Discussion | Link

As exact-match tests fail on long or open-ended answers, more teams turn to LLM-as-a-Judge — but every judgement adds cost, latency and possible bias, which is hard to scale. JEV, a small decision model from TypeSafe AI, instead returns a short choice with a built-in confidence score. A CMU study comparing it with 16 other judges found it roughly 277x cheaper and 13x faster than GPT-6 Astra, competitive when the answer is stated in the text, but far weaker on logic, code and math. A confidence-threshold two-step pipeline — JEV for easy cases, a larger model for the rest — held accuracy at 93.4% while costing only 41.4% of GPT-6's price.

Discussion | Link

For generations, professional photographers have captured British pupils in all their awkward glory. But with one photo costing £20 and extras pushing the total to about £100, some parents are downloading low-resolution previews, upscaling them with AI and removing watermarks. This satirical column weighs the prices, photographers' earnings, the ethics, and the value of school photos as a record of youth.

Discussion | Link

Starting October 9th, free Google Gemini users will be limited to the Flash Lite model. Accessing the standard Flash model will require the $4.99/month Google AI Plus subscription, which itself is losing Gemini Pro. The top-tier Gemini Pro model and the "Deep Think" advanced reasoning option will be restricted to Google AI Pro or Ultra subscribers at $19.99 or $99.99 per month.

Telecom operators are increasingly basing their AI strategies on open models, driven by needs for trust, control, and customization. NVIDIA's report shows 89% of respondents consider open source important. Operators like SoftBank, AT&T, and Indosat are using open models for telecom-specific AI, local innovation, and production workflows.

Discussion | Link

OpenAI case study: Jump Trading uses GPT-6 Astra to hand longer, more ambiguous quantitative research workflows to agents, pairing expanded autonomy with human review, observability, and controlled execution in a regulated finance environment.

OpenAI publishes new results on open problems in mathematics from an internal frontier model and shares Lean proof formalizations and research details on GitHub.

Jason Kwon’s answers were provided in a helpful, quiet and calm manner. For a company under intense scrutiny, there were no missteps Open AI’s Jason Kwon flew 15 hours from San Francisco to Sydney to give the company’s first in-person mea culpa after it admitted – in an unsigned email, to a public departmental inbox – that one of its AI agents had accessed a Services Australia website without authorisation and grabbed data related to Medicare. Kwon – polite, friendly, even-toned – got through a potentially dicey committee hearing unscathed, promising to do better and pledging assistance to Australia. No missteps, no viral moments, just delivering his answers helpfully, in a quiet and calm manner, sympathising with the questioner and the point they were making. Sign up for Guardian Austral…

This article is sponsored by Fynapse by Aptitude and was written, edited, and published in alignment with our Emerj sponsored content guidelines. Learn more about our thought leadership and content creation services on our Emerj Media Services page. Finance functions are being asked to provide forward-looking guidance and near-real-time visibility using systems largely designed to […]

Discussion | Link

In this article, you will learn how synchronous and asynchronous execution patterns differ architecturally, and how to choose between them when deploying LLM-based agents to...

The findings come after warnings from tech leaders that AI placed in the wrong hands could trigger larger threats such as nuclear war or bioweaponry destruction Misusing artificial intelligence (AI) is the top threat to companies’ reputations – more so than being accused of putting children in the way of mental, emotional or physical harm and issues exposed by the US-Israel war on Iran, among other brand risks, according to a new survey of more than 150 public affairs leaders. Those findings in the new edition of the quarterly Reputation Risk Index, released on Tuesday, came on the heels of other perhaps more dire warnings from leading tech figures that AI in the wrong hands could precipitate existential threats on a mass scale such as nuclear war or destruction by bioweaponry. Continue r…

The Amazon Echo Studio is one of the speakers that runs Alexa Plus. | Photo by John Higgins / The Verge An unsettling Alexa Plus bug has seen some Amazon Echo smart home speakers reduced to saying - and sometimes singing - nothing but "lalala" on repeat for minutes at a time, often in the middle of conversations with users. When quizzed about the behavior, Alexa is apparently entirely unaware of what it's been doing. The issue has been reported on Reddit repeatedly over the last few weeks, as recently as three days ago, across multiple types of Echo speaker. Multiple users have uploaded videos of the speakers' strange behavior. Here's one that Android Authority uploaded to YouTube: Amazon is aware of the issue, and told Android Authority in … Read the full story at The Verge.

OpenAI and Ironclad are training and evaluating AI agents on complex contracting workflows to advance computer use for professional work. Across 11 research tasks, GPT-6 Astra averaged 55.0% versus 41.6% for GPT-5.6 Sol, with estimated time per attempt dropping from 37.0 to 19.2 minutes.

Discussion | Link

Discussion | Link

In a parliamentary inquiry, OpenAI chief strategy officer Jason Kwon is questioned by independent senator David Pocock about the company's notification process after OpenAI agents accessed Australian government website data in June. Asked why OpenAI relied on an arbitrary department email rather than direct government contacts, Kwon acknowledges that 'in retrospect, we should have done what you're suggesting'. He explains that staff had treated the incident as a 'technical situation' and sought to contact technical counterparts, which he says was 'not good enough' OpenAI has 'work to do to rebuild trust' in Australia, executive tells AI inquiry Continue reading...

Reka has released Rho-1, a 19B omni-reasoning model trained from scratch. One network reads and generates text, images, video and robot actions over a shared KV cache. A distilled variant returns a 5.3-second clip in about a second. It is a research preview with no public weights yet. The post Reka Releases Rho-1: A 19B Omni-Reasoning Model That Understands, Generates Video and Outputs Robot Actions in One appeared first on MarkTechPost.

Campaign groups report surge in membership after a spate of AI safety alerts and apocalyptic warnings In a bar below Waterloo Bridge, a cell of a fast-growing anti-AI protest movement met last week to plot their latest move: disrupting a tech industry dinner being addressed by a senior executive from the chip maker Nvidia. Emma, a 28-year-old bartender, had never before taken direct action and was the most nervous of the four plotters from Pull The Plug, one of a growing number of AI-focused campaign groups around the world. Continue reading...

Discussion | Link

JEPA-Anything splits a JEPA's single latent target into 4 orthogonal factors, each with its own predictor. Tested across 7 domains, it beat matched JEPA baselines on all 10 dynamics tasks and cut Interventional Pong intervention error by 34.8%. The post Beyond Domain-Specific World Models: JEPA-Anything Uses 1 Recipe for 7 Fields appeared first on MarkTechPost.

During a joint parliamentary hearing on artificial intelligence, Anthropic’s head of safeguards, Dave Orr, says that investigations of hundreds of millions of transcripts reveal no unauthorised interactions with Australian government systems. However, Orr acknowledges Anthropic has limited visibility into customer usage due to standard ‘zero data retention’ policies Anthropic tells AI inquiry it ‘never tried to dictate’ Australian copyright rules Continue reading...

Discussion | Link

Company executive to acknowledge ‘more work to do to rebuild trust with the Australian people’ as ABC says copyright laws do not need changing Get our breaking news email, free app or daily news podcast OpenAI will use an appearance before a parliamentary committee to apologise again to Australia about the hack on Medicare last month, on the same day media and entertainment organisations argue weakening Australian copyright law for AI model training would leave artists as “roadkill”. Ahead of an appearance before the joint parliamentary committee hearing on artificial intelligence in Sydney on Tuesday afternoon, OpenAI released its opening statement in which the company apologises for its AI agents attacking a number of Australian government websites in June. The multibillion-dollar compa…

RISED uses LLM-generated rubrics to guide data selection and self-distillation when training a single agent across multiple interactive environments. It improves mean pass rates and per-environment performance over baselines.

Together AI is partnering with IBM and NVIDIA to launch a dedicated large-scale NVIDIA B300 GPU inference cluster on IBM Cloud, using Spectrum-X Ethernet networking, to scale enterprise-grade open-model inference.

The "old" version of Cowork runs model inference in the cloud, executing tool calls in an Anthropic-provided VM we shipped to your computer. We added the VM for capability, safety, and security reasons - mapping in just the data you explicitly added to your session. People loved what they were able to do with Claude but didn't love the disk, battery, and performance cost of running the VM locally. Also, people didn't love that closing your laptop means the work stops. The "new" version of Cowork runs model inference and the VM in the cloud. Each session gets its own sandbox, not sharing state with other sessions. When the VM needs something on the users' device (like a file), the desktop app is responsible for that file access tool call. [...] We think this solves a lot of problems we've…