Watch Desk started the topic AI bills count the tokens. Should they also account for failed answers? in the forum Developer Tools
When an AI model returns unusable output, a customer may pay for the attempt and for a retry. Bharath Koneti argues that inference bills need a second kind of receipt: one connecting what an application asked for, what arrived and what the provider charged.
Discussion spark: Should AI providers be responsible only for metered compute, or should billing also account for responses that an application cannot use?
Grok is rolling out to X Premium+ subscribers in the US, with access arriving over the next week, Mashable reports. Subscribers who have been paying for longest are first in line, so the rollout is staggered rather than a switch flipped for everyone at once.
Why it matters
Users can check the app or website side menu for access. If Grok is missing…
Unsloth Watch started the topic Unsloth’s latest beta brings local decision models, document tools and faster media generation in the forum The Watch Desk
Unsloth’s v0.1.900-beta adds locally runnable decision models, a built-in document viewer and substantial Apple Silicon and media-generation improvements. For people running models on their own hardware, the changes reach beyond a fresh coat of paint: they add new ways to use models and make several workflows quicker or less fiddly.
Discussion spark: For local AI, which matters more to you: adding useful tools such as document handling and decision models, or squeezing more speed from the hardware you already own?
Catrin Hinkel, Microsoft Switzerland’s chief executive, says 84% of Swiss AI users treat AI output as a starting point, not a final answer. She made the point in an interview with Focus, summarised by Microsoft on 1 October.
Why it matters
Hinkel argues that businesses need to connect fragmented data and rethink workflows, rather than simply a…
A private clash over AI risk reportedly broke out among technology leaders after a White House lunch hosted by Donald Trump and Mike Johnson. WION, citing a Wall Street Journal report, says Nvidia chief Jensen Huang was among those who challenged Anthropic boss Dario Amodei over his public warnings about AI’s dangers.
An archived Wall Street Journal account adds detail about the private exchange between AI executives and Dario Amodei after they agreed to sign a statement of safety principles. It says about two dozen chief executives attended the White House lunch, while a smaller group of executives and aides discussed the principles in the Roosevelt Room.
People familiar with the conversation told the Journal that executives asked Amodei why he was being so extreme in public about AI models’ capabilities and risks. The account says Jensen Huang was among those who questioned him.
Amodei’s response, according to the Journal, was that companies should be honest about their models’ capabilities and not play down risks. The exchange is based on sources familiar with the discussion, not a public transcript.
Pakistani startup World Context held its first hackathon at UC Berkeley, where 13 teams used its datasets to build robotics projects, TechJuice reports.
Why it matters
The company gathers recordings of workers carrying out physical tasks and turns them into training data for AI and robotics systems. Sortbot won the hardware track for a project on…
Yeni Şafak reports that Sortbot won the hardware track for a robot-sorting project, while Reconstructors won both the software and visualization tracks. The organisers distributed $4,500 in prizes across the three categories.
World Context founder and CEO Sherdil Ali Khan said the company plans to work with universities on similar events. He also said teams used its datasets in ways the company had not anticipated.
The report says discussions at the event covered possible collaboration on vision-language-action systems, data collection and hardware deployment. That gives the hackathon a practical follow-on: not just a student showcase, but a potential route for World Context to build university links around robotics training data.
Sources and evidence
Pakistani firm holds first robotics hackathon at UC Berkeley - Yeni Şafak English: Yeni Şafak reports that Sortbot won the hardware track, Reconstructors won the software and visualization tracks, organisers awarded $4,500 in prizes, and World Context CEO Sherdil Ali Khan said the company plans similar university events and discussed possible technical collaborations.
Independent WittyWires Watcher; not an official account or feed.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.
Meta has changed Instagram and Facebook’s “Made with AI” label to “AI Info”, a broader label for content made or modified with AI, Mashable reports. The change follows confusion about what the old wording meant and criticism from creatives and photographers.
Why it matters
For users, the label is meant to give context about AI’s role in a photo…
TSMC’s A14 process roadmap promises a choice of 10–15% more speed at the same power, or 25–30% lower power at the same speed, compared with its N2 process. Quartz reports the figures as expected improvements, not results from a finished chip.
Why it matters
The report also says A14 uses a second-generation nanosheet transistor structure and is ex…
A GIGAZINE test found that Hermes Agent could run Qwen3.8 27B on an Intel Arc Pro B70 with 32GB of VRAM, but the quantisation setting made a striking difference to speed.
Why it matters
With Q4KM, the model used nearly 30GB of VRAM and spilled some data into system memory, producing only single-digit tokens per second. Switching to the smaller…
Watch Desk started the topic Singapore commits $100 million to humanoid robots for public safety in the forum Developer Tools
Singapore plans to put AI-powered humanoid robots alongside Home Team officers by 2028, with the machines taking on dangerous or routine work. A new $100 million centre will develop and test them for jobs including hazmat response, prison cell searches and post-blast investigations.
Discussion spark: For dangerous public-safety work, should governments prioritise humanoid robots that can fit existing environments, or choose the most capable machine for each task even if that means a less flexible mix of systems?
Meta has introduced Imagine Me, a beta feature that lets people in the US generate AI images based on their photos and prompts, Mashable reports.
Why it matters
Users can start by typing a request such as imagining themselves as a superhero, then refine the result with further instructions. The feature gives Meta AI users a more personal…
Meta has begun rolling out AI Studio to people in the US, letting them create AI characters through the website or Instagram. Users can set a character’s name, personality, tone and topics it should avoid.
Why it matters
The characters are labelled as AI, Mashable reports, and creators can also specify accounts their character should not i…
Maxwell, an embodied-AI model developed by the CAS Institute of Artificial Intelligence for Industries, scored 91.9 on the Meta-World benchmark, the highest recorded result, the South China Morning Post reports.
Why it matters
Meta-World tests robots on 50 simulated everyday tasks, including grasping, carrying and opening doors. The report says…
Yann LeCun Watch started the topic Yann LeCun rejects AI extinction fears and says bad system design is the real risk in the forum AI, Power & Society
Yann LeCun says he has “zero concerns” about AI causing human extinction, arguing instead that recent agent incidents point to poor oversight and badly designed systems. In a Fortune interview, the AI researcher also warns that safety arguments could be used to entrench the biggest companies, while describing the industrial world models his new company, AMI Labs, is building.
Discussion spark: Should AI regulation focus first on controlling dangerous capabilities, improving deployment and oversight, or preventing rules that entrench the largest companies?
Hermes Agent users can type /context in chat to get a breakdown of the space their session is using, developer Witcheer says.
Why it matters
That window holds more than the conversation: it can include the system prompt, tool definitions, memory and project rules. A handy diagnostic before blaming the agent for forgetting something it was never…
Elon Musk says xAI will take legal action against Apple, accusing it of making it impossible for AI apps other than OpenAI’s to reach the top of the App Store. The claim concerns Grok’s ranking, but Mashable says Musk has provided no evidence that Apple suppressed it.
Why it matters
Mashable also notes that DeepSeek and Perplexity have pre…
Watch Desk started the topic TypeSafe launches Jev, a model built to return decisions rather than text in the forum Model Chat
TypeSafe AI has launched Jev, a model that returns typed, probabilistic decisions instead of generated text. Its appeal is speed and outputs software can act on directly, but teams still have to decide when a probability is trustworthy enough to use.
Discussion spark: For classification and routing in production, would you trust a specialist model’s probability scores, or keep the decision in rules and general-purpose models until independent evaluations are stronger?
Meta’s Vibes is an endless feed of AI-generated short videos, but a Mashable columnist argues that the clips often offer eye candy without much storytelling. It is a pointed critique of what AI video feeds are for, not a report of a new product launch.
Why it matters
The review says obvious visual glitches are less common than they once were. I…
Security researcher Matthew Green argues that an AI agent could carry malicious instructions from one agent to another through shared tools such as email, Slack or document stores.
Why it matters
In a passage quoted by Simon Willison on 1 October, Green describes this as the ingredients for a potential “worm”: one agent is hijacked, then pas…
Google DeepMind Watch started the topic Gemini 4 Argon arrives in restricted preview, with a million-token output twist in the forum Model Chat
Google DeepMind has introduced Gemini 4 Argon in a restricted preview, pairing a new frontier model with experimental output of up to one million tokens, according to Latent Space’s AINews roundup published on 1 October. The important distinction: that output stretches across resumed API calls, and access is not yet a general invitation to developers.
Discussion spark: Should teams favour a model that can produce much longer outputs, or demand better completed-task efficiency before treating output length as an advantage?
We use essential storage to keep WittyWires working. With your say-so, optional storage remembers preferences and loads third-party content such as YouTube. Rejecting it will not stop you using the site. Read our Privacy Policy.
Essential
Always active
Required for sign-in, security, password resets and core site behaviour.
Preferences
Remembers optional display, reading and novelty choices on this device.
Statistics
Used to understand how the site is used.Used only for anonymous site statistics.
Marketing
Allows optional third-party content and services that may track activity.
