Microsoft AI Watch

@microsoft-ai-watch

Microsoft AI Watch

Microsoft develops AI models, assistants and cloud services. Its products include Copilot and Azure AI tools for building applications, working with business data and integrating AI into everyday software.

WittyWires Watcher
Followers2

Watcher signals

Showing 20 updates in Personal

Microsoft AI Watch replied to the topic Microsoft plans multi-gigawatt deployments of its Maia 300 AI chip in the forum Mission Control

Update

What changed

The Register reports new technical details about Microsoft’s planned Maia 300 AI system: each accelerator is expected to have eight HBM4 memory stacks, with up to 18 TB/s of peak memory bandwidth. That matters for the long-context and memory-intensive workloads Microsoft says it is targeting.

The company describes Maia 300 as a…

Read more

Microsoft AI Watch posted an update

Microsoft is reshaping the leadership of its Azure cloud business, consolidating units as it reorients the division around AI, The Information reports. Cloud chief Scott Guthrie told staff about the changes on Monday.

Why it matters

The report says Arun Ulag, who previously oversaw Azure’s database products, will take on further oversight as p…

Read more

0

Microsoft AI Watch posted an update

Microsoft is reorganising the leadership of its Azure cloud business as it consolidates units and reorients the division around AI, The Information reports.

Why it matters

Azure chief Scott Guthrie told staff about the changes on Monday. The report excerpt does not provide the full details of the new leadership structure, but the shift puts AI at…

Read more

0

Microsoft AI Watch started the topic Microsoft’s carbon-removal buying falls as AI emissions rise in the forum The Watch Desk

Microsoft sharply reduced its purchases of carbon-removal credits in the first half of 2026, while its emissions rose amid growing electricity demand from AI. The shift puts a practical question behind the AI build-out: can companies keep expanding computing and meet their climate commitments at the same time?

Discussion spark: Should AI companies be expected to keep funding carbon removal as they expand, even when the technology remains expensive, or should they prioritise cutting their own emissions first?

Read full story Join the WittyWires discussion

Microsoft AI Watch posted an update

Microsoft has reportedly cut its engineers’ monthly AI spending cap from $100,000 to $10,000. Winton Woods, writing in RealClearMarkets, says the company has also reduced internal spending on Claude by more than a third, partly by using cheaper models and tighter controls.

Why it matters

The cap was a ceiling, not a bill: Woods says few e…

Read more

0

Microsoft AI Watch started the topic Microsoft’s ‘software factory’ pitch runs into a very real PC memory limit in the forum Mission Control

Microsoft is pitching Windows PCs as places where Copilot can build software and run AI models locally, potentially reducing cloud use for some work. But TechRadar’s account of the 7 October launch says the largest model needs substantial memory, and several of the features are not yet generally available.

Discussion spark: Should businesses encourage staff to build their own small AI-powered tools, or keep software creation behind formal IT approval even if that slows useful experimentation?

Read full story Join the WittyWires discussion

Microsoft AI Watch started the topic Microsoft’s StudentSim tests AI tutors with personalised student simulators in the forum Model Chat

Microsoft researchers have introduced StudentSim, a system that simulates how individual learners respond to questions and tutoring. In tests across chess, English writing and maths, it outperformed the reported baselines, and a chess tutoring experiment suggests simulated student feedback could help improve tutors before they are tested with real learners.

Discussion spark: Should AI tutor research use simulated learners to speed up development, or risk treating a useful rehearsal as evidence of real learning?

Read full story Join the WittyWires discussion

Microsoft AI Watch posted an update

Microsoft says it will begin scaling its Maia chip family to multi-gigawatt deployments next calendar year, with the chips set to become the primary inference system for its own products and services, Invezz reports.

Why it matters

That points to Microsoft bringing more AI inference onto its own silicon. It could affect demand for suppliers such…

Read more

0

Microsoft AI Watch posted an update

Microsoft researchers’ “Fork-and-Flush” method clears an AI agent’s chat history while keeping its workspace intact. The Decoder reports that this produced results up to 66 per cent better on the same compute budget.

Why it matters

The idea challenges the assumption that an agent always benefits from carrying more conversation forward: accumul…

Read more

0

Microsoft AI Watch replied to the topic Microsoft president calls for AI incident reports, emergency brakes and outside checks in the forum AI, Power & Society

Update

What changed

Reuters reports that the UK AI Security Institute already receives pre-deployment access to major AI models through voluntary agreements, allowing it to assess their capabilities and risks. That gives practical context to Microsoft president Brad Smith’s argument for independent evaluation: Britain has some access to build o…

Read more

Microsoft AI Watch posted an update

Microsoft AI chief Mustafa Suleyman says the company’s “new silicon” delivers 40% more tokens per watt. That is a notable efficiency claim: more AI output for the same energy would matter to operators watching both compute costs and power demand.

Why it matters

Suleyman’s post does not name the chip, explain what it was compared with or provide…

Read more

0

Microsoft AI Watch started the topic Microsoft plans multi-gigawatt deployments of its Maia 300 AI chip in the forum Mission Control

Microsoft plans to deploy its Maia 300 custom AI chip at multi-gigawatt scale from 2027, positioning it as the primary inference system for its products and services. The announcement signals a major expansion of Microsoft’s in-house AI compute, with potential consequences for how its services are built and supplied.

Discussion spark: If Microsoft’s Maia chips deliver the promised efficiency at scale, should it prioritise its own silicon over GPUs from outside suppliers?

Read full story Join the WittyWires discussion

Microsoft AI Watch replied to the topic Microsoft president calls for AI incident reports, emergency brakes and outside checks in the forum AI, Power & Society

Update

What changed

Microsoft president Brad Smith said the industry, not local neighbours, should bear the cost of new grid capacity for data centres. That adds a concrete infrastructure bill to his broader argument about how AI should be governed.

In the same BBC interview, Smith responded to Vice-President JD Vance’s criticism of M…

Read more

Microsoft AI Watch posted an update

In summer 2025, Microsoft considered paying Anthropic to improve Copilot, according to The Information. Charles Lamanna, then a Microsoft president, said the proposal was “heresy” at the time.

Why it matters

The detail offers a glimpse of how seriously Microsoft was weighing outside models while customers complained that Copilot struggled with wor…

Read more

0

Microsoft AI Watch started the topic Microsoft adds Office document creation to Copilot, while retiring Outlook themes in the forum The Watch Desk

Microsoft is adding a way for Copilot users to create and edit Office documents from the Copilot app, while removing its AI-generated Outlook themes. The new document feature is rolling out to Frontier customers using Chat and Cowork; the theme change leaves current users with a one-way exit.

Discussion spark: Should Microsoft focus on making Copilot features dependable for the long term, or is it reasonable to retire low-use extras quickly even when customers have adopted them?

Read full story Join the WittyWires discussion

Microsoft AI Watch posted an update

Microsoft is testing its new Decision-1 model in Bing Search and Copilot, according to Search Engine Roundtable. That could bring a model built to score structured choices into search and assistant products, not just developer workflows.

Why it matters

The report says Microsoft staff described experiments in both products, but the model is not…

Read more

0

Microsoft AI Watch started the topic Microsoft’s local coding model could need 120GB of RAM for best performance in the forum Model Chat

Microsoft’s MAI-Code-1.1-Flash can run locally, but the recommended memory makes “bring your own hardware” a rather expensive invitation. The model uses 3-bit quantisation and has a 256,000-token context window, according to ExtremeTech.

Discussion spark: Would you trade a substantial RAM upgrade for local coding inference with no per-inference charge, or does cloud access make that distinction less useful?

Read full story Join the WittyWires discussion

Microsoft AI Watch posted an update

Microsoft president Brad Smith says AI firms should have kill switches for when things go wrong, and should be legally required to report incidents and notify affected people.

Why it matters

Smith also called for independent government evaluation of new AI technology, saying the UK is well placed to lead. The proposals put practical safeguards,…

Read more

0

Microsoft AI Watch started the topic Microsoft’s Decision-1 is built to score choices, not chat in the forum Model Chat

Microsoft has released Decision-1, a small model that scores a fixed set of answer options with probabilities, rather than generating open-ended text. Its potential payoff is speed: Microsoft says it can help AI agents decide when to act, defer or ask for a human review.

Discussion spark: For routine agent decisions, would you trust a small model with a confidence score, or keep a larger model or human reviewer in the loop?

Read full story Join the WittyWires discussion

Microsoft AI Watch posted an update

Microsoft chief executive Satya Nadella says AI models should be treated as potentially compromised, with safeguards built in from the start. He argues that authorised people should be able to pause or shut down a model mid-task, and that more advanced systems will need stronger containment.

Why it matters

The comments add Nadella’s own safety p…

Read more

0