AI Memory and Personalization @GoogleDeepMind. Gemini team || https://xitter.cf/t.co/wj3YgTdHue
Zurich, Switzerland
Joined September 2014
- Tweets501
- Following718
- Followers286
- Likes2K
Pinned Tweet
Remembering is not understanding. 🧠
We can give an AI "long-term memory" to store 10,000 facts about you, but that doesn't mean the system actually gets you. A thread on why AI memory needs a rewrite. 🧵
AI inverted software's cost function: divergence (generating options) is nearly free, making convergence (filtering and deciding) the bottleneck.
Engineers: Diverge on architecture; prune generated code ruthlessly.
PMs: Define the loss function and kill criteria before prototyping.
10 demos ≠ 1 shipped product.
Cool to see how expert knowledge is turned into AI knowledge in industry
Hello world! 🎥🤖
Generating an AI video is easy.
Producing the right video repeatedly for a specific brand is an engineering problem.
Without built-in evaluation, faster AI generation just moves the bottleneck downstream and creates more work for human reviewers.
We're breaking it down how we do it at @Replai_io in
replai.io/blog/from-expert-j…
Congrats to the team!
Meet Mistral Large 4, aka Le Chonk.
• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
• Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure.
• Available to all via API today. Working with cybersecurity partners privately.
Open weights release end of October.
The holy trinity of Lisbon to Porto transit options, ranked by how they test your patience:
Train (Alfa Pendular): ~3 hours | ~€35
Billed as "high-speed," it is not.
Car (Motorway): ~3h 15 min | ~€60–€65 (€25 tolls + ~€35–€40 fuel)
Three hours of someone tailgating if you follow the rules.
Plane (TAP): 50 min in air / ~4 hours door-to-door | ~€50–€100+
Take a metro, arrive 1 hour early, wait for a delayed boarding bus, spend 40min inside the airplane because there is a lot of air traffic, spend 35 minutes in the air, plane is stranded in Porto taxi way, take uber.
Pure efficiency.
As an aerospace engineer, this is gold for me :)
I remember the first meeting where the team pitched the idea of putting compute in space - exciting idea!
Today's successful launch with our partner @planet prototype satellite shows how far we've come...and how far there is to go. Thanks for the lift, @SpaceX, another spectacular booster landing too🚀
Hugo Lopes retweeted
Lots of discussion out there about our next model(!), so I wanted to give an early look as soon as possible. Introducing Gemini 4 Argon!
It shows frontier performance in complex workflows, cyber defense and software engineering. Teams are using it extensively at Google, from coding to quantum computing, great feedback.
Here’s a look at the benchmarks:
Huge congratulations to everyone with an accepted paper at the Personalized, aligned, long-term memory for AI systems workshop @PALMNeurips2026 @NeurIPSConf
> 62 papers accepted pushing boundaries across memory architectures, test-time learning, and user alignment.
> 5 selected for oral presentations highlighting standout breakthroughs.
In my opinion, we should not weaken open-weight competition, it is where academics can perform their work and advance science and technology.
What separates a generic chatbot from a genuinely helpful AI assistant?
Memory. 🧠
I’m taking the stage this Saturday (Oct 3) at the Sword AI Summit in Porto to talk AI Persistent Memory Architectures.
Lineup is stacked—who's around? 👇 aisummit.swordhealth.com/#ev…
Remember: Compete on human impact, not benchmarks.
Today we have set out how we’re building AI to accelerate science and improve people’s lives. Just some examples in the last week or so:
- Mapped all 9B possible single letter genetic changes across the human genome with AlphaGenome Atlas and made it openly available to researchers.
- Billions of decisions depend on weather predictions so we introduced WeatherNext 3, our most accurate and capable global weather AI model to date.
- We published AI & Economy ATLAS, a comprehensive open-access look at how people are using AI globally.
- AI has enabled extraordinary advances in language translation. Today our services are available in nearly 300 languages, spoken by 7B people
We’re focusing our efforts on four key areas: health, natural disaster and weather resilience, learning, and economic opportunity. blog.google/innovation-and-a…
Go go go! This is going to be an amazing workshop. Lots of interesting people joining!
⏳ Deadline reminder & call for reviewers
📅 PALM workshop @NeurIPSConf 2026: submit on OpenReview by Aug 29, AoE.
📝 Non-archival — work in progress, position papers and negative results welcome
🙋 We're also still welcoming new reviewers.
🔗 Link in comments
I started Aerospace Engineering at @istecnico 20 years ago, when choosing a degree felt like a predictable map for life.
Today, facing AI and global shifts, my advice to 18-year-olds isn't about picking a title, it's choosing an environment that sharpens critical thinking, multidisciplinary bridges, and our most human skills.
A university degree isn’t the destination. It’s just a tool to help you build what you’re truly capable of.
Embrace it.