Meta AI News & Updates - Latest AI Coverage
241 articles • Page 1 of 13
Meta runs the largest open-weights AI research effort of any major tech company. The Llama family — Llama, Llama 2, Llama 3, Llama 4 — is downloadable, modifiable, and free to use commercially up to a usage threshold, which has made it the default base model for thousands of fine-tuned derivatives on Hugging Face. Meta also builds AI features into Facebook, Instagram, WhatsApp, and Messenger (AI Studio, Meta AI assistant), and ships specialized models like Segment Anything (computer vision) and AudioCraft (audio generation). Research output runs through FAIR (Fundamental AI Research). Stories cover Llama releases, FAIR papers, AI integration into Meta's consumer apps, the Ray-Ban Meta smart glasses, and Mark Zuckerberg's recurring statements about AGI timelines.
OpenAI hires 630 ex-Meta staff as ChatGPT memory may turn data into ads
OpenAI is becoming Meta, one employee at a time. It has hired 630 of them, enough to fill an internal Slack channel and worry the old guard.
Developers Clash Over Google’s Cursor Killer Antigravity IDE Amid Hype
The coffee machine at every Y Combinator startup is now forking Google’s Antigravity IDE.
DFlash drafts whole token blocks, achieving 15× throughput on NVIDIA Blackwell
Token-by-token generation is the bottleneck that has kept large language models tethered to a serial fate. DFlash breaks that chain.
Wiola Architecture Introduces Five Novel Components for Efficient Small Language Models
The obsession with trillion-parameter models has become a bad joke. A useful one is happening with small ones.
DeepSeek AI unveils DeepSeek‑V4 with compressed attention for 1 M‑token contexts
The AI arms race has always been a battle of scale, until now. DeepSeek’s latest release, DeepSeek-V4, doesn’t just push context windows to a million...
LWiAI Podcast #238: GPT 5.4 Mini, OpenAI Pivot, Mamba 3, Attention Residuals
The dust has barely settled on last week’s AI news cycle, and the landscape has already shifted again.
Alibaba's Qwen3.5-9B outperforms OpenAI's gpt-oss-120B on laptop benchmarks
A nine billion parameter model just whipped a one hundred twenty billion parameter model.
Figma launches AI motion graphics, shader tools, code layers, and new creative materials
Figma's latest update targets more than interface clutter. It takes aim at the org chart.
WhatsApp launches Meta AI Incognito Chat, cuts latency for privacy
The infinite feedback loop of your data feeding the AI, and the AI feeding Meta, just got a kill switch.
LWiAI Podcast #235: Sonnet 4.6, Deep‑Thinking Tokens, Anthropic vs Pentagon
It is rare to hear a company’s CEO stare down the Pentagon in public. But that is exactly what Anthropic’s Dario Amodei did this week, insisting that...
Nvidia and Meta ink deal; Nvidia touts hardware for inference and AI training
Nvidia locked in another computing deal this week, this time with Meta. While the contract's value stays under wraps, it specifically covers Nvidia...
Meta's DreamGym boosts AI agent success by 30% over baseline methods
Reinforcement learning is famously data-hungry, and painfully expensive when that data must come from the real world.
How Ollama’s Context Length Setting Impacts Local Model Memory
Picture this: You load an entire codebase into a local model, expecting an insightful analysis.
Meta says AI moderators make 13% fewer errors than humans, defends rollout speed
Thirteen percent. Meta's entire case for automating content moderation hinges on that single statistic.
Meta's SPICE framework beats baselines, boosts math and general reasoning
Teaching logic to an AI is a famously stubborn problem. Meta’s researchers just attacked it with a new, pugilistic framework called SPICE.
Poetiq’s Meta‑System Improves LLMs on LiveCodeBench for Reasoning, Retrieval
Large language models are powerful, until they’re asked to think, to find hidden facts, or to write code that actually compiles. Reasoning stalls.
IEEE launches five‑course online program on large language models
The engineering class is finally catching up to the hype. A new five-course program from the IEEE aims to turn people who use AI into people who can...
ZML releases free tool to speed AI inference across mixed chips
ZML, a Paris-based AI startup backed by Turing Award winner Yann LeCun, has released a free inference server called ZML/LLMD built to run open-source...
Zuckerberg Unveils Meta Compute to Build Global AI Infrastructure
Mark Zuckerberg is done renting. Meta will now build its own power plants. The company’s audacious plan to construct a global computing network for...
ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
ChatGPT briefly, bizarrely, fell for goblins. The whole fling started with a single broken piece of code.