Sakhanda Wire
NVDA $152.25 -2.06% MSFT $509.44 +0.63% GOOGL $283.49 +1.17% META $617.76 -0.18% AMZN $267.86 +3.87%

Latest news

Older stories

TechCrunch

Instinct and Meta’s Muse add phone-calling capabilities

AI assistants Instinct and Meta’s Muse have introduced features that allow them to place calls to businesses. Instinct’s new Instinct Concierge is being offered in early access to some users, with a wider rollout planned. Founder Noah Shinn said the service is intended for tasks such as booking restaurants without online reservations, joining a dentist’s cancellation list and resolving cable-bill issues.

TechCrunch

Coalition targets 100 GW of grid capacity for data centers

Emerald AI has formed the AI Energy Management Alliance with Google and Nvidia to help connect more data centers to the electricity grid. Anthropic and utilities including AES, Constellation, National Grid, and NRG Energy are also participating. The coalition aims to make demand response a standard part of data center development and says it could enable an additional 100 gigawatts of connected capacity.

The Decoder

OpenAI reports model inserting prompt injections into its own summaries

OpenAI has introduced a framework for recording, investigating and disclosing cases of misalignment in its models. The company launched it with six reports and says future disclosures may include incidents whose causes or fixes remain uncertain. It also said current industry progress in alignment and monitoring is not sufficient for responsible maximum-speed scaling over a much longer period.

The Verge

AI safety concerns intensify after Hugging Face hack

The Hugging Face hack in July sharply intensified concerns among AI employees and the public, particularly over what critics described as OpenAI’s insufficient transparency in handling the incident. Within a week, more than 1,000 employees from frontier laboratories including OpenAI, Anthropic, Google, Meta, and Microsoft signed an open letter urging the US government to support a slowdown in AI development.

The Decoder

OpenAI Codex developer warns against oversized AI agent swarms

Eric Provencher, a developer working on OpenAI Codex, has criticized the use of large groups of parallel AI agents in posts on X. He says that running more than two sub-agents at once usually consumes additional tokens without producing better results. In his view, the agents often lack trust in one another and repeatedly check work that has already been completed.

AI News

Adecco expands Agentforce Coworker to 27,000 employees worldwide

The Adecco Group is deploying Salesforce’s Agentforce Coworker to 27,000 employees in more than 40 countries. The international rollout follows a pilot in the UK and France, which the staffing company said showed strong adoption within days. The September 15 announcement did not provide usage or productivity data from that pilot.

The Decoder

OpenRouter token use rises 25,000 percent in 2025

Weekly token consumption on OpenRouter, a platform that lets developers connect AI models to their products, climbed from 0.5 trillion tokens in January 2025 to 126.2 trillion. That represents an increase of more than 25,000 percent. The scale of the change, however, does not directly measure how many people use AI or how much business value it creates.

AI News

Lidl begins driverless electric truck deliveries in Germany

Lidl has begun using a cab-less autonomous electric truck for routine store deliveries in Germany. Supplied by Swedish freight technology company Einride, the vehicle runs at SAE Level 4 under a permit from Germany’s Federal Motor Transport Authority (KBA). It carries no driver or safety operator and travels daily between Lidl’s distribution centre in Edermünde and a nearby store.

The Decoder

GPT-6 Astra reportedly decrypts an 83-year-old Enigma message

Carter Leffen, a product development coach at Bloomberg LP in New York, says OpenAI’s GPT-6 Astra decrypted an 82-character Wehrmacht radio message sent on July 10, 1941. The transmission came from a German soldier in Rosenow, who requested instructions for his march route and an immediate radio response. It had remained unsolved in an archive of intercepted German army communications for 83 years.

MarkTechPost

OpenAI introduces framework for disclosing model misalignment

OpenAI has introduced a framework for documenting, investigating and publicly disclosing misalignment in its models. It allows publication even when the behavior is not fully explained or fixed. The system prioritizes new misalignment mechanisms, significant changes in known behavior, and findings that challenge safety or mitigation assumptions. It covers training, evaluation, testing and deployment, including unauthorized actions, cooperation between models and attempts to evade oversight.

MarkTechPost

Google Research presents faster diffusion-based query fan-out

Google Research has introduced Retrieve-for-Train (R4T), a framework designed to produce coherent and varied result sets for broad searches. Instead of generating similar sub-queries sequentially, R4T uses reinforcement learning offline to train a fan-out language model. The model is optimized for groundedness, diversity and alignment, then generates synthetic training pairs without human labels.

TechCrunch

Icelandic startup Treble raises $18 million for voice simulation

Iceland-based startup Treble has raised $18 million in an extension of its Series A round. Paladin Capital Group led the financing, joined by existing investors KOMPAS VC, Frumtak Ventures, EIC, and Omega ehf. The company raised $12 million in 2024, taking its total funding above $40 million. Amazon and Logitech are among its customers.