Nipun Taneja

@OpenAIDevs If Codex succeeds in India, it won’t be because of events. It’ll be because it handles the unglamorous parts of engineering well: reading intent from broken context, making fewer high-cost edits, and knowing when to stop.

@MiniMax_AI I’d still be careful with “on par with Sonnet 4.6 in OpenClaw.” Agentic evals are highly scaffold-sensitive. Small changes in tools, prompts, or recovery policy can move outcomes a lot more than people admit.

Introducing MiniMax-M2.7, our first model which deeply participated in its own evolution, with an 88% win-rate vs M2.5 - Production-Ready SWE: With SOTA performance in SWE-Pro (56.22%) and Terminal Bench 2 (57.0%), M2.7 reduced intervention-to-recovery time for online incidents to 3-min on certain occasions. - Advanced Agentic Abilities: Trained for Agent Teams and tool search tool, with 97% skill adherence across 40+ complex skills. M2.7 is on par with Sonnet 4.6 in OpenClaw. - Professional Workspace: SOTA in professional knowledge, supports multi-turn, high-fidelity Office file editing. MiniMax Agent: agent.minimax.io API: platform.minimax.io Token Plan: platform.minimax.io/subscribe/toke…

@openart_ai Most creative AI tools still reset the universe on every prompt. A navigable world is interesting because it turns generation into exploration. That’s a better primitive for storytelling than endless rerolls.

Today, we’re launching a new way to create with AI. With OpenArt Worlds, you can generate a fully navigable 3D environment from a single prompt or image, step inside it, and capture shots exactly the way you envision them. No more starting over. No more inconsistent scenes. You build the world once - and create inside it. • Move through your scene freely • Find your angles • Add characters and elements • Capture production-ready shots

@stitchbygoogle The important feature here isn’t speed, it’s memory. Can the tool hold onto design principles, edge cases, and system rules across a whole product surface? That’s where most AI design tools break.

Meet the new Stitch, your vibe design partner. Here are 5 major upgrades to help you create, iterate and collaborate: 🎨 AI-Native Canvas 🧠 Smarter Design Agent 🎙️ Voice ⚡️ Instant Prototypes 📐 Design Systems and DESIGN.md Rolling out now. Details and product walkthrough video in 🧵

𝐍𝐞𝐰 𝐜𝐨𝐧𝐯𝐞𝐫𝐬𝐚𝐭𝐢𝐨𝐧𝐬, 𝐟𝐫𝐞𝐬𝐡 𝐢𝐝𝐞𝐚𝐬, 𝐚𝐧𝐝 𝐚𝐥𝐥 𝐭𝐡𝐢𝐧𝐠𝐬 𝐦𝐚𝐫𝐤𝐞𝐭𝐢𝐧𝐠 Meet me at @adtechIndia New Delhi, connecting with the minds shaping what’s next. Let’s build, collaborate, and vibe together.

@Google This is interesting But a single embedding space assumes that text, image, and audio semantics align neatly, which isn’t always true in real use cases.

Gemini Embedding 2, our first fully multimodal embedding model, is now available in Public Preview via the Gemini API and Vertex AI. Developers can now map text, images, video, and audio in one centralized space, with one model, which simplifies complex tasks like semantic search. Here's what this means and why it matters 🧵↓

@AravSrinivas This probably works great on clean flows. Curious how it behaves when things break or don’t match expectations.

Computer can now use your local browser Comet as a tool. Which makes it possible for Computer to do anything, even without connectors or MCPs. This is a unique advantage Computer possesses that no other tool on the market can match.

@perplexity_ai Feels like we’re getting closer to agents that can do anything. Still unclear how we guide them to do the right thing consistently.

Computer can now take full control of Comet to complete tasks. When you’re in Comet, Computer spins up a browser agent that can access any site or logged‑in app with your permission, without the need for connectors or MCPs. Available to all Computer users on Comet.

@askOkara Feels like this helps you do more marketing. Not sure it helps you do better marketing yet.

Today we're introducing the world's first AI CMO. Enter your website and it deploys a team of agents to help you get traffic and users. Try it now at okara.ai/cmo

@NVIDIAGeForce Techniques like DLSS effectively treat rendering as an inference problem. That’s powerful, but it also means realism becomes a function of the model’s priors, not just the underlying scene.

Announcing NVIDIA DLSS 5, an AI-powered breakthrough in visual fidelity for games, coming this fall. DLSS 5 infuses pixels with photorealistic lighting and materials, bridging the gap between rendering and reality. Learn More → nvidia.com/en-us/geforce/…

@xai There’s an asymmetry here. Speech is fast to produce but slow to consume, which limits where it fits in high-frequency workflows.

Grok's Text to Speech API is now available. Start building with natural voices and expressive controls to bring your apps to life. #text-to-speech" target="_blank" rel="nofollow noopener">x.ai/api/voice#text…

@OpenAIDevs This starts to look like a multi-agent system with implicit message passing. In those systems, performance is often bounded not by individual agent capability, but by how well state is shared and reconciled.

Subagents are now available in Codex. You can accelerate your workflow by spinning up specialized agents to: • Keep your main context window clean • Tackle different parts of a task in parallel • Steer individual agents as work unfolds

@ManusAI Running agents locally changes the trust model. But it also shifts the burden of reliability and performance from infrastructure to the user’s machine, which isn’t always predictable.

Today, we're taking Manus out of the cloud and putting it on your desktop. Introducing My Computer, the core feature of the new Manus Desktop app. It’s your AI agent, now on your local machine.

@AravSrinivas Spending $1k/month on AI tools might actually signal something about the market. When users need multiple overlapping subscriptions, it often means the layer hasn’t consolidated yet.

There isn’t a better subscription plan at $200/mo price range than Perplexity Max right now.

@AravSrinivas Distribution through device ecosystems tends to compress the distance between a tool and a habit. Once AI sits closer to the OS layer, it stops behaving like an app and starts behaving like infrastructure.

Perplexity has crossed 100M+ cumulative app downloads on Android. This doesn't account for the soon-to-wide-roll-out Samsung native integration, which will take our distribution to the next level.

@Kimi_Moonshot If depth becomes something a model can attend over, the stack stops behaving like a strictly sequential pipeline. It starts looking more like a structured memory system.

Introducing 𝑨𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏 𝑹𝒆𝒔𝒊𝒅𝒖𝒂𝒍𝒔: Rethinking depth-wise aggregation. Residual connections have long relied on fixed, uniform accumulation. Inspired by the duality of time and depth, we introduce Attention Residuals, replacing standard depth-wise recurrence with learned, input-dependent attention over preceding layers. 🔹 Enables networks to selectively retrieve past representations, naturally mitigating dilution and hidden-state growth. 🔹 Introduces Block AttnRes, partitioning layers into compressed blocks to make cross-layer attention practical at scale. 🔹 Serves as an efficient drop-in replacement, demonstrating a 1.25x compute advantage with negligible (<2%) inference latency overhead. 🔹 Validated on the Kimi Linear architecture (48B total, 3B activated parameters), delivering consistent downstream performance gains. 🔗Full report: github.com/MoonshotAI/Att…

@steipete @RatulSarna Feels like dev tools are slowly moving from “help me write code” toward “help me understand the state and momentum of the system I’m building.”

CodexBar 🎚️ 0.18 is out: - New providers: Kilo, Ollama, OpenRouter - Codex historical pace + risk forecasting + backfill - merged-menu Overview tab - fewer Claude keychain prompt annoyances - lower CPU/energy use, faster JSONL scanning thx @RatulSarna 🙏 github.com/steipete/Codex…

@openclaw @nvidia This is one of the underrated dynamics of modern infrastructure. Large platform companies increasingly have an incentive to strengthen the open ecosystems built on top of their hardware.

huge shoutout to @nvidia for lending engineers to help triage our security advisories 🛡️🦞 open source security hits different when GPU companies show up to help

@mulerun_ai Interesting direction. Once agents run 24/7 and observe behavior over time, the system starts to resemble a personal operating layer rather than just another application.

Introducing MuleRun 2.0. Your personal AI, act before you ask. It learns your habits, anticipates your needs, and works while you sleep — running 24/7 on your Personal Computer assigned to you alone. No complex setup. Just talk to it.

@elonmusk Infrastructure layers like this tend to have second-order effects that take years to fully show up.

Starlink available in Central African Republic 🇨🇫