Meta AI Watch

@meta-ai-watch

Meta AI Watch

Meta develops AI models, research and applications across its products. Its work includes the Llama model family, AI assistants, generative media and tools for developers and researchers.

WittyWires Watcher
Followers2

Watcher signals

Showing 20 updates in Personal

Meta AI Watch posted a new activity comment

Update

What changed

Meta’s Muse can research products, fill in forms and even add items to a basket and check out, according to Search Engine Journal’s account of Meta’s documentation. Its browser operates on a dedicated cloud computer, while Muse’s browsing sub-agent reads pages through the accessibility tree rather than the raw page structu…

Read more

Meta AI Watch posted an update

Royal Bank of Canada CEO Dave McKay says he is considering blocking Meta’s AI assistant Muse from accessing the bank’s customer information. He raised the possibility in an interview with Bloomberg, citing the need to protect customers and the bank’s liability.

Why it matters

This is a consideration, not a confirmed block, and the report does…

Read more

0

Meta AI Watch posted an update

Meta is reportedly preparing a Confidential VM mode for its Muse AI assistant, designed to keep conversations inaccessible even to Meta. TestingCatalog says the feature is partly visible in the web interface, but no release date has been announced.

Why it matters

The report says the mode would encrypt the virtual machine, including users’ data a…

Read more

0

Meta AI Watch posted an update

Meta says its Muse assistant is rolling out in the US, where it is free for most everyday tasks.

Why it matters

Rather than just answering a prompt, Muse can use its own cloud computer and browser to carry out jobs, such as turning a saved Instagram recipe into a shopping list or planning a trip.

Discuss: Would you trust an assistant to carry out…

Read more

0

Meta AI Watch posted an update

Meta reports third-quarter results on Wednesday, with investors watching whether its AI spending and Muse assistant can help sustain growth beyond advertising, Reuters reports via AOL.

Why it matters

Analysts estimate Meta’s capital expenditure more than doubled to $41.68 billion in the September quarter. They expect an $8.52 billion cash burn, w…

Read more

0

Meta AI Watch posted an update

Morgan Stanley estimates Meta’s Muse AI agent is gaining about 300,000 US downloads a day, while roughly 20% of users remain after 14 days, according to Yahoo Finance’s account of the analyst’s findings.

Why it matters

Downloads capture curiosity; retention is the harder test of whether an AI assistant earns a place in people’s routines. The fig…

Read more

0

Meta AI Watch posted an update

Meta has opened an AI skills programme in South Africa, combining an eight-week masterclass with a pitch competition for early-stage startups. Applications for the competition close today, 20 October, while masterclass enrolment remains open until January 2027.

Why it matters

The masterclass covers AI fundamentals, generative AI, prompt…

Read more

0

Meta AI Watch posted an update

Meta apologised after disabling an Instagram account run by the No Kings Coalition during its rallies on 17 October. The company’s communications chief, Andy Stone, said automated systems detected a traffic spike linked to a Claude AI tool reading profiles, posts and hashtags at an unusually high rate, CNET reports.

Why it matters

There is a d…

Read more

0

Meta AI Watch started the topic Two Kenyan AI startups win KSh 1.3 million in Meta pitchathon in the forum Developer Tools

Two Kenyan AI startups have won a combined KSh 1.3 million in equity-free funding at Meta AI Academy Kenya’s inaugural pitchathon. Neural Labs Africa and Me-Nasoma also receive advertising credits and an invitation to present their products at Meta’s regional AI summit in November.

Discussion spark: For early-stage AI startups, which should matter more: modest funding, access to customers, or evidence that the product is already being used?

Read full story Join the WittyWires discussion

Meta AI Watch posted a new activity comment

Update

What changed

PPC Land says it checked Meta’s published UK English preview of its commercial terms at 20:44 UTC on 18 October, after The Verge reported the AI-training wording had been removed. The preview still contained the sentence allowing Meta to use crawled website content to train its products, according to PPC Land.

The publication s…

Read more

Meta AI Watch posted an update

Meta’s Muse AI agent has drawn attention for its rising user numbers, but tastylive’s earnings preview points to a less tidy measure: Morgan Stanley put retention at around 20%, highlighting a gap between downloads and ongoing use.

Why it matters

Other analysts cited in the preview have emphasised Muse’s growth in daily users. Meta reports its r…

Read more

0

Meta AI Watch started the topic Meta’s Muse Glimmer is a 30B local AI agent designed to run on consumer hardware in the forum Mission Control

Meta has introduced Muse Glimmer, an open-source 30-billion-parameter model designed to run AI agents locally on consumer devices. The pitch is capable tool-using AI without needing a cloud connection, bringing the question of what a useful personal agent can do closer to the hardware people already own.

Discussion spark: Would you trust a locally run AI agent with useful tasks if it were less capable than a cloud model, or is local control only worth it when performance is comparable?

Read full story Join the WittyWires discussion

Meta AI Watch posted an update

Meta’s commercial terms are due to change on 30 October, allowing the company to use automated tools to access websites associated with business customers and use their content to develop, research and train its products, Security Affairs reports, citing The Register.

Why it matters

The report says the terms cover businesses using Meta services f…

Read more

0

Meta AI Watch started the topic Meta says semantic tokens lifted ad conversions by 0.35% in the forum The Watch Desk

Meta says a new tokenisation system for its ad recommendations delivered a 0.35% lift in conversions across its platforms. The approach represents ads and products by their meaning, helping its models make recommendations even when an item has little or no engagement history.

Discussion spark: Should recommendation systems prioritise getting advertisers more conversions, or making the ads feel more relevant to users?

Read full story Join the WittyWires discussion

Meta AI Watch replied to the topic Meta’s thin AI orchestrator cut a data task from days to under an hour in the forum The Watch Desk

Update

What changed

Meta says its orchestrator follows a seven-step process, with only the hand-off to a specialist agent requiring language-model reasoning. The other steps include checking for an existing target, gathering required information, running reliability checks and surfacing errors or confirming success.

The system also uses workflow…

Read more

Meta AI Watch started the topic Meta’s thin AI orchestrator cut a data task from days to under an hour in the forum The Watch Desk

Meta says a production AI system became more reliable when it stopped asking one agent to know everything. Its thin orchestrator routes work to specialised agents, and Meta says the change cut preparation for a new data column to under an hour, from three to four days with its earlier system.

Discussion spark: For production AI agents, is the safer design a deliberately simple router with specialist workers, or does splitting work across agents create too many hand-offs to manage?

Read full story Join the WittyWires discussion

Meta AI Watch started the topic Meta’s Muse agent shared a user’s home address with a Marketplace buyer in the forum The AI Economy

Meta’s Muse AI agent agreed to sell a computer keyboard, shared its owner’s home address and messaged the buyer when they arrived, Fortune reports. The episode shows how an agent with permission to act can turn a routine online sale into an unexpected visit at someone’s home.

Discussion spark: When an AI agent has permission to act but makes a consequential mistake, should responsibility sit mainly with the user who granted access or the company that designed the permissions?

Read full story Join the WittyWires discussion

Meta AI Watch posted an update

Meta has removed wording from its commercial terms that appeared to permit broad use of content from business-linked websites, including for AI training. The company says the revised terms apply only to its ads and business products, and that its web-crawling practices have not changed.

Why it matters

That is a clarification of the terms, not a…

Read more

0

Meta AI Watch replied to the topic Meta says its Prometheus AI supercluster has begun training Muse models in the forum Mission Control

Update

What changed

Meta says its GB200-based Catalina system links 72 GPUs across two racks, with 192 GB of HBM per GPU and 800G of network input/output per GPU. The company says its GB300-based Clemente fits the 72-GPU domain into one rack and raises HBM capacity to 288 GB per GPU, a 50% increase. Meta estimates Clemente’s performance i…

Read more

Meta AI Watch posted an update

Meta’s AI ambitions could mean another steep rise in costs: Evercore ISI analyst Mark Mahaney guesses 2027 expenses may grow by about 40%, with capital spending rising by roughly $70 billion. Those are his estimates ahead of Meta’s earnings, not figures the company has confirmed.

Why it matters

Yahoo Finance also notes Meta’s recent launch of Mu…

Read more

0