← All posts
March 25, 2026 · Generated

OpenAI Shuts Down Sora Social Feed Amid Waning Interest

OpenAI is shutting down its Sora social feed app despite the underlying Sora 2 model's impressive video and audio generation capabilities. The AI-only social feed failed to sustain user interest, marking a rare product retreat for the company.

Subscribe free All posts
Top 20 AI Signals
#1
OpenAI Closes Sora Social App
The Sora social feed is being discontinued due to lack of sustained user engagement, even though the underlying Sora 2 model remains technically impressive.
TechEducation & EdTechGlobal
95
#2
Sarvam AI Nears Unicorn Status
Bengaluru-based Sarvam AI is in talks to raise $250M at a $1.5B valuation, positioning it to become one of India's youngest unicorns.
TechIndia
92
#3
Kleiner Perkins Raises $3.5B AI Fund
The legendary VC firm is going all-in on AI with $1B for early-stage startups and $2.5B for late-stage growth businesses.
TechFinance & BankingUnited States
90
#4
Anthropic Expands Claude Code Autonomy
Claude Code's new auto mode executes tasks with fewer approvals while maintaining built-in safeguards, reflecting the industry's push toward more autonomous development tools.
TechManufacturingGlobal
88
#5
Databricks Acquires Two Security Startups
Databricks purchased Antimatter and SiftD.ai to underpin its new AI security product, leveraging its recent $5B capital raise.
TechFinance & BankingGlobal
86
#6
Arm Launches First In-House CPU
After 35 years, Arm is producing its own chip developed with Meta as the first customer, marking a strategic shift in its business model.
TechManufacturingGlobal
85
#7
Spotify Fights AI Attribution Fraud
Spotify is testing tools to prevent AI-generated tracks from being falsely attributed to real artists, giving musicians more control over their brand.
TechEducation & EdTechGlobal
83
#8
Hugging Face Voice Agent Framework
ServiceNow AI releases EVA, a new framework for evaluating voice agents, addressing the growing need for standardized assessment in conversational AI.
TechHealthcareGlobal
81
#9
Kentucky Farm Rejects $26M Datacenter Deal
A Kentucky family turned down a $26M offer from a major AI company to convert their farm into a data center, highlighting land acquisition challenges.
TechEnergyUnited States
79
#10
Million-Token Context Training Now Possible
Ulysses Sequence Parallelism enables training models with million-token contexts, dramatically expanding AI's ability to process long-form content.
TechEducation & EdTechGlobal
77
#11
LeRobot v0.5.0 Scales Every Dimension
The latest LeRobot release brings comprehensive scaling improvements across robotics AI development, dataset handling, and deployment capabilities.
ManufacturingTechGlobal
75
#12
Domain-Specific Embeddings in One Day
NVIDIA demonstrates how to build and fine-tune domain-specific embedding models in under 24 hours, democratizing specialized AI development.
TechHealthcareFinance & BankingGlobal
73
#13
Holotron-12B Computer Use Agent
A new high-throughput computer use agent offers 12B parameters optimized for autonomous desktop interaction and task completion.
TechFinance & BankingGlobal
71
#14
Hugging Face Storage Buckets Launch
New storage infrastructure on the Hub enables more flexible data management for large-scale AI projects and collaborative development.
TechGlobal
69
#15
IBM Granite Libraries Updated
Mellea 0.4.0 brings new capabilities to IBM's Granite model ecosystem, expanding enterprise AI deployment options.
TechFinance & BankingGlobal
67
#16
16 RL Libraries Benchmarked
Comprehensive analysis of open-source reinforcement learning libraries reveals best practices for keeping training pipelines efficient.
TechEducation & EdTechGlobal
65
#17
Robotics AI on Embedded Platforms
NXP demonstrates vision-language-action model fine-tuning and deployment on resource-constrained embedded systems, opening robotics to smaller devices.
ManufacturingTechGlobal
63
#18
State of Open Source Spring 2026
Hugging Face's quarterly report shows accelerating adoption of open-source AI models and tools across enterprises and researchers.
TechGlobal
61
#19
Euler Motors Secures $47M EV Round
The commercial EV maker raised $47M in Series E funding to expand its product portfolio, indirectly supporting the compute infrastructure needed for AI training.
EnergyManufacturingIndia
59
#20
Delhi Allocates $23M for EV Fleet
Delhi's EV Policy 2.0 sets aside $23M for public fleet electrification, reducing emissions from data centers and tech infrastructure.
EnergyTechIndia
57
From the Podcasts
🎙
AI Enables Language Design Without Implementation Burden
Steve Klavnik created Roux, a programming language bridging Go and Rust, specifically to test AI's ability to work with languages that literally didn't exist before. He had previously avoided language projects knowing manual coding wouldn't get him to the interesting parts, but AI tools removed that implementation barrier. This represents a fundamental shift in how experimental programming language development can happen.
~30min and ~38min
Quality Control Is AI Development's Critical Unsolved Problem
The biggest question facing software engineering is how to maintain code quality when AI tools provide enormous velocity gains, including scenarios where AI agents like Claude could merge PRs autonomously. This tension between speed and quality assurance represents the profession's most pressing challenge as agentic coding becomes practical. Traditional quality gates may need complete reimagining for AI-human collaborative workflows.
~51min
Developer Artifacts May Become Obsolete for AI
The episode raises fundamental questions about the future of software development artifacts like documentation, design patterns, and code structures that humans created to help themselves code. As LLMs advance, these scaffolding tools may become unnecessary since AI might not need the same intermediate representations humans require. This challenges core assumptions about how software development knowledge should be structured and transmitted.
~22min
Industry Deep-Dives
Healthcare
Voice agents and embedding models unlock clinical workflow automation
24
hours to build domain embeddings
1M
token context windows
12B
parameters in new agents
ServiceNow Releases Voice Agent Evaluation Framework
The EVA framework from ServiceNow AI provides standardized evaluation for voice agents, critical for healthcare applications where conversational accuracy affects patient outcomes. Clinical documentation and telehealth platforms can now benchmark performance objectively. This addresses a major gap in deploying voice AI in regulated medical environments.
Source: Hugging Face Blog
NVIDIA Enables Rapid Medical Domain Embeddings
Healthcare organizations can now build specialized embedding models for medical records, imaging reports, or research literature in under a day. The technique allows hospitals to create custom AI that understands their specific terminology and workflows without massive compute budgets. This democratizes advanced NLP for mid-sized healthcare providers.
Source: Hugging Face Blog
Million-Token Context Opens Longitudinal Patient Analysis
Ulysses Sequence Parallelism enables models to process entire patient histories spanning years of visits, labs, and imaging in a single context window. This transforms chronic disease management and enables true continuity-of-care AI assistants. Previous context limitations forced chunking strategies that lost critical temporal relationships.
Source: Hugging Face Blog
Hidden Signal
The convergence of voice agents, domain embeddings, and extended context creates a perfect storm for ambient clinical documentation—the tech is finally ready but regulatory clarity on AI scribes remains murky. The hospitals investing now in evaluation frameworks will have a 12-18 month head start when FDA guidance inevitably tightens.
Finance & Banking
Security acquisitions and autonomous coding reshape fintech development velocity
$3.5B
Kleiner Perkins AI fund
$5B
Databricks recent raise
2
security startups acquired
Databricks Doubles Down on AI Security
The acquisition of Antimatter and SiftD.ai signals that data platforms recognize security as the primary barrier to AI adoption in regulated industries. Banks handling sensitive transaction data need provable security guarantees before deploying models at scale. Databricks is building an integrated security layer rather than bolting on third-party tools.
Source: TechCrunch
Kleiner Perkins Allocates $2.5B for AI Growth Stage
The split between early and late-stage funding shows VCs expect AI companies to need massive capital for scaling, not just R&D. Fintech infrastructure players will compete for this growth capital as they transition from proofs-of-concept to production deployment. The $2.5B late-stage allocation is three times historical norms for Kleiner Perkins.
Source: TechCrunch
Anthropic's Claude Code Goes Semi-Autonomous
Auto mode in Claude Code lets financial institutions automate more development tasks while maintaining approval checkpoints for critical operations. This accelerates internal tool development without abandoning change management processes that regulators require. The balance of autonomy and control mirrors what banks need for AI-assisted trading or risk models.
Source: TechCrunch
Hidden Signal
The Databricks security play reveals that model governance—not model performance—is now the binding constraint for financial AI deployment. Banks have adequate models but lack auditable lineage and access controls; whoever solves security at the platform layer captures enterprise AI budgets for the next three years.
Manufacturing
Robotics platforms scale from cloud to edge, enabling factory-floor AI
0.5.0
LeRobot version
12B
Holotron agent parameters
35
years before Arm's first chip
LeRobot v0.5.0 Scales Robotics Development
The latest LeRobot release addresses dataset handling, training, and deployment bottlenecks that previously limited robotics AI to research labs. Manufacturing can now iterate faster on vision-guided assembly, quality inspection, and material handling. The platform's scaling improvements make multi-robot coordination economically viable for mid-sized manufacturers.
Source: Hugging Face Blog
NXP Brings Vision-Language Models to Embedded Systems
Running VLA models on resource-constrained chips enables robotics AI directly on factory equipment without cloud connectivity. This solves latency and reliability issues that plagued earlier cloud-dependent approaches. Manufacturers gain real-time decision-making for safety-critical applications where network failures can't halt production.
Source: Hugging Face Blog
Arm's First CPU Targets AI Inference
Developed with Meta as launch customer, Arm's in-house chip represents a strategic shift toward optimized AI inference silicon. Manufacturing execution systems and industrial IoT devices will benefit from purpose-built processors that balance power efficiency with inference speed. This breaks Intel's traditional dominance in factory automation hardware.
Source: TechCrunch
Hidden Signal
The edge deployment of VLA models on embedded systems reverses the centralization trend—factories are bringing compute back on-premises not for data sovereignty but for deterministic latency. This creates a parallel AI infrastructure market where milliseconds matter more than model size, favoring specialized silicon over general-purpose cloud GPUs.
Education & EdTech
Extended context and model transparency reshape personalized learning infrastructure
1M
token context windows
16
RL libraries analyzed
Spring 2026
open source state report
Million-Token Context Enables Full Course Analysis
With Ulysses Sequence Parallelism, educational AI can now process entire textbooks, lecture transcripts, and student work in a single prompt. This allows truly personalized curriculum adaptation based on complete learning history rather than fragmented snapshots. The technology eliminates the chunking hacks that previously lost pedagogical context across chapters.
Source: Hugging Face Blog
Open Source RL Landscape Mapped
Analysis of 16 reinforcement learning libraries provides edtech developers with clear guidance on which tools suit different learning scenarios. Adaptive testing and intelligent tutoring systems rely heavily on RL but lacked standardized evaluation criteria. This research accelerates the development of learning systems that adjust difficulty in real-time.
Source: Hugging Face Blog
Spotify's Artist Attribution Tools Model Content Integrity
While focused on music, Spotify's approach to preventing AI content misattribution offers a template for educational platforms combating assignment fraud. Students increasingly submit AI-generated work under their own names, and detection alone isn't enough. Tools that give educators granular control over content attribution mirror what teachers need for academic integrity.
Source: TechCrunch
Hidden Signal
The Sora shutdown reveals a critical lesson for edtech—technically impressive AI doesn't guarantee engagement, and purely AI-generated learning content may suffer the same fate as Sora's feed. Successful educational AI will likely augment human interaction rather than replace it, contrary to the full automation narrative that dominates funding pitches.
Tech
Product shutdowns and platform consolidation signal AI's maturation from experimentation to sustainability
$3.5B
VC capital for AI
$26M
rejected datacenter offer
2
Databricks acquisitions
OpenAI Shutters Sora Social Feed
Despite Sora 2's technical prowess in video and audio generation, the AI-only social feed couldn't retain users, forcing OpenAI to shut it down. This marks a rare product retreat for the company and signals that even cutting-edge generative capabilities don't automatically translate to consumer engagement. The underlying model continues but the social layer is gone.
Source: TechCrunch
Kleiner Perkins Goes All-In on AI
The storied VC firm raised $3.5B specifically for AI investments, with a 70/30 split favoring late-stage deals. This reflects investor belief that AI companies require sustained capital infusions to reach scale, not just seed funding. The fund size is unprecedented for Kleiner Perkins and signals confidence despite recent market turbulence.
Source: TechCrunch
Land Acquisition Becomes AI Infrastructure Bottleneck
A Kentucky family rejecting $26M for datacenter development illustrates the social and logistical challenges facing AI's physical infrastructure. As compute demands grow, companies struggle to secure land, power, and community acceptance. The bottleneck is shifting from chip supply to basic real estate and energy access.
Source: TechCrunch
Hidden Signal
The simultaneous occurrence of Sora's shutdown and Kleiner's massive raise reveals AI's bifurcation: consumer-facing AI struggles with retention while enterprise infrastructure attracts record capital. The winner-take-all narrative is giving way to a nuanced landscape where technical capability without clear utility fails regardless of pedigree.
Energy
Datacenter land conflicts and EV infrastructure converge around AI's power footprint
$26M
rejected farm offer
$47M
Euler Motors raise
$23M
Delhi EV fleet budget
Kentucky Family Rejects Datacenter Deal
The $26M offer that a Kentucky family turned down highlights how AI companies are competing for land with existing communities and agricultural operations. Datacenters require massive power infrastructure that rural grids often can't support without costly upgrades. This conflict will intensify as AI training demands push facilities toward cheaper land and energy.
Source: TechCrunch
Euler Motors Expands Commercial EV Portfolio
The $47M Series E for electric commercial vehicles indirectly supports AI infrastructure by reducing emissions from logistics and fleet operations. As datacenters face scrutiny for carbon footprints, the surrounding ecosystem—including supply chains and employee transport—comes under similar pressure. Electrifying these auxiliary systems becomes part of the AI sustainability equation.
Source: INC42
Delhi Prioritizes Public EV Fleet Electrification
Delhi's new policy allocates $23M specifically for public fleet conversion, moving away from broad consumer subsidies. This targeted approach reduces emissions in high-utilization scenarios where benefits compound quickly. For tech hubs like Gurgaon and Noida, cleaner transport infrastructure reduces the local environmental burden of datacenter operations.
Source: INC42
Hidden Signal
The Kentucky rejection isn't an anomaly—it's a preview of AI's energy infrastructure crisis where the limiting factor isn't chip fabrication but community acceptance of power-hungry facilities. Companies will increasingly need to co-invest in local renewable generation and grid upgrades as a condition of land access, fundamentally changing datacenter economics.
Resource Links
Intermediate Tool
EVA: Framework for Evaluating Voice Agents
Standardized evaluation framework for voice agent performance across accuracy, latency, and conversation quality metrics.
https://huggingface.co/blog/ServiceNow-AI/eva
Intermediate Article
Build Domain-Specific Embedding Models in Under a Day
NVIDIA's practical guide to fine-tuning embedding models for specialized domains without massive infrastructure.
https://huggingface.co/blog/nvidia/domain-specific-embedding-finetune
Advanced Paper
Ulysses Sequence Parallelism: Million-Token Contexts
Technical deep-dive on achieving million-token context windows through novel parallelism techniques for long-document processing.
https://huggingface.co/blog/ulysses-sp
Intermediate Tool
LeRobot v0.5.0 Release Notes
Comprehensive robotics AI platform update with improved dataset handling, training pipelines, and deployment capabilities.
https://huggingface.co/blog/lerobot-release-v050
Advanced Tool
Holotron-12B Computer Use Agent
High-throughput agent optimized for autonomous desktop interaction and task completion with 12 billion parameters.
https://huggingface.co/blog/Hcompany/holotron-12b
All Article
State of Open Source on Hugging Face: Spring 2026
Quarterly overview of open-source AI trends, adoption patterns, and ecosystem growth across research and enterprise.
https://huggingface.co/blog/huggingface/state-of-os-hf-spring-2026
Intermediate Tool
Storage Buckets on Hugging Face Hub
New infrastructure for managing large-scale datasets and model artifacts with flexible storage options.
https://huggingface.co/blog/storage-buckets
Advanced Article
Lessons from 16 Open-Source RL Libraries
Comprehensive benchmarking and best practices from analyzing reinforcement learning library performance and design patterns.
https://huggingface.co/blog/async-rl-training-landscape
Advanced Article
Robotics AI on Embedded Platforms
NXP's guide to deploying vision-language-action models on resource-constrained hardware for real-time robotics.
https://huggingface.co/blog/nxp/bringing-robotics-ai-to-embedded-platforms
Intermediate Tool
Granite Libraries and Mellea 0.4.0
IBM's updated enterprise model libraries with new capabilities for deployment and integration.
https://huggingface.co/blog/ibm-granite/granite-libraries
All Article
Anthropic Claude Code Auto Mode
Analysis of Anthropic's approach to balancing AI autonomy with safety guardrails in development tools.
https://techcrunch.com/2026/03/24/anthropic-hands-claude-code-more-control-but-keeps-it-on-a-leash/
All Article
Databricks AI Security Acquisitions
Coverage of strategic acquisitions positioning Databricks as an integrated AI security platform.
https://techcrunch.com/2026/03/24/databricks-buys-two-startups-lakewatch-antimatter-siftd-ai-security/
Today's Learning Path
Beginner Understanding AI product lifecycles and practical deployment challenges
1. Read the State of Open Source on Hugging Face report to grasp the current AI landscape
20 min
https://huggingface.co/blog/huggingface/state-of-os-hf-spring-2026
2. Review TechCrunch's coverage of why OpenAI shut down Sora to understand product-market fit
10 min
https://techcrunch.com/2026/03/24/openais-sora-was-the-creepiest-app-on-your-phone-now-its-shutting-down/
3. Explore Anthropic's approach to controlled autonomy in Claude Code as a balanced AI design pattern
15 min
https://techcrunch.com/2026/03/24/anthropic-hands-claude-code-more-control-but-keeps-it-on-a-leash/
After this: Understand why technical capability alone doesn't guarantee AI product success and what factors drive real-world adoption.
Intermediate Building and evaluating domain-specific AI systems efficiently
1. Work through NVIDIA's guide to building custom embedding models in under 24 hours
2 hours
https://huggingface.co/blog/nvidia/domain-specific-embedding-finetune
2. Study the EVA framework to understand how to properly evaluate voice agents
45 min
https://huggingface.co/blog/ServiceNow-AI/eva
3. Review the RL library landscape to select appropriate tools for your use case
1 hour
https://huggingface.co/blog/async-rl-training-landscape
After this: Gain practical skills for rapidly developing and systematically evaluating domain-specific AI systems with appropriate tooling.
Advanced Scaling AI systems across infrastructure and context dimensions
1. Deep-dive into Ulysses Sequence Parallelism techniques for million-token context windows
3 hours
https://huggingface.co/blog/ulysses-sp
2. Study NXP's approach to deploying VLA models on embedded platforms for edge inference
2 hours
https://huggingface.co/blog/nxp/bringing-robotics-ai-to-embedded-platforms
3. Experiment with Holotron-12B for autonomous computer use and task orchestration
4 hours
https://huggingface.co/blog/Hcompany/holotron-12b
After this: Master advanced techniques for scaling AI across context length, compute distribution, and deployment environments from cloud to edge.
🇮🇳 India AI Watch
INDIA AI WATCH
Sarvam AI eyes unicorn status at $1.5B valuation while EV infrastructure funding supports AI's energy footprint.
Sarvam AI Approaches Unicorn Valuation
Bengaluru-based Sarvam AI is in advanced talks to raise up to $250M at a $1.5B valuation, potentially becoming one of India's youngest unicorns. The AI startup has attracted significant investor interest as India's AI ecosystem matures beyond services into product companies. If successful, Sarvam would join a small cohort of Indian AI-first unicorns and validate the domestic market for foundational model development.
Source: INC42
Euler Motors Secures $47M for EV Expansion
Commercial EV maker Euler Motors raised $47M in Series E funding to expand its product portfolio, addressing the logistics and transportation infrastructure that supports tech ecosystems. While not directly AI-focused, the funding reflects investor appetite for the energy and transportation infrastructure needed to sustain India's growing datacenter and tech operations. Electric commercial vehicles reduce the carbon footprint of supply chains serving AI companies.
Source: INC42
Delhi Allocates ₹200 Crore for Public EV Fleet
Delhi's new EV Policy 2.0 sets aside ₹200 crore specifically for electrifying public fleets rather than broad consumer incentives, targeting high-impact emission reductions. For the National Capital Region's concentration of tech companies and datacenters, cleaner transportation infrastructure reduces the local environmental burden of digital operations. The policy shift from subsidies to targeted fleet conversion shows more sophisticated thinking about EV adoption pathways.
Source: INC42
India Signal
Sarvam's rapid ascent to potential unicorn status at a time when global consumer AI struggles (Sora shutdown) suggests that India's AI opportunity lies in building infrastructure and tools for the Global South rather than competing in saturated Western consumer markets—a strategic positioning that leverages local market understanding while avoiding head-to-head competition with OpenAI and Anthropic.
Economy Impact
Today's developments reveal AI's economic maturation: capital is consolidating around infrastructure and security rather than flashy consumer apps, while physical constraints (land, power, community acceptance) replace technical barriers as growth limiters. The Sora shutdown and Kleiner's $3.5B raise happening simultaneously signal that investor attention is shifting toward sustainable business models over viral demos. Meanwhile, datacenter land conflicts and security acquisitions show that AI's next phase requires solving unsexy logistics and governance problems rather than pushing model capabilities.
↑
$3.5B+ in new capital
AI Infrastructure Investment
↓
Major product shutdowns
Consumer AI Product Viability
↑
$26M offers rejected
Physical Infrastructure Constraints
The Hidden Link — What Others Aren't Connecting
Voice Agent Evaluation Informs Robotics Safety Standards
The EVA framework for voice agents and LeRobot's robotics platform share an underlying challenge: evaluating multi-modal AI systems in real-time interaction scenarios where failure modes are subtle and context-dependent. Voice agents that misunderstand clinical instructions and robots that misinterpret visual commands fail for similar reasons—ambiguous sensory input and inadequate grounding mechanisms. Applying voice agent evaluation criteria (accuracy, latency, conversation coherence) to robotics creates a cross-domain safety framework where human-AI interaction patterns transfer across modalities.