I=4
Meta releases Meta Motivo, an AI model that is aimed at enhancing Metaverse experience and Meta Video Seal, a model for video watermarking.
“Today, we’re sharing Meta Motivo, a first-of-its-kind behavioral foundation model that controls the movements of a virtual embodied humanoid agent to perform complex tasks,” Meta wrote in a statement.
“Video Seal adds a watermark (with an optional hidden message) into videos that is imperceptible to the naked eye and can later be uncovered to determine a video’s origin. The watermark has proven resilience against common video editing efforts like blurring or cropping, as well as compression algorithms commonly used when sharing content online,” the statement added.
Meta’s AI engineer will arrive in 2025 and will be sort of a mid-level engineer
During the recent interview with Joe Rogan, Zuckerberg said Meta will have the AI engineer in 2025 that will be sort of a mid-level engineer but will be expensive to run at the beginning.
"I think this year, probably in 2025, we at Meta, as well as the other companies that are basically working on this, are going to have an AI that can effectively be a sort of mid-level engineer that you have at your company, that can write code. And once you have that, then in the beginning it will be really expensive to run, then you can get it to be more efficient, and then over time we’ll get to the point where a lot of the code in our apps, and including the AI that we generate, is actually going to be built by AI engineers instead of people engineers, " he said. “But I don’t know. I think that that’ll augment the people working on it. So my view on this is the future, people are just going to be so much more creative, and they’re going to be freed up to do kind of crazy things,” he added. (min 2:08:15).
This comes after Google said last quarter that more than 25% of new code in the company is now generated by AI. Similarly, Amazon said its developer agent had saved the company more than $260 million. Salesforce also said they are not adding any software engineers this year since they have increased productivity Agent Force and other Ai technology that they are using by more than 30%.
Devin, an AI software engineer developed by Cognition AI looks particularly interesting, it is said to be able to perform Upwork developer jobs.
Based on this commentary, it’s likely that the AI engineer will initially replace the mid-level jobs and not the junior and senior developers (min 9:53).
The other interesting AI agent is Open AI’s Operator, which is in beta version. According to a live YouTube test, the agent was able to order stuff from Instacart and only experienced minimal issues.
Assessment
Based on the above, it appears that Meta was late to the party. However, real-world AI engineers already exist, giving lowering the risks of its deployment.
Meta said its Meta AI will now remember things you share with it 1:1 on WhatsApp and Messenger.
Meta is also bringing greater personalization to Meta AI. For instance, it would give suggestions based on where you live, Reels activity and its memory of your family members.
Meta is scrambling ‘war rooms’ of engineers to figure out how DeepSeek’s model is outperforming every other model at a much lower cost, The Information Reported.
Meta AI infrastructure director Mathew Oldham has reportedly told coworkers that the model could outperforms Meta’s upcoming Llama 4 model which is expected to be released in early 2025.
Meta is in discussion to buy South Korea’s AI Chip startup FuriosaAI, a strategy that could enable it to make its own chips and be less dependent on Nvidia, at undisclosed valuation.
Founded in 2017, FuriosaAI has raised $115 million to date, and had reported a 10 times increase in revenue to 3.6 billion won ($2.5 million) in 2023.
It claims that its RNGD chip has three times better performance than Nvidia’s H100 chips.
I=8 Meta is planning a $200 billion AI data center in United States, the Information reported
Meta is planning to construct a $200 billion AI data center in either Louisiana, Wyoming or Texas, the Information reported citing people familiar with the matter.
Meta hasn’t responded to a Reuters request for comment.
I=7 Meta plans to debut standalone Meta AI App and test a paid subscription for it
Meta Platforms plans to release Meta AI app in the second quarter, according to people familiar with the matter.
It also plans to test a paid subscription service for Meta AI.
In January, Zuckerberg publicly agreed with a Threads user who said Meta should create a standalone App for Meta AI.
According to Business of Apps, Meta AI standalone website generates less than 10 million views per month, which is significantly lower than that of GPT or the likes of Anthropic.
I=5 Meta’s business chatbots are facing hurdles, the Information reported citing a current Meta employee
Businesses that have tested Meta’s chatbots have complained that the chatbots can take months to tailor responses to a brand’s voice, the Information reported citing a current Meta employee.
Also, Meta’s chatbots have failed to learn from interactions between human representatives of a business and its customers.
Due to these hurdles, employees working on these business chatbots have been assigned monthly goals instead of the typical half-yearly goals.
According to the Information, Meta has also considered using AI models from DeepSeek in its tools for advertisers in order to provide them with better experience.
In an interview with Morgan Stanley’s Technology, Media and Telecom Conference, Chief product officer Chris Cox said the upcoming Llama 4 model will have what is expected in a new model today such as agentic capabilities and being an omnimodel.
“The second is just the basis of what’s expected in a new model today, which is reasoning. Agentic use cases, just meaning tool use, ability to use a browser, ability to use other tools,” he said. And then the third piece is being an omni-model. So basically interacting with image and voice natively. So rather than translating voice into text and then text into LLMs getting text out, turning that back into speech, having speech be native. This is a big deal."
Regarding when the Llama 4 will be launched, he said “no date forthcoming”.
Clara Shih, Meta’s head of business AI said the company expects more businesses to use AI agents to automate complex tasks (from min 6).
“On the business side, we’re building business AIs from Meta, and we already have these trusted relationships with 200 million small businesses around the world,” she said. They’re using WhatsApp, they’re using Facebook, they’re using Instagram, both to acquire customers, but also engage and deepen each of those relationships. Very soon, each of those businesses are going to have these AIs that can represent them and help automate redundant tasks, help speak in their voice, help them find more customers and provide almost like a concierge service to every single one of their customers, 24x7."
I=5
Meta is planning to introduce improved voice features in its upcoming Llama 4 model, reflecting the company’s bets that the future of AI agents will be conversational instead of text-based, the Financial Times reported citing people familiar with the matter.
Meta turns to distillation process to reduce the cost of training small models, following DeepSeek’s success
Following DeepSeek’s breakthrough, AI companies including Meta are now turning to a process called “distillation” to reduce the cost of training the models.
Here, companies use a large language model to train smaller models.
“We’re going to use [distillation] and put it in our products right away,” said Yann LeCun, Meta’s chief AI scientist. “That’s the whole idea of open source. You profit from everyone and everyone else’s progress as long as those processes are open.”
Meta’s Chief product officer Chris Cox confirmed the other day that they are currently practicing distillation.
“We’ve finished pretraining the smaller model. The main thing – first, just from an intelligence perspective, we’re trying to pack basically, the intelligence of the large Llama 3 series down into really small models, which can then be used with low latency for low-cost on devices on a single host. So basically getting a lot of the intelligence down into the smaller form factor, that’s one of the most important things we can do,” he said.
Open AI believes that DeepSeek used GPT to create its model using distillation process.
Meta combines GPT and Llama in its internal coding tool, signaling limitations of its model despite praises from Zuckerberg
In December last year, the Fortune reported that Meta’s internal AI-powered coding assistant, Metamate, was switching between GPT and Llama depending on the query.
One of the sources told Fortune that Metamate was on par with other AI coding tools such as Microsoft’s GitHub Copilot.
This is despite Zuckerberg saying it was “competitive with the most advanced models and leading in some areas.”
Reports indicate that Meta AI is likely not delivering the best results, this could limit its progress on AI agents
While it’s hard to make an accurate assessment on the trajectory of Meta AI since DeepSeek disrupted the industry and the Llama 4 is yet to be released, current reports suggests that not everything is going on well with it. For instance, there are reports that Meta business AI agents are not delivering as expected and the company is even considering using DeepSeek’s model. The report that Meta’s internal coding agent is using GPT also signals underperformance of Meta’s Llama model. There were also reports that the upcoming model is likely to underperform DeepSeek and Meta’s engineering team was scrambling to resolve this situation. Similarly, during a call with Morgan Stanley, Meta’s chief of product Chris Cox failed to confirm when the Llama 4 will be released despite Mark Zuckerberg having promised that it will be released in early 2025. He only confirmed that they have started using “distillation” process to reduce training costs. However, distillation process will only lead to great results if the parent model is great. While Meta’s relationship with around 200 million small businesses will give it an hedge when it comes to monetizing the AI agents, these agents will need a great model to work efficiently.
Manus, an AI agent that was launched on March 6 has received a lot of hype, with some people calling it the “second Deepseek”. Manus says it uses a combination of existing and fine-tuned AI models which include Anthropic’s Claude and Alibaba’s Qwen, to perform tasks such as buying real estate and programming video games. Its scientist Yichao “Peak” Ji has even said the Manus outperforms Open AI’s deep research and Operator. However, some testers have found it promising but not perfect yet. For instance;
A test by Business Insider established that while Manus structures tasks correctly, it stumbles in execution and even uses fake or simulated data.
Kyle Wiggers, Techcrunch’s said Manus crushed 10 minutes into the first test: " order a fried chicken sandwich from a top-rated fast food joint in my delivery range". On the second test, it didn’t complete the ordering process or provide a checkout link. Manus also provided broken links on the second test which involved booking a flight from NYC to Japan. It also errored when asked to build Naruto-inspired fighting game.
A test by Caiwei Chen of MIT found it to be highly intuitive but with limited scope and high system crashes and server overload. He said in two of the three tasks, it provided better results than ChatGPT DeepResearch, though it took longer time.
“I found that using it feels like collaborating with a highly intelligent and efficient intern: While it occasionally lacks understanding of what it’s being asked to do, makes incorrect assumptions, or cuts corners to expedite tasks, it explains its reasoning clearly, is remarkably adaptable, and can improve substantially when provided with detailed instructions or feedback.” He wrote.
Analyst Dan Salmon of New Street Research believes that the likes of Meta will still have a significant advantage over the likes of Manus due to their existing user base.
“Like other newly developed products from startups, Manus lacks established businesses with billions of users already to whom new AI products can be pushed by Google, Amazon and Meta,” Salmon wrote. “These companies will continue to hold this significant advantage for some time.”
Assessment
In my opinion, the fact that Manus is able to receive such a hype by using the less celebrated Anthropic’s Claude and Alibaba’s Qwen reduces the risks for Meta Platforms. For instance, if Meta is unable to deliver superior performance through its upcoming models, it could use that of Open AI or DeepSeek to build AI agents (while working on its own models) as it has been rumored.
