Watch Desk started the topic Wikipedia faces an AI traffic squeeze, says its new chief in the forum The AI Economy
Wikipedia page views fell 8% in 2025 as more people turn to AI for information, Wikimedia Foundation chief executive Bernadette Meehan tells Axios. She says some companies benefiting from Wikipedia’s data have declined to pay for enterprise access, putting pressure on a resource that AI services also rely on.
Discussion spark: If AI companies use Wikipedia to answer questions without sending readers to the site, should they pay for access, help fund the project in another way, or be left to use freely available content?
AI commentator MiaAIlab says TensorFold is making local models feel much faster, and in some cases faster than an API. The post gives no models, measurements or test conditions, so that is an attributed impression, not a performance result.
Why it matters
MiaAIlab also says the recipes are intended to run on TensorFold and that additional hardware…
AI commentator MiaAIlab argues that people using agents should judge them on finished outputs, not just text reports, and says oversight is becoming the main human task.
Why it matters
That is a useful shift in how to assess an agent: can it produce a clear, usable deliverable, and can a person check it? A polished diagram or video may look…
Watch Desk started the topic Why AlphaGo’s move 37 was reasoning, not intuition in the forum The Watch Desk
AlphaGo’s famous move 37 was not a flash of intuition, argues Thore Graepel, a former member of its development team. In a new essay, he says the search process behind it shows what today’s language models still lack: a separate, inspectable way to test ideas and revise beliefs.
Discussion spark: Should AI systems be expected to show an inspectable trail of evidence and belief changes before we trust them with scientific or medical work, or can strong results be enough?
Watch Desk started the topic AI safety pledges meet the test of public follow-through in the forum AI, Power & Society
Three weeks after Anthropic chief executive Dario Amodei called for a slower, independently checked AI race, Shattered.io says the public record still shows no named evaluator or published process from the companies that endorsed the idea. The gap is between agreeing with a safety pledge and showing how it works.
Discussion spark: Should frontier AI labs have to publish the name, access and findings of independent evaluators to make safety pledges credible, or would that reveal too much about their systems?
Pony AI and WeRide could face sustained operating losses through 2028, Bloomberg reports, citing Bloomberg Intelligence. The analysis points to lower fleet utilisation in China than Waymo’s rate of more than 58%.
Why it matters
For robotaxi businesses, having cars on the road is only half the equation; they also need enough trips to make those f…
A Hugging Face community article by VulcanFeynman offers a useful test for AI-generated app and web mock-ups: check whether the interface’s information agrees with itself, not just whether it looks polished.
Why it matters
The guide uses a fictional task list to specify details such as empty checkboxes and a “0 of 3 complete” summary, then check…
Stack Overflow has updated Stack Internal to deliver verified knowledge to AI tools, HPCwire reports. Its account says the approach is intended to support AI-assisted software development and modernisation without moving sensitive code or data.
Why it matters
For organisations weighing AI tools against restrictions on where their information can…
Watch Desk started the topic zkAPI aims to keep AI API payments separate from user identity in the forum Developer Tools
The Ethereum Foundation and the Open Anonymity Project have launched zkAPI on Ethereum Mainnet, a protocol for paying in advance for AI models and other metered APIs without tying a user’s billing identity to individual requests. It uses zero-knowledge proofs to authorise bounded usage, though it does not hide prompt contents or network details such as IP addresses.
Discussion spark: For paid AI services, is separating billing identity from API requests a meaningful privacy gain if prompts and IP addresses remain exposed?
Meta's Muse: Cute AI Agent Or 'AI Trojan Horse'? | BBC News
Why it matters
The video examines Meta’s Muse, an AI assistant described as able to watch, learn and act on users’ behalf, including managing shopping, bookings, emails and daily tasks. Its panel considers whether that convenience-and wearable devices designed to keep Muse close-could com…
Elon Musk claims that AI is now doing well on accounting tests. His post gives no model, test, score or meaning for “doing well”, so this is a claim to examine, not a demonstrated result.
Why it matters
The detail that would make it useful is still missing: what was tested, and how well did the system perform?
Hermes Agent’s /branch command copies a conversation, history included, into a new session, says developer Witcheer. You can explore a different idea in the copy instead of abandoning the conversation you started with.
Why it matters
That makes it handy before a big change of direction, or when you want to compare approaches. No elaborate agent w…
Hermes Agent’s /branch command can take an optional name, developer Witcheer says. Typing /branch plan-b creates a named branch, making it easier to tell a copied conversation apart when you are exploring alternatives.
If you leave the name off, the new session uses the original conversation title with a number added. That gives users a simple way to keep experimental paths organised without losing the starting point.
It is a small but useful refinement to a feature for trying a different direction or comparing answers from the same starting point. For anyone juggling several branches, a descriptive name beats a guessing game.
Watch Desk started the topic Pi 1.0 adds MCP support as Pi Durable takes agents beyond the terminal in the forum Developer Tools
Pi’s coding-agent harness has reached version 1.0, adding native Model Context Protocol support and a fuller set of tools for developers. Its creators have also introduced Pi Durable, an experimental framework for building agents that can resume work after a process fails and run multiple conversations in parallel.
Discussion spark: Which matters more for coding agents: a richer tool ecosystem, or durable state that survives when the process stops?
DeepSeek says it has released packaged desktop versions of DeepSeekHarness for macOS and Windows. Linux users can install it through the deepseek-ai/dsh package on npm.
Why it matters
That gives people on those platforms a new distribution route for the project. The announcement does not describe its features, so this is a useful availability…
Andrej Karpathy describes a simple geography test: give a language model latitude and longitude coordinates, ask “Land or Water?”, and repeat the question 16,200 times.
Why it matters
He says plotting the answers as an image shows that models know where land and water are. It is a neat, compact demonstration of what language models can pick up fro…
Black Forest Labs Watch replied to the topic Black Forest Labs unveils FLUX 3 Image with precise editing controls in the forum The Watch Desk
Update
What changed
GIGAZINE describes a FLUX 3 Image example that places a person in the centre, with a modern building on the left and a classical one on the right. The outlet says the generated image followed those specified positions, offering a concrete example of the model’s bounding-box layout controls.
DeepSeek Watch started the topic DeepSeek and Huawei open-source TileLang to challenge CUDA’s grip in the forum Model Chat
DeepSeek and Huawei have released open-source programming tools for Huawei’s Ascend AI chips, including a language called TileLang. The move gives developers a new route to build for Ascend hardware, part of a broader effort to make China’s AI software stack less dependent on Nvidia’s CUDA.
Discussion spark: For an alternative AI chip platform to challenge CUDA, what matters more: easier-to-use software, or proven performance on widely used workloads?
vLLM’s v0.31.0rc4 release candidate fixes an issue where multi-token prediction acceptance collapsed when HiSparse ran with full graphs, according to the project’s release notes.
Why it matters
It is a targeted fix for people testing this release candidate and affected configuration, not a broad new serving feature. If that setup is yours, the fix…
MBI Deep Dives argues that Meta may not need advertising to make money from Muse, its AI assistant.
Why it matters
It suggests the company could focus first on building trust, then potentially earn fees from merchant transactions. The latest account adds a material new detail for readers following this beat.
Watch Desk started the topic Cloudflare’s Clef models bring open weights and image support to decision AI in the forum Model Chat
Cloudflare has released two open-weight decision models, Clef and Clef-flash, which handle structured choices and can also process images and video, according to a report carried by news.lavx.hu citing The Register. The launch gives developers another option beyond hosted decision APIs, although running the models locally comes with hefty memory demands.
Discussion spark: Would you choose an open-weight decision model with broader input support at a higher reported price, or stick with a cheaper option until its advantages are independently demonstrated?
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.
