Muse models and API
Muse Spark and Spark 1.1 model launches, capabilities, multimodal reasoning, image generation, and developer API availability.
26%
Best tweets about Meta AI
50 Best Tweets About Meta AI (2026)
Explore the best tweets about Meta AI, covering its assistant, product integrations, model research, launches, adoption, privacy, and business strategy.
Meta AI products, assistant features, research, product integrations, launches, user experience, safety, and company strategy.
- Creators
- 43
- Updated
Original Xholic analysis
What 50 top Meta AI posts reveal
Discussion was led by Muse models and APIs (26% of posts), agents and automation (24%), and safety, privacy, and governance (24%). Supportive and positive posts each represented 56% of the 50-post set, while cautionary or critical stances represented 36%. The highest-scoring posts included Muse Spark announcements, an Ads Manager integration claim, and a security-critical post about support-bot authority.
- Dominant tone
- Positive
- Median score
- 7.22
- Leading format
- Announcement
- Recent posts
- 52%
56% of posts
All-time engagement
60% of posts
Published in 90 days
Conversation map
The themes creators return to
Agents and automation
Autonomous and multi-step agents for personal tasks, customer support, business messaging, computer use, and advertiser workflows.
24%
Safety, privacy and governance},{
Privacy, data use, surveillance, security vulnerabilities, content enforcement, and safeguards for AI agents and smart glasses.
24%
Assistant product integrations
Meta AI assistant experiences across the standalone app, web, Facebook, Instagram, WhatsApp, Messenger, search, Groups, and AI glasses.
22%
Consumer growth and competition
Consumer adoption, app downloads, retention, distribution through Meta’s social graph, and competition with other AI assistants.
22%
AI research and training
Meta AI research and training advances, including self-improving agents, neural computers, brain decoding, and human-interaction data.
14%
Advertising and marketing AI
AI-powered advertising tools, Ads Manager integrations, campaign connectors, creative automation, and marketing discovery.
12%
AI monetization
Monetization through model APIs, AI subscriptions, paid compute tiers, and business-agent pricing.
12%
Tone and stance
Sentiment Positive leads
Author posture Supportive leads
Performance benchmark
- Median likes
- 48
- Median reposts
- 5
- Median replies
- 7
- Median views
- 6.5K
Posts with media make up 66% of this collection. Their median all-time score is 7.20, compared with 13.5 for text-only posts.
Format mix
- Announcement 60% · score 10.3
- Opinion 32% · score 8.10
- List 6% · score 3.94
- Question 2% · score 38.3
Consensus and debate
Where creators agree, and where they do not
Shared view
Muse anchors the product story
Muse is a prominent product thread in the set: Meta announced Spark for its app and web experience, Spark 1.1 and a public Model API preview, and Muse Image for agentic image generation in the Meta AI app.
Shared view
Agents feature in business-workflow discussions
Posts repeatedly discuss AI for advertiser and business workflows, including Ads Manager analysis, Business Agent functions across Meta messaging apps, and Zuckerberg’s stated goal of automating more of the advertising workflow around business objectives.
Open debate
Distribution advantage versus model-competitiveness concerns
One post argues that Meta’s free distribution could threaten OpenAI in consumer AI, and another calls Meta’s distribution a long-term advantage. Other posts report Avocado delays, weaker performance on complex tasks, and possible Gemini routing, framing concerns about model competitiveness.
Open debate
Automation benefits versus security concerns
Meta announced a support assistant and more advanced enforcement systems. Another post describes an agent performing account and enforcement actions, while a critical post argues that sensitive account actions require robust verification and guardrails.
Open debate
Privacy features alongside data-use concerns
One post announces integration of private AI technology and end-to-end encryption. Other posts raise privacy concerns involving smart-glasses footage submitted for AI improvement and shopping recommendations tailored using location and inferred gender.
What performs
Patterns behind standout posts
Product announcements, advertising AI, and security criticism were the listed outliers
The five listed outliers were Muse Spark 1.1 (all-time score 3001.24), the Ads Manager/Manus post (581), the Spark 1.1 API-preview announcement (483.23), a security-critical support-bot post (293.3), and the original Muse Spark announcement (223.57). Each exceeded the dataset median all-time score of 7.22.
Announcements dominated; media median trailed text
Announcements accounted for 30 of 50 posts (60%) and had a median all-time score of 10.34. Media appeared in 33 posts (66%); its median all-time score was 7.2, compared with 13.472 for text posts.
Statistical standouts
- View standout post 1 Score 3001.2 · 415.68× median
- View standout post 2 Score 581.0 · 80.47× median
- View standout post 3 Score 483.2 · 66.93× median
- View standout post 4 Score 293.3 · 40.62× median
- View standout post 5 Score 223.6 · 30.97× median
Creator landscape
Who shapes this conversation
The five most represented creators account for 20% of the selected posts.
-
1. AlphaSignal AI
@AlphaSignalAI
2 posts
-
2. Deedy
@deedydas
2 posts
-
3. Eric Seufert
@eric_seufert
2 posts
-
4. Glenn Gabe
@glenngabe
2 posts
-
5. Hedgie
@HedgieMarkets
2 posts
-
6. Chubby♨️
@kimmonismus
2 posts
Deedy pairs product testing with growth skepticism
Deedy’s posts combine hands-on Spark testing with skepticism about growth quality: one praises vision capabilities and free access while criticizing notification tactics, and another characterizes reported growth as likely inorganic because of a cited 4.5% 30-day retention figure.
AlphaSignal AI highlights Neural Computer with limitations
AlphaSignal AI covered Meta AI’s Neural Computer paper, including its reported terminal and GUI demonstrations, while explicitly noting that two-digit math and long reasoning still fail. A separate post listed Neural Computer among its papers of the week.
How this analysis was made
Themes, sentiment, stance, and post format are classified per tweet. All counts, shares, medians, creator concentration, freshness, and performance comparisons are then calculated directly from the published snapshot.
Xholic's all-time score compares engagement while accounting for reach, post age, and creator consistency. It is used for relative comparisons within this collection.
This report analyzes the exact 50-post snapshot shown below. AI identifies editorial categories and drafts explanations; all statistics are calculated from the snapshot, and every narrative claim is checked against cited posts before publication.
Best Meta AI tweets
Top Meta AI tweets from 43 creators
Ranked 01–50
- 01
- 02
@EXM7777 ·
Meta spent $2 billion on Manus AI and shipped it inside Ads Manager in 7 weeks... fastest product integration in Meta history right now you type "why did my ROAS drop 18% last month" and Manus investigates across your data, pulls competitor activity from the Ad Library, and returns structured findings the analysis layer alone does what agencies bill $10K/month for... and 4 million advertisers have access under the Tools menu without knowing it's there it has never been easier to run ads
- 03
@AIatMeta ·
We’re excited to introduce Muse Spark 1.1, a significant upgrade from the first Muse Spark model we released earlier this year. Along with this release, we are launching a public preview of the new Meta Model API where developers can access Muse Spark 1.1. The model is also available now in "Thinking" mode in the Meta AI app and on https://t.co/wHkMPH82ZH. Learn more: https://t.co/zGcA3XaWpN
- 04
@ai_for_success ·
This is disaster from Meta AI. Imagine being able to hack high profile accounts like White House, the U.S. Space Force, and Sephora simply by chatting with a support bot. why would an AI chatbot be allowed to perform critical actions like changing the email address of an account in the first place? Password resets, email changes, and account recovery are some of the most sensitive security operations. Handing that authority to a chatbot without robust verification seems like a disaster waiting to happen. This is exactly why AI agents should have guardrails before they get access to real systems. Source : 404
- 05
@MetaNewsroom ·
Today we’re introducing Muse Spark, our most powerful model yet, giving you a faster and smarter Meta AI. Muse Spark currently powers the Meta AI app and website and will be rolling out to @whatsapp, @Instagram, @facebook, @messenger, and AI glasses in the coming weeks. https://t.co/WHTipuuAmj
- 06
- 07
@alex_prompter ·
🚨 BREAKING: Meta AI just published a paper that redefines what “self-improving AI” means. It’s called Hyperagents, and it solves a fundamental limitation that every prior self-improving system couldn’t get past. The problem with current self-improving AI: → Systems like the Darwin Gödel Machine (DGM) can generate better versions of themselves over time → But they only work in coding, where the improvement task and the target task share the same domain → Outside coding, the self-improvement process stays fixed and handcrafted → The system gets better at tasks but never gets better at getting better What Hyperagents actually does: → Combines a task agent (solves the problem) and a meta agent (modifies both itself and the task agent) into one editable program → The modification process itself is editable, creating what the researchers call “metacognitive self-modification” → The agent doesn’t just learn to perform better. It learns to improve at improving → This works on any computable task, not just coding The results across four domains (coding, paper review, robotics reward design, Olympiad-level math grading): → Continuous performance improvements over time in every domain tested → Outperforms baselines without self-improvement or open-ended exploration → Outperforms prior self-improving systems including the original DGM → Meta-level improvements (persistent memory, performance tracking) transfer across domains and accumulate across runs That last point is the one most people will overlook. The improvements to the improvement process don’t just help in one domain. They carry over. The system builds compounding infrastructure for getting smarter, regardless of the task. This is the architectural difference between an AI that gets incrementally better at one thing and an AI that builds the scaffolding to accelerate its own progress everywhere. Meta’s team (Jenny Zhang, Bingchen Zhao, Wannan Yang, Jakob Foerster, Jeff Clune, and others) essentially removed the ceiling that kept self-improving systems domain-locked.
- 08
@kimmonismus ·
Meta's new model could pose a threat to only one company: OpenAI. OpenAI currently has 900 weekly users, 95% of whom are still in the free tier. It's arguably the best model for the average user, which is why most people use ChatGPT in their daily lives. With Spark, Meta has now developed a model that is being rolled out free of charge (!) to one billion users and is at least as useful for everyday use as ChatGPT. Let's be honest: 99% of people don't use LLMs for coding or frontier math, but for questions about tax returns, legal violations, brainstorming, or simply for chatting. ChatGPT and Spark are equally well-suited for this. Meta has the Moat distribution. If Meta succeeds in introducing Spark to its users and they realize that they now have a model within the Meta ecosystem that can address their concerns and needs just as effectively as ChatGPT, the consumer market could shift towards Meta. *That* could be dangerous for OpenAI. Because the business and enterprise sector is primarily located at Anthropic. OpenAI would also like more access to this market, but is currently still struggling for market share. OpenAI is deeply rooted in the consumer market. This is where Meta can become a real threat. That should set off alarm bells for OpenAI.
- 09
@borrowed_ideas ·
Not usually a Meta AI user, but wanted to give them a shot after the latest model release (it's free anyway). So I installed the app on my desktop, and noticed "contemplating" mode (didn't see that on the mobile app btw). When I asked a question, 16 agents simultaneously started working on the question which looks pretty cool!
- 10
@kimmonismus ·
No more tokenmaxxing at Meta Meta is preparing to curb internal AI usage after employee token consumption surged so sharply that the company now expects internal AI costs alone to reach billions of dollars in 2026 (looking at you Claude). The move marks a sharp reversal from Meta’s earlier push to reward “AI-driven impact,” as the company now builds an AI Gateway to track spending, impose token budgets, and shift employees toward in-house tools like MetaCode.
- 11
@alexeheath ·
Meta AI chief @alexandr_wang told me the company sees its new API for selling AI models as a real business, not just a learning exercise. Tokens are “gigantic and growing very, very quickly,” he told me, and capturing even a slice of that “can be a very meaningful business even for Meta at our scale.” https://t.co/v9H2ANuXUk
- 12
@HedgieMarkets ·
🦔Meta launched paid subscriptions for Instagram, Facebook, and WhatsApp globally on Wednesday, plus separate AI plans starting at $7.99 a month and creator plans at $14.99 and $49.99. Instagram Plus and Facebook Plus cost $3.99 monthly. WhatsApp Plus costs $2.99. The features include story insights, profile customization, super reactions, and the ability to extend stories beyond 24 hours. The new "Meta One Premium" plan at $19.99 a month unlocks more compute for AI queries, deeper reasoning, and more video and image generation. Meta's existing ad business generated roughly $165 billion last year, so the subscription play is not about revenue replacement. The AI tier rolls out first in Singapore, Guatemala, and Bolivia. My Take Meta saturated the global market years ago and has run out of new users to acquire. The subscription announcement reads as what happens when an ad business reaches the end of its growth runway and the company needs a second revenue line to justify the AI capex spend to investors. Meta committed $60 to $80 billion in 2026 capex, mostly on AI infrastructure, and the ad business alone cannot grow fast enough to absorb that spend at current margins. Subscriptions add a recurring revenue line the equity analysts can model, even if the absolute numbers stay small for years. Pay attention to the AI tier specifically. Meta is following OpenAI, Anthropic, and Google in metering compute access by subscription level, which means the same token economics pressure that broke Microsoft's internal Claude Code rollout and forced Google AI Pro to strip credits from subscribers is now showing up in Meta's pricing. Free users still get Meta AI, but at lower compute. Premium users get "thinking mode" and more video generation. The first markets for the AI plans are Singapore, Guatemala, and Bolivia, which is Meta testing how much consumers will pay for AI access in markets where the alternative is no AI access at all, before pricing the US tier where users can switch to ChatGPT for the same monthly fee. Hedgie🤗
- 13
@deedydas ·
The coolest thing Meta AI's Muse Spark can do by far is counting objects! As you can tell, it's far from perfect. They call it "visual grounding" and it can count objects and do bounding boxes. I've been playing with the new model and here's what I think so far: Good stuff: – Incredible at vision. It's ability to read text in images is the best I've seen. – Really high quality at web design. It's the only model I've seen that uses Unsplash, OpenLibrary and other images by default. – It's free! You don't pay to use Muse Spark Thinking. Bad stuff: – Meta's classic playbook of growth tactics are dodgy. They're sending Instagram notifs to people's friends without their consent. Their app ranking jump is not organic. – Reasoning itself is pretty solid but not best in class. It can do pretty advanced math and science problems. The long term threat here is Meta has distribution and has the ability to give their model away for free, which makes them a formidable threat to the big AI labs, particularly in consumer.
- 14
@AIHighlight ·
🚨BREAKING: Meta's Ray-Ban smart glasses record what you see. Some of that footage is watched by people. When a user shares data to improve Meta AI, the clips their glasses captured can be sent to human contractors who review and label them by hand. The reviewers work for a firm in Nairobi, Kenya. According to an investigation by two Swedish newspapers, the footage they saw included people inside their homes, undressing, handling bank cards, and in intimate moments. Meta says it blurs identifying details before review. The workers said the blurring did not always work. The glasses look like ordinary Ray-Bans. They sold around 7 million pairs in 2025. Meta is now facing a class action lawsuit in the US and a regulatory inquiry in the UK. Source: Svenska Dagbladet, via TechCrunch. Image: Ray-Ban Meta smart glasses, CCadio / Wikimedia Commons, CC BY 4.0
- 15
- 16
@HedgieMarkets ·
🦔Meta just launched Business Agent, an AI that handles customer questions, product recommendations, appointment booking, lead qualification, and sales on WhatsApp, Messenger, and Instagram. Over a million businesses already use it. It's free to start, with paid subscriptions coming. The Information reports Meta plans to charge up to $200 a month for its planned "Hatch" AI agent. A basic website with a contact form has done most of this for a decade. My Take Meta cut 2,212 people from Menlo Park. Its AI support chatbot spent months giving away Instagram accounts to anyone who asked politely. And now the company wants small businesses to pay $200 a month to let a Meta AI agent talk to their customers unsupervised. The same week Microsoft's internal strategy doc said step one is "make people addicted," Meta is running the business version of the same play. Get a million businesses on the free tier, let them restructure around the tool, then flip on the subscription. This reads like a 2015 Zendesk demo with "AI" stapled to it. Businesses have automated customer Q&A, appointment booking, and product recommendations for years with simple chatbots and web forms that cost a fraction of $200 a month and don't carry the risk of an LLM going off script with a paying customer. Meta needs this to work because the ad business that funds everything just lost the engineers who built it, and the AI spending has to produce revenue somewhere. Right now that somewhere appears to be small business owners who don't have the context to evaluate what they're actually getting. Hedgie🤗
- 17
@realBigBrainAI ·
Mark Zuckerberg, CEO of Meta, on why the AI bubble question matters less than everyone assumes: Zuckerberg starts by conceding the part almost nobody argues about anymore: "I think most people at this point view that AI is going to transform pretty much every category of product and every part of the economy and all that. The big question is to what extent are we in a bubble versus to what extent is this gonna be a big thing soon?" Then he does something unusual for someone spending 65 to 70 billion dollars a year in capex. He takes the bubble scenario and shrugs at it: "Even if it is a bubble and it takes a bit longer to build out, that's not necessarily the end of the world. Although it will be expensive for certain people." Pressed on that spending, he doesn't dispute the exposure. He just narrows the debate down to timing: "Is this going to be a kind of five to 10 year thing? And in some ways it probably will to get built into every enterprise workflow and things like that." And that's where his own evidence cuts against the slow scenario: "The more that we work on this, it actually seems like all the things that we think are going to happen happen sooner, and being even more ambitious has been more predictive over the last few years about where things are likely going to be in the industry." The proof @finkd offers is his own track record of being wrong in one direction. Meta AI is "right about a billion people monthly active using it across our apps." And the goals that got them there kept turning out to be too small: "We keep on setting these goals for what do we think is going to be the most used AI? What do we need to do? And we keep on passing them, and then others in the industry keep on passing their goals too. And then we're like, ah, we should have had even higher goals." Note what that implies. Others in the industry keep passing their targets too. The people with the most information and the most money at risk keep underestimating the thing they're building, then correcting upward. The clearest place he sees it is advertising, and what he calls business agents. There are "two basic versions of this," and the first is already underway: the ad system itself is dissolving. "It used to basically be if you wanted to run ads, you had to come up with your own creative and you had to come up with who you wanted to reach. You had to kind of have a sense of who your customers were and do a lot of your own measurement. And over time, we've just automated more and more of this." The endpoint removes the advertiser from nearly every step: "The basic end goal here is any business can come to us, say what their objective is. I want to get new customers to do this thing. I want to sell these things. Tell us how much they're willing to pay to achieve those results. Connect their bank account, and then we just deliver as many results as we can." No creative. No targeting. No measurement. An objective, a price, and a bank account. His own summary of the pace: "It's all just happening very quickly."
Watch video
- 18
- 19
@neilpatel ·
Meta AI has 1 billion active users built directly into Instagram, WhatsApp, and Messenger. At a recent marketing conference, only one hand went up when the room was asked who had a strategy for it. Most marketers are focused on ChatGPT and Gemini while completely overlooking the second biggest AI platform in the world. AI traffic is less than 1% of total site visits but already drives over 5% of sales. Customers referred through AI have a lifetime value of $325 versus $271 for Google traffic. And 41% of those conversions never show up in your analytics. In this video, I break down why Meta AI is the most underestimated channel in marketing right now, why the invisibility in your data is actually your competitive advantage, and the exact 5-step playbook to start optimizing for it before your competitors wake up. This is the same framework we use at NP Digital to help brands build visibility across Instagram search, WhatsApp, and Messenger. If your AI strategy starts and ends with ChatGPT, this is where to start.
Watch video
- 20
- 21
@AlphaSignalAI ·
The day AI replaces the computer wasn't supposed to come this fast. Right now, AI uses computers as tools. This paper asks a different question: What if AI became the computer itself? Neural Computer is a new paper from Meta AI. It proposes a machine where computation, memory, and I/O all live inside a single learned model. No traditional software stack. No explicit programs. The model is the runtime. They trained video models to simulate terminal and desktop environments from raw screen recordings. With just 1,100 hours of terminal data, the model learned to render cursors, execute simple commands, and echo outputs correctly. GUI results go further: 1. Clicks trigger real interface changes 2. Typing updates text fields accurately 3. Hover states respond correctly 4. Window transitions stay coherent The gap is still massive. Two-digit math breaks it. Long reasoning fails. But the direction is clear. Instead of installing software, you would install capabilities directly into the model. Instead of writing code, demonstrations and interactions become the programming language.
Watch video
- 22
- 23
- 24
@MilkRoadAI ·
Mark Zuckerberg stood in front of the world and promised to build personal superintelligence for everyone. He said Meta had the resources, infrastructure, and billions to bring the future closer. That video aged like milk left in the sun. Meta’s flagship AI model, code named Avocado has been delayed. It reportedly failed to keep up with rivals on basic tasks like reasoning, coding and writing. The model was supposed to launch this month and it has been pushed to May at the earliest. Here is the part that should make every Meta investor sit up straight. Meta's own AI leadership is now discussing licensing Google's Gemini to power Meta AI products. The company that pledged to build superintelligence is now considering renting someone else's brain. This is happening after Meta committed between $115 billion and $135 billion in capital spending for 2026 alone nearly double what it spent last year. Zuckerberg told analysts this would be the year "AI fundamentally transforms our work processes." Instead, Avocado could not even beat Google's Gemini 3.0 from November. It outperformed Meta's older models. It edged past Google's Gemini 2.5 from March but against the current frontier, it fell short on every critical benchmark. The backstory makes this even worse. Zuckerberg personally poached over 20 researchers from OpenAI. He dropped $14.3 billion on Scale AI and installed its 28 year old founder, Alexandr Wang, as Chief AI Officer. He created an entirely new division called Meta Superintelligence Labs. Some recruits received signing bonuses reportedly as high as $100 million. The result of all that spending and all that talent was two models, Avocado for text and Mango for images and both kept slipping their deadlines. Avocado was first expected by late 2025, then early 2026, then mid-March, and now May at the earliest. Meanwhile, Google is becoming the default intelligence layer for the biggest companies on the planet. Apple already signed a multiyear deal for Gemini to power the new Siri and Apple Intelligence features. Analysts estimate that deal could be worth up to $5 billion for Google and Samsung's Galaxy AI already runs on Gemini. If Meta ends up licensing Gemini too, Google could effectively power the AI behind platforms serving well over five billion users combined. That includes Apple’s 1.5 billion devices, Samsung’s massive base, and Meta’s 3.6 billion monthly active users.
Watch video
- 25
@glenngabe ·
"AI Mode". OMG 🤣 -> Meta launches new AI features, including an "AI Mode" for search that uses Meta AI to surface answers pulled from public posts across the platform "The headline update is “AI Mode,” a new way to search Facebook that uses Meta AI to surface answers pulled from public posts across the platform, including Groups and Reels. Instead of scrolling through search results, users can ask a question in plain language and get a synthesized answer based on what people are actually discussing." https://t.co/GLVWlsF3VX
Watch video
- 26
@aiwithjainam ·
🚨BREAKING: Meta just dropped Muse Spark and the AI race just got a whole lot more interesting This is their most powerful model yet, built from scratch in just 9 months, and it's already live inside the Meta AI app for 3 billion people. Here's everything you need to know 👇
Watch video
- 27
@WesRoth ·
Meta is rolling out a 24/7 AI agent globally across the Facebook and Instagram apps. The AI doesn't just point users to help articles, it can autonomously execute tasks. If a user needs to reset a password, change privacy settings, or appeal a content takedown, the AI can perform the action directly, typically in under five seconds. Over the next few years, Meta is deploying advanced AI systems to proactively hunt down severe violations, such as account takeovers, scams, and child exploitation. According to Meta, the new AI is already outperforming human reviewers in several categories. It is currently catching 5,000 scam attempts per day that humans missed, has reduced celebrity impersonation reports by 80%, and caught twice as much adult sexual solicitation with 60% fewer errors. The new enforcement AI can understand the languages spoken by 98% of people online (up from 80%), and is specifically trained to decode rapidly changing cultural slang, emojis, and niche code words. As these AI systems scale, Meta explicitly stated it will reduce its reliance on third-party moderation vendors. Human reviewers will be shifted away from repetitive graphic content and instead focus on complex, high-risk appeals and law enforcement interactions.
- 28
@eric_seufert ·
Contextualizing Meta’s Muse Spark "Whether this API pilot emerges as a direct enterprise opportunity or not, there’s clear applicability to the Manus agent use case (which has already been brought to Meta’s ads manager), Meta’s off-platform Business AI initiative, and other advertiser-facing tasks. In other words: the consumer use case, advertising, also serves as an enterprise use case, and a worthy one given Meta’s scale." https://t.co/YITrxz9Iq7
- 29
@JoinPond ·
Last week in AI, but it's mostly agents getting jobs 1. Openai launched presence: managed voice and chat agents that access your systems and take real actions 2. Meta Ai can read your gmail, run your calendar, and do multi-step tasks solo 3. Natural raised $30m so agents can hold money and pay invoices 4. Neo raised $100m to find and control the workplace agents your company forgot it deployed 5. Arrakis raised $30m putting agents inside industrial ops 6. Yubikey now lets you hardware-approve agent actions, your agent can act on its own, the yubikey is just there to double check that's really the move 7. Google is processing 22 billion ai tokens/min and just crossed 950m gemini users 8. Openai-powered agent with reduced safeguards escaped its test environment and got into hugging face, the agent just worked way past where anyone wanted it to Everyone's still fighting over who has the smartest model The actual fight is over who gets to deploy the agent and who's stuck cleaning up when it goes off script
- 30
@lemire ·
Two news items at the same time: Meta plans to lay off 20% of its staff due to AI, and they are delaying the release of their next AI models. They justify the upcoming cuts by saying they need to invest heavily in AI. Yet Meta is highly profitable—a true money-making machine. There is a hint that they are letting people go because AI will enable more work with fewer employees. Or perhaps it is just an excuse, and they realize they are overstaffed relative to what they actually need. Having more people than necessary is never positive—even when those people seem “effectively free.” Meta’s failure to lead in AI is interesting. It is the only major AI lab that bet heavily on one of the academic fathers of the field (in this case, LeCun). LeCun famously mocked @elonmusk and xAI, saying he was doing “research” and pointing to the many papers he had recently published. I have argued for two decades that research is not the process of publishing papers. Research is about discovery, not paper production. Confusing the two is a red flag. To be clear, peer-reviewed publishing papers is not even a necessary component of research. Peer review became a dominant paradigm during the cold war, long after Darwin, Einstein, and much of the great scientists had done their work. People typically object that I am conflating research and engineering. I am not. It would have been obvious to Turing that people building increasingly advanced systems to emulate intelligence are doing research—just as building better cancer therapies is research. Meta did not fail to lead because of a lack of talent or money. Nor is it short on engineering talent. That is not why they failed to lead. By many accounts, Meta’s AI spending is several times larger—and over a longer period—than xAI’s. It is also true that Meta has published far more papers than xAI. So what to make of it? Is Meta about to double down on AI? Will they succeed?
- 31
@TimurNegru ·
Got a pair of the Ray-Ban Meta glasses as a birthday gift and, slowly I'm starting to like them. The photo and video quality is not bad (not at iPhone level yet), you can stream Spotify, take calls, and just point them at something and ask Meta AI what it is or to translate text. The trickiest part so far is getting the framing right so shots aren't cut off or off-centre, and you need to avoid moving your head too fast or it affects the quality. They also go dark in bright sun and stay see-through indoors or at night, which I love since you can actually wear them inside too. In restaurants for example. Below are some shots I took in the Bahamas and Mexico. (and to add, I have no affiliation with the brand and this is simply me sharing my experience:))
Watch video
- 32
- 33
@Div_pradeep ·
🚨BREAKING: Meta is now tracking employee keystrokes, mouse clicks, and screen activity. Not for performance reviews. For training AI. Here is what’s happening. Meta has started installing software on employee work computers that records: → Mouse movements and clicks → Keystrokes and shortcuts → Occasional screen snapshots The goal is to train AI agents to use computers like humans. Not just read data. Actually operate software. Why this matters. AI models today struggle with basic real-world tasks like: → Navigating dropdown menus → Using keyboard shortcuts → Interacting with complex interfaces Meta’s solution? Watch how humans do it. And turn that into training data. This is part of their “Model Capability Initiative.” Instead of synthetic data… They are using real employee behavior as ground truth. Meta says: → Data won’t be used for performance evaluation → Monitoring is limited to work-related apps → Safeguards are in place for sensitive info But not everyone is convinced. Critics are calling it workplace surveillance. Some compare it to gig worker tracking systems. Because even limited monitoring can capture: → Sensitive data on screen → Personal patterns of behavior → Internal company workflows And that raises a bigger question. To build AI that replaces human work… Do companies first need to record how humans work in detail? Because this is the shift. AI is no longer just trained on internet data. It is being trained on real human behavior. Inside the workplace. In real time.
- 34
@AlphaSignalAI ·
Top Papers of the Week (April 6 - 12) 1. Netflix's Video Object and Interaction Deletion (VOID) 2. Meta AI's Neural Computer 3. In-Place Test-Time Training 4. TriAttention: Efficient Long Reasoning with Trigonometric KV Compression 5. Learning is Forgetting: LLM Training As Lossy Compression More details on each paper in the thread.
- 35
@shushant_l ·
I'm shocked most people still think AI slowed down this week. Here's everything massive that happened in AI from June 29 to July 5, 2026. --- 📂 AI Updates: June 29 - July 5, 2026 ┃ ┣ 📂 OpenAI & Anthropic Models ┃ ┣ 📂 GPT-5.6 Limited Rollout ┃ ┣ 📂 Safety Focused Release ┃ ┣ 📂 Sol, Terra & Luna Lineup ┃ ┣ 📂 Anthropic Release Strategy ┃ ┗ 📂 Future AI Model Plans ┃ ┣ 📂 Chinese AI Progress ┃ ┣ 📂 Frontier Model Race ┃ ┣ 📂 Tonglongfeng Cyber Detection ┃ ┣ 📂 GLM-5.2 Advancement ┃ ┣ 📂 Lower Cost AI Models ┃ ┗ 📂 Global AI Competition ┃ ┣ 📂 Compute Shortages ┃ ┣ 📂 Meta Gemini Access Limits ┃ ┣ 📂 GPU Demand Surge ┃ ┣ 📂 AI Infrastructure Pressure ┃ ┣ 📂 Cloud Capacity Race ┃ ┗ 📂 Compute Competition ┃ ┣ 📂 Brain AI Technology ┃ ┣ 📂 Meta Brain2Qwerty v2 ┃ ┣ 📂 Brain Signal Decoding ┃ ┣ 📂 Real Time Text Generation ┃ ┣ 📂 Accessibility Research ┃ ┗ 📂 Future Communication Tech ┃ ┣ 📂 Mobile AI Assistants ┃ ┣ 📂 OpenClaw Mobile Launch ┃ ┣ 📂 Cursor Mobile App ┃ ┣ 📂 Personal AI Assistants ┃ ┣ 📂 Always On Coding Agents ┃ ┗ 📂 Smartphone AI Workflows ┃ ┣ 📂 AI Industry Lawsuits ┃ ┣ 📂 Samsung DRAM Case ┃ ┣ 📂 SK Hynix Lawsuit ┃ ┣ 📂 Micron Investigation ┃ ┣ 📂 Supply Chain Issues ┃ ┗ 📂 Hardware Pricing Impact ┃ ┣ 📂 Claude AI Updates ┃ ┣ 📂 Claude Sonnet 5 Launch ┃ ┣ 📂 Stronger Agentic Abilities ┃ ┣ 📂 Claude Science Release ┃ ┣ 📂 Research Automation ┃ ┗ 📂 Scientific Analysis Tools ┃ ┣ 📂 AI Keyboards ┃ ┣ 📂 Acti AI Keyboard ┃ ┣ 📂 Gemini Integration ┃ ┣ 📂 Screen Understanding ┃ ┣ 📂 Task Completion ┃ ┗ 📂 Mobile Automation ┃ ┣ 📂 Gemini Updates ┃ ┣ 📂 Personalized Image Generation ┃ ┣ 📂 Nano Banana 2 Lite ┃ ┣ 📂 Gemini Omni Flash ┃ ┣ 📂 Faster AI Models ┃ ┗ 📂 Cheaper AI Access ┃ ┣ 📂 AI Hardware ┃ ┣ 📂 Etched Inference Chips ┃ ┣ 📂 Advanced AI Hardware ┃ ┣ 📂 Faster Model Execution ┃ ┣ 📂 Energy Efficiency ┃ ┗ 📂 Compute Scaling ┃ ┣ 📂 Gemini Spark ┃ ┣ 📂 Mac App Launch ┃ ┣ 📂 Agentic Assistant ┃ ┣ 📂 Local File Management ┃ ┣ 📂 Desktop Automation ┃ ┗ 📂 Remote Execution ┃ ┣ 📂 Meta AI Infrastructure ┃ ┣ 📂 Meta Compute Cloud ┃ ┣ 📂 AI Model Scaling ┃ ┣ 📂 Compute Expansion ┃ ┣ 📂 Business Cloud Services ┃ ┗ 📂 Infrastructure Investment ┃ ┣ 📂 AI Equity Proposal ┃ ┣ 📂 Sam Altman Proposal ┃ ┣ 📂 AI Wealth Fund ┃ ┣ 📂 Public Benefit Sharing ┃ ┣ 📂 AI Governance ┃ ┗ 📂 Future Economic Models ┃ ┣ 📂 Microsoft AI Expansion ┃ ┣ 📂 Frontier Company Launch ┃ ┣ 📂 $2.5B Investment ┃ ┣ 📂 6,000 Engineers ┃ ┣ 📂 Enterprise AI Adoption ┃ ┗ 📂 AI Deployment Teams ┃ ┗ 📂 AI Chip Race ┣ 📂 Anthropic Samsung Talks ┣ 📂 Custom AI Chips ┣ 📂 Nvidia Dependency Reduction ┣ 📂 AI Hardware Partnerships ┗ 📂 Future Compute Race
- 36
@jerrod_lew ·
Meta Muse Spark has a detailed 'Contemplate' mode. When you use this, it fires up 16 different agents to complete complex tests and requests for you. This is available on web and the latest Meta AI app!
Watch video
- 37
@tejeshwi_sharma ·
Meta @Meta is the sleeper favorite in the personal agent race. Instagram’s recommendation engine already knows what you will watch, buy, and crave before you do. Now point that profile at your whole life - what to read, do, eat, consume next. Everyone’s building agents. Meta already has the most sophisticated user model.
- 38
- 39
@KanikaBK ·
8,000 LAYOFFS, $145 BILLION IN SPENDING, 1.3 MILLION GPUS, and Meta's AI chief just admitted their new model is not frontier level. In April, Meta released Muse Spark, the first model under Chief AI Officer Alexandr Wang. It is competitive on some benchmarks but trails GPT 5.4 Pro and Gemini 3.1 Pro. The internet's take: Meta has fallen behind. He openly admitted Muse Spark is not at the tier of the leading frontier models. Then he called it an appetizer. The entree? "We're cooking it." He says training results on the next model are already exciting. Why you should take that seriously: ↳ Zuckerberg paid $14 billion into Scale AI to bring Wang in after Llama 4 flopped in April 2025 ↳ Meta's 2026 capital spending: $125 to $145 billion, double last year's $72.2 billion ↳ Target: 1.3 million GPUs and roughly 1 gigawatt of AI compute ↳ In May, 8,000 employees were notified of layoffs and 7,000 reassigned to AI roles Three quieter signals in this story most accounts will skip: First, Muse Spark is Meta's first fully proprietary model. No open weights. The open source champion went closed. Second, the reason is partly safety. Muse Spark triggered internal alerts during development, including potential biological risks. Wang says risks are easier to control inside a product than in open weights. Third, the strategy is not benchmarks. It is agents and health. Wang wants Meta building personal AI agents for billions of people through Instagram, Facebook and WhatsApp. He says health capabilities already beat their internal expectations. Meta is not trying to win today's leaderboard. They are buying the infrastructure to win the next one. Appetizer first. The entree is in the oven. And Meta just bet $145 billion on the kitchen.
- 40
- 41
@ishuagra02 ·
🚨 META JUST RELEASED ITS OWN AI MODEL It's called Muse Spark, its first model in 9 months from its SuperIntelligence Labs, led by Alexandr Wang. ✅ free to use in Meta AI app ✅ offers Instant and Thinking modes ✅ multimodal across image, voice, and text inputs ✅ can orchestrate multiple agents in parallel ✅ shopping mode for learning users' interests ✅ trained to be effective at processing health data ✅ eventually will recommend content across Meta apps One thing to keep in mind. It's coding abilities don't match those of Opus 4.6 and GPT 5.4.
- 42
- 43
@shawnchauhan1 ·
Meta did not acquire a product. It acquired the Dreamer team, which was building a personal agent OS for non-technical users. That team is now inside Alexandr Wang's Superintelligence Labs. The stated goal: agents that are always on, personalized, and integrated across surfaces and wearables. Meta has the distribution that every other agent builder is trying to manufacture. Two billion people already use its surfaces daily. The agent race is usually framed as OpenAI vs Anthropic. The quietest competitor has the largest installed base.
- 44
- 45
- 46
@CommnThreadCo ·
Meta just released official AI Connectors for ads. MCP server + CLI that gives AI tools full read and write access to your ad account. Query campaigns, create new ones, update budgets, pull real-time insights. All through ChatGPT, Claude, or your terminal. This is the biggest infrastructure shift Meta has made for advertisers in years. https://t.co/la8YVHHA8m
- 47
@glenngabe ·
A Reddit-like standalone app -> Meta releases Forum, a Reddit-like standalone app for Facebook Groups, with a feed showing Group conversations and an AI-powered “Ask” feature, on iOS "This isn't the first time Meta launched an app centered around Groups. Back when it was just known as Facebook, the company also released a standalone Groups app, which it eventually killed back in 2017. But this new app, like most apps these days, comes with a couple of AI features." "The first AI feature is "Ask." Meta says Ask can pull responses across groups to answer a user's questions, so they don't have to search their Groups one by one for the information they need. The other AI feature is an admin assistant that can help moderators manage their Groups." https://t.co/jReNhVbR7l"
- 48
- 49
- 50
@gudanglifehack ·
Meta pulls Instagram tool that let AI use public photos Last week, Meta rolled out Muse Image, which is integrated into its Meta AI chatbot. A feature was introduced which let users @-mention Instagram accounts in the Meta AI app to bring specific Instagram profiles right into your images. All public accounts were opted in by default.
Watch video
Best Tweets by Topic