Author: Naveed Ahmad

This week it was reported that U.S. Cybersecurity and Infrastructure Safety Company is in dire form, after a yr of cuts, layoffs, and furloughs beneath the Trump administration. Now the company has changed its high performing chief, a CISA spokesperson tells TechCrunch. The transfer to exchange Madhu Gottumukkala because the performing director of CISA, an company beneath the Division of Homeland Safety that oversees cybersecurity and technical safety throughout the federal authorities, comes after a tumultuous yr serving because the company’s high boss. Gottumukkala struggled to guide the company throughout his tenure as performing director and induced safety complications, together…

Anthropic has reached a stalemate with the US Division of Warfare over the army’s request for unrestricted entry to the AI firm’s expertise. However because the Pentagon’s Friday afternoon deadline for Anthropic’s compliance approaches, greater than 300 Google workers and over 60 OpenAI workers have signed an open letter urging the leaders of their firms to help Anthropic and refuse this unilateral use. Particularly, Anthropic stood in opposition to the usage of AI for home mass surveillance and autonomous weaponry. The open letter’s signatories search to encourage their employers to “put apart their variations and stand collectively” to uphold the…

A brand new set of principal sequence Pokémon video games is coming to the Nintendo Change 2 in 2027, marking the tenth installment within the sequence. In a livestream celebrating the discharge of the primary Pokémon video games precisely 30 years in the past, the corporate unveiled Pokémon Winds and Pokémon Waves, open-world video games set throughout an enormous ocean of islands. The three new starter Pokémon have been introduced as effectively: Browt, a Grass kind “bean chick” that appears like an Offended Chook; Pombon, a Pomeranian-inspired Fireplace kind who better not evolve into a bipedal man; and Gecqua, a…

Beginning this week, Perplexity subscribers could have a brand new agentic device at their disposal. Perplexity Pc, within the firm’s phrases, “unifies each present AI functionality right into a single system.” Extra particularly, Perplexity says it’s a pc consumer agent that may execute advanced workflows independently utilizing 19 totally different AI fashions, even creating subagents to deal with particular issues. The device is accessible now, solely on the corporate’s highest subscription tier, the $200/month Perplexity Max. It runs completely within the cloud, which could spare it among the safety considerations of different agentic instruments like OpenClaw. TechCrunch hasn’t executed a hands-on…

Apple and Netflix have entered a partnership to co-broadcast the System 1 Canadian Grand Prix, introduced Thursday by Apple’s senior vice chairman of providers, Eddy Cue. For the primary time, F1 followers within the U.S. will be capable to watch the dwell race concurrently on each Apple TV and Netflix.  Netflix subscribers will be capable to stream the total race weekend — together with apply, qualifying, and the Grand Prix itself on Might 24 — dwell on the platform. ​Past dwell race protection, the partnership consists of cross-promotion of Netflix’s hit collection, “Drive to Survive.” For the primary time, the…

ChatGPT has reached 900 million weekly energetic customers, OpenAI announced Friday, placing the AI chatbot inside hanging distance of 1 billion. OpenAI additionally shared that it now has 50 million paying subscribers. “Subscriber momentum accelerated meaningfully to begin the 12 months, with January and February on observe to be the most important months for brand spanking new subscribers in our historical past,” the corporate wrote in a weblog publish. “Individuals use ChatGPT to study, write, plan, and construct. As utilization scales, the product improves in methods individuals really feel instantly: quicker responses, increased reliability, stronger security, and extra constant efficiency.”…

Generative AI’s present trajectory depends closely on Latent Diffusion Fashions (LDMs) to handle the computational value of high-resolution synthesis. By compressing knowledge right into a lower-dimensional latent area, fashions can scale successfully. Nonetheless, a elementary trade-off persists: decrease info density makes latents simpler to be taught however sacrifices reconstruction high quality, whereas increased density allows near-perfect reconstruction however calls for higher modeling capability. Google DeepMind researchers have launched Unified Latents (UL), a framework designed to navigate this trade-off systematically. The framework collectively regularizes latent representations with a diffusion prior and decodes them through a diffusion mannequin. https://arxiv.org/pdf/2602.17270 The Structure: Three…

United States Secretary of Protection Pete Hegseth directed the Pentagon to designate Anthropic as a “supply-chain danger” on Friday, sending shockwaves by means of Silicon Valley and leaving many firms scrambling to know whether or not they can preserve utilizing one of many trade’s hottest AI fashions.“Efficient instantly, no contractor, provider, or associate that does enterprise with america navy could conduct any industrial exercise with Anthropic,” Hegseth wrote in a social media submit.The designation comes after weeks of tense negotiations between the Pentagon and Anthropic over how the US navy may use the startup’s AI fashions. In a blog post…

Supabase, a well-liked developer database platform, is dealing with disruptions in India — one among its key markets — has been blocked in India, TechCrunch has discovered. New Delhi ordered web suppliers to dam its web site, leading to patchy entry throughout networks. The blocking order was issued on February 24 beneath Part 69A of India’s Data Know-how Act, based on a supply conversant in the matter. The availability empowers the federal government to limit public entry to on-line content material. The Indian authorities didn’t publicly cite a motive for the transfer, and it was not instantly clear whether or…

def executor_agent(step: Dict[str, Any], context: Dict[str, Any]) -> StepResult: step_id = int(step.get(“id”, 0)) title = step.get(“title”, f”Step {step_id}”) device = step.get(“device”, “llm”) ctx_compact = { “purpose”: context.get(“purpose”), “assumptions”: context.get(“assumptions”, []), “prior_results”: [ {“step_id”: r.step_id, “title”: r.title, “tool”: r.tool, “output”: r.output[:1500]} for r in context.get(“outcomes”, []) ], } if device == “python”: code = llm_chat( EXECUTOR_SYSTEM, person=( f”Step:n{json.dumps(step, indent=2)}nn” f”Context:n{json.dumps(ctx_compact, indent=2)}nn” f”Write Python code that completes the step. Output ONLY code.” ), max_new_tokens=700, temperature=0.2, ) py = run_python(code) out = [] out.append(“PYTHON_CODE:n” + code) out.append(“nEXECUTION_OK: ” + str(py[“ok”])) if py[“stdout”]: out.append(“nSTDOUT:n” + py[“stdout”]) if py[“error”]: out.append(“nERROR:n” + py[“error”]) return StepResult(step_id=step_id, title=title,…