I=5
Meta’s MSL and other product teams use vibe coding to speed up app development
“PMs are actually vibe coding products, and we’re showing them to Zuck and leadership, and it’s allowing us to iterate and explore the space really fast,” said Joseph Spisak, a product director in Meta’s Superintelligence Labs (MSL), onstage at the TechEquity AI Summit in Sunnyvale, California.
A post was merged into an existing topic: AI Developments
I=5
Meta strikes several agreements with news publishers allowing Meta AI to access real-time news and events
The news publishers include USA Today, People Inc., CNN, Fox News, The Daily Caller, Washington Examiner and Le Monde.
Like its previous deal with Reuters, the agreements are multiyear deals where publishers are compensated for their content.
I=9
Meta Plans to release Llama’s successor in Q1 2026 but its strategy is causing internal confusion, CNBC reported citing company insiders
Assessment
Meta Platforms shares are down about 1.5% today. The move may partly reflect investor concern about a potential delay or internal uncertainty.
For me, the update is mildly positive because we now know Meta is targeting a Q1 2026 launch for Avocado, and the shift from late 2025 represents only about a one-quarter delay. Considering that this update comes roughly four months after the launch of the Meta Superintelligence Lab, the revised timing doesn’t appear alarming.
Additionally, based on earlier reports suggesting Meta might temporarily rely on third-party AI models, I had expected that Meta’s next frontier model could take longer to arrive. In that sense, receiving a defined Q1 2026 target actually provides more clarity and reduces the risk of an extended delay.
Thanks, I agree with your Assessment.
I=7
Chief AI Officer Alexander Wang has impressed his new colleagues and Meta shifts plans to sell access to its upcoming model code-named Avocado
Assessment
It was already public knowledge that Meta was shifting its AI strategy away from open-source after they started focusing on superintelligence.
The main new insight is the working relationship between Wang and other executives, which seems to be going on well. I had expected internal tensions between Wang and the likes of experienced AI executives like Nat Friedman (who now reports to him).
Micromanagement from Zuckerberg may be something to watch.
I=6
Tensions are emerging between Meta’s core team and Meta Superintelligence Lab (MSL) over training data and compute
I=6
Meta Platforms released SAM Audio for separating any sound from complex audio mixtures using text, visual, or span prompts
Assessment
Initial reviews indicate that the model is impressive. I will continue assessing it (Notion-SAM Audio). I also think that the fact that Meta continues releasing such impressive models despite executive departures at FAIR indicate that the division continues to execute well.
I=6
Meta is developing a new image and video-focused model called Mango, in early stages of exploring developing world-models
https://www.wsj.com/tech/ai/meta-developing-new-ai-image-and-video-model-code-named-mango-16e785c7
I=8
Meta acquires Manus for $2 billion
I=5
Meta’s former Chief Scientist Yann LeCun described Alexander Wang as “young” and “inexperienced,” but noted that he learns fast
“He learns fast, he knows what he doesn’t know . . . There’s no experience with research or how you practise research, how you do it. Or what would be attractive or repulsive to a researcher,” he said.
He predicts more old Meta GenAI employees will leave since Zuckerberg lost confidence in them after the Llama 4 flopped.
“Mark was really upset and basically lost confidence in everyone who was involved in this. And so basically sidelined the entire GenAI organisation. A lot of people have left, a lot of people who haven’t yet left will leave.”
I=4
China is reviewing Meta’s $2 billion deal to buy Manus
I=6
Meta Platforms signed deals to acquire more than 6 gigawatts of electricity, costing at least $5 billion per year
Where are those calculation coming from? If it’s your own place always make them public and link them in any post.
I=5
Meta establishes Meta Compute, a top-level initiative for AI infrastructure
Zuckerberg said they plan to build tens of gigawatts this decade and hundreds of gigawatts over time.
According to Jefferies senior analyst Brent Thill, the hundreds of gigawatts that Meta plans to build will cost at least $400 billion (min 0:30). He added that although Meta is currently the most hated internet stock, he likes what Meta is doing.
Is this an acceleration of previous ai spend plans or just a reiteration of it? If it’s an acceleration by how much and does it change assumptions of the valuation model?
I think it would be cool to build a timeline of key spend announcements in the right Notion page + link it here, to have an easier overview of key developments and make judging the topics seamless.
I=4
Former Meta’s head of Generative AI, Ahmad Al-Dahle, departs to Airbnb
Under his leadership, Meta’s Llama 4 model significantly underperformed.
He seems to be blaming Meta for focusing on powerful models only and not people’s experiences?
Yes possibly. I think the post is also potentially another small sign that the new model will not be open source
I=9
Meta’s Superintelligence Lab has delivered its first key models internally this month, CTO Bosworth said
He added that the models were “very good”, Reuters reported citing conversation with Bosworth.
Bosworth didn’t specify which models have been delivered internally but said they are not yet finished.
“There’s a tremendous amount of work to do post-training” for AI, “to actually deliver the model in a way that’s usable internally and by consumers,” he said, speaking in general terms about the development cycle.
The Wall Street Journal reported in December 2025 that Meta was working on a text AI model codenamed Avocado and a new image and video-focused model called Mango, which are slated to be released in Q1 2026.
