Wiredframe Radar logo

Category intelligence

Social Media Briefing — April 28, 2026

493 current items analyzed and ranked.

Executive synthesis

Social Media Summary

The OpenAI-Microsoft partnership restructuring dominated the day. Sam Altman announced OpenAI can now offer products across all clouds while Microsoft remains primary partner. Simon Willison highlighted the removal of the AGI clause from revenue-sharing terms. Altman also celebrated strong developer reception to GPT-5.5.

  • China blocked Meta's $2B acquisition of Manus, signaling Beijing now treats AI software and talent as strategic national assets ahead of the Trump-Xi meeting
  • Andrew Ng published a detailed essay on AI-native software teams — engineers becoming generalists, 1:1 engineer-to-PM ratios, and new downstream bottlenecks
  • hardmaru presented an ICLR 2026 paper on a Conductor model trained via RL to orchestrate pools of LLMs, powering Sakana AI's Fugu system
  • vLLM announced day-0 support for DeepSeek V4 base models, continuing rapid open-source infrastructure buildout
  • Yudkowsky made a substantive case for international AI governance treaties, drawing nuclear arms control parallels
  • Ethan Mollick argued every AI debate reduces to the shape of the capability S-curve, offering an influential framing for policy and business discussions
  • A Google researcher's paper claiming AI consciousness is mathematically impossible sparked philosophical debate across the community

Key Themes

OpenAI-Microsoft Partnership Restructuring · 4AI-Native Team Transformation · 1US-China AI Geopolitics / Manus Acquisition · 2Multi-Agent AI Systems Research · 1Vibe Coding Disasters & AI Agent Safety · 16GPT-5.5 Reception and OpenAI · 3DeepSeek V4 Ecosystem · 2AI Safety, Governance, and International Treaties · 7GPT-5.5 Early Demonstrations · 4AI S-Curve & Capability Assessment · 3

Primary evidence

Top Ranked Signals

Social Twitter Apr 27

we have updated our partnership with microsoft. microsoft will remain our primary cloud partner, bu...

By @sama

95 score
AI Analysis

Sam Altman announces updated OpenAI-Microsoft partnership: Microsoft remains primary cloud partner but OpenAI can now offer products across all clouds. Revenue share through 2030, model/product provision through 2032.

we have updated our partnership with microsoft. microsoft will remain our primary cloud partner, but we are now able to make our products and services available across all clouds. will continue to provide them with models and products until 2032, and a revenue share through 2030.
OpenAI-Microsoft relationshipcloud strategyAI industry dynamicsbusiness strategy
Social Twitter Apr 27

AI-native software engineering teams operate very differently than traditional teams. The obvious di...

By @AndrewYNg

82 score
AI Analysis

Andrew Ng writes a detailed essay on how AI-native software teams operate: engineers becoming generalists, 1:1 engineer-to-PM ratios, small co-located teams, coding speed creating bottlenecks in marketing/legal/design, and the value of generalists in small teams.

AI-native software engineering teams operate very differently than traditional teams. The obvious difference is that AI-native teams use coding agents to build products much faster, but this leads to many other changes in how we operate. For example, some great engineers now play broader roles than just writing code. They are partly product managers, designers, sometimes marketers. Further, small teams who work in the same office, where they can communicate face-to-face, can move incredibly quic
AI-native teamsfuture of worksoftware engineeringagentic codingorganizational change
Social Twitter Apr 27

China just blocked Meta's $ 2B December acquisition of Manus and told Zuck to unwind the deal. Manu...

By @TheRundownAI

82 score
AI Analysis

China has blocked Meta's $2B acquisition of Manus (AI startup). Though Singapore-incorporated, Manus has Chinese founders and offices. Teams are already merged with engineers working at Meta's Singapore office. Some Manus execs restricted from leaving China. Seen as Beijing drawing a line on Chinese AI talent before Trump-Xi meeting.

China just blocked Meta's $ 2B December acquisition of Manus and told Zuck to unwind the deal. Manus is technically Singapore-incorporated, but its founders, parent company, and offices in Beijing and Wuhan are all Chinese. Awkwardly, the teams are already merged, with Manus engineers reportedly working at Meta's Singapore office for months. Several of its execs have also been restricted from leaving China. Just weeks before Trump and Xi sit down, Beijing just drew a major line for Chinese A
geopoliticsus_china_ai_competitionmetaai_acquisitionsregulation
Social Twitter Apr 27

For the past few years, humans have been doing “prompt engineering” to coax the best performance out...

By @hardmaru

39 score
AI Analysis

As first reported in Social yesterday, hardmaru presents an ICLR 2026 paper on training a 'Conductor' model with RL to manage a pool of LLMs — learning to write custom instructions, dynamically adapting strategies (single model for simple tasks, multi-agent pipelines for complex ones). Connected to Sakana's Fugu system.

For the past few years, humans have been doing “prompt engineering” to coax the best performance out of different LLMs. In this work, we explored what happens if we train an AI to do that job instead. By training a Conductor model with RL, we found that it naturally learns to write highly effective, custom instructions for a whole pool of other models. It essentially learns to ‘manage’ them in natural language. What surprised me most was how it dynamically adapts. For simple factual questions,
multi-agent systemsAI researchreinforcement learningAI orchestrationICLR 2026
Social Twitter Apr 27

so fun to see the reception to 5.5! there is almost nothing that feels more gratifying to me than b...

By @sama

39 score
AI Analysis

Continuing the GPT-5.5 launch buzz from Social yesterday, Sam Altman celebrates the positive reception to GPT-5.5, expressing gratitude for builders finding OpenAI's tools useful.

so fun to see the reception to 5.5! there is almost nothing that feels more gratifying to me than builders saying they find our tools useful.
gpt5.5_receptionopenai_strategydeveloper_tools
Social Twitter Apr 27

Total AI disaster, totally predictable

By @GaryMarcus

72 score
AI Analysis

Gary Marcus declares a 'Total AI disaster, totally predictable' — a highly viral post with 2.2M views, likely referencing a specific AI failure incident.

Total AI disaster, totally predictable
ai_failureai_hype_criticismai_safety
Social Twitter Apr 27

vLLM support for DeepSeek V4 base models is on the way! The V4 release includes 4 models: base/inst...

By @vllm_project

72 score
AI Analysis

Following yesterday's News coverage of DeepSeek V4, vLLM announces support for DeepSeek V4 base models is coming. V4 includes 4 models (base/instruct × flash/pro). They collaborated with DeepSeek to add expert_dtype field to distinguish fp4 (instruct) vs fp8 (base) models.

vLLM support for DeepSeek V4 base models is on the way! The V4 release includes 4 models: base/instruct × flash/pro. Initial support covers the instruct versions. To extend support to the base models, we worked with @deepseek_ai to add an expert_dtype field in the config, making it easy to distinguish between them (fp4 for instruct, fp8 for base). Config commit: t.co/EZRpjQhwES vLLM PR: t.co/ioAZ1acnou Model support isn't easy, but close collaboration with model vendors keeps
deepseek_v4vllmai_infrastructureopen_sourcemodel_serving
Social Twitter Apr 27

@SecScottBessent Unless we unwarrantedly reject concerns dating back to the 1920s and hundreds of mo...

By @ESYudkowsky

68 score
AI Analysis

Yudkowsky makes a detailed case for international AI governance treaties, comparing the situation to nuclear arms control. Argues for GPU export controls, ASML equipment restrictions, and coordination with China to prevent uncontrolled machine superintelligence.

@SecScottBessent Unless we unwarrantedly reject concerns dating back to the 1920s and hundreds of modern expert statements that AI extinction risk is a concern, the world needs to step back. The USA should not try to halt alone; that wouldn't work. So Senator Sanders seems to me to be straightforwardly taking the correct posture here -- unless we unwarrantedly and without any reasonable argument reject the possibility of superhuman AI; or even more bizarrely, suppose that it will serve humanit
ai_governanceai_safetyinternational_treatiesgpu_export_controlsgeopolitics
Social Twitter Apr 27

A Google researcher just proved AI consciousness is mathematically impossible. Not in 10 years. No...

By @AlphaSignalAI

68 score
AI Analysis

A Google researcher published a paper arguing AI consciousness is mathematically impossible, based on the 'Abstraction Fallacy' - computation is a description of a process, not the process itself. The paper argues simulation is not instantiation, and if a machine were ever aware it would be from physical makeup, not code.

A Google researcher just proved AI consciousness is mathematically impossible. Not in 10 years. Not in 100. Ever. The argument is structural, not technical. Computation is a description of a process, not the process itself. For something to "compute," a conscious observer must first carve reality into symbols and assign meaning. Without that observer, there are only voltage gradients. The paper calls this the Abstraction Fallacy. The analogy that makes it click: > A GPU can simulate p
ai_consciousnessphilosophy_of_mindai_limitationsresearch_papers
Social Twitter Apr 27

Every AI discussion ultimately rests on two questions: how good can AI get? And how fast? They are p...

By @emollick

65 score
AI Analysis

Mollick argues every AI discussion reduces to two questions: how good can AI get, and how fast (the S-curve shape). Job impact, risks, etc. are all downstream of these.

Every AI discussion ultimately rests on two questions: how good can AI get? And how fast? They are predictions about the s-curve shape. Everything else (job impact, potential risks, etc.) is downstream of those questions. I think it would be useful to focus on them more often. t.co/bmlrW8cfHk
AI predictionsS-curve analysisAI capabilitiesfuture of work
Social Twitter Apr 27

Lately I've been having fun with running coding agents fully locally. The setup I landed on is: - P...

By @patloeber

65 score
AI Analysis

Pat Loeber shares his setup for running coding agents fully locally: Pi agent + Gemma 4 26B A4B + LM Studio/Ollama/llama.cpp. Includes a step-by-step guide.

Lately I've been having fun with running coding agents fully locally. The setup I landed on is:
  • Pi agent
  • Gemma 4 26B A4B
  • Server of choice: LM Studio/Ollama/llama.cpp
I wrote a step-by-step guide with instructions on how to set it up: t.co/g21F22JFAI t.co/qeKHIfAkaD
local-aicoding-agentsgemma-4developer-toolsopen-models